Characterising Outcomes in Rheumatoid Arthritis
An Investigation of Methods for Predicting Rheumatoid Arthritis Patient Response in Clinical Trials from Clinical Biomarkers and Patient CharacteristicsThesis submitted in Fulfilment for the Degree of Doctor of PhilosophySchool of Health and Related Research (ScHARR)Paul Mahoney 2017AcknowledgementsThis work would not have been completed without the support of individuals and organisations. First and foremost, I would like to express my gratitude to my supervisors, Professor Steven Julious and Professor Mike Campbell, for their invaluable advice and unwavering support during the course of this research. Their patience has been tested to the limit. I would like to thank the Roche Products Limited for providing funding support, study time, use of company facilities and the clinical trial data which form the basis of this research. DedicationTo Michelle, for the encouragement, review and support (and cajoling when needed) to get me to this point. To Luke, Katie and Jeannie in believing I could do it. Disclaimer and Author’s DeclarationThe views expressed in this research are those of the author and not necessarily those of the organisations of which the author is affiliated. The author has no competing interests to declare.The author declares that this thesis is his original work and that none of the material contained in this thesis has previously been submitted for a degree to any awarding institution. The work contained in this thesis has been undertaken by me, with the support from those individuals or collaborators mentioned in the Acknowledgements section. AbstractIntroduction: Rheumatoid Arthritis (RA) is a chronic, destructive, autoimmune disorder of unknown cause and with no known cure. RA is a multi-factorial disease with many factors influencing onset, severity and outcome.Research Question: Of interest is to understand how selection of patients might influence patient outcome in clinical trials. Specifically, which are the patient characteristics or biomarkers measurable at enrolment that would be important in predicting response? In this PhD we will attempt to answer the question where response is the dichotomous outcome ACR20.Methods: The research for this dissertation accessed clinical trial data from over 11,700 patients enrolled into 16 late stage RA clinical trials from 1998 to 2008. Through systematic review and a review and selection process, logistic regression and CART were selected to compare head to head in simulations. In simulations each method was compared using Variable Selection (VaSe) plots developed in the PhD. In most scenarios CART outperformed logistic regression, and was selected to apply to the clinical database. Results: The CART analysis, generated a model that had an overall predictive accuracy for ACR20 of just 60%, although the explanatory variables selected in the model were plausible for predicting patient outcome: baseline tender joint count, region, joint space narrowing score, number of previous treatments and race. When one considered the 8 individual components of the ACR20, the predictive accuracy increased from 60% to between 67% - 82%. Conclusions: Through the use of a large RA dataset, we were unable to predict ACR20 response to a satisfactory level. However, we were able to predict each component of the ACR20 well. This is likely due to the complexity of the ACR20 as a tool for patient response as well as the heterogeneity of this multifactorial condition. From a patient perspective it may be of more value to be able to predict a disease symptom such as number of swollen joints rather than a composite score. This dissertation provides a framework for investigating predictive markers for patient response in clinical trial in RA and other disease areas. Table of Contents TOC \o "1-3" \h \z \u Acknowledgements PAGEREF _Toc518027138 \h iiDedication PAGEREF _Toc518027139 \h iiiDisclaimer and Author’s Declaration PAGEREF _Toc518027140 \h ivAbstract PAGEREF _Toc518027141 \h vChapter 1 : Introduction PAGEREF _Toc518027142 \h 11.1.Background PAGEREF _Toc518027143 \h 21.2.Research question and rationale PAGEREF _Toc518027144 \h 31.3.Specific Objectives PAGEREF _Toc518027145 \h 31.4.Scope of the research PAGEREF _Toc518027146 \h 41.5.Thesis Roadmap PAGEREF _Toc518027147 \h 41.6.Summary PAGEREF _Toc518027148 \h 5Chapter 2 : Background to Rheumatoid Arthritis PAGEREF _Toc518027149 \h 62.1.Introduction PAGEREF _Toc518027150 \h 72.2.Aims PAGEREF _Toc518027151 \h 72.3.Rheumatoid Arthritis PAGEREF _Toc518027152 \h 72.4.Descriptive Epidemiology PAGEREF _Toc518027153 \h 82.5.Causes of Rheumatoid Arthritis PAGEREF _Toc518027154 \h 102.6.Diagnosis PAGEREF _Toc518027155 \h 102.7.Manifestations and co-morbidities associated with rheumatoid arthritis PAGEREF _Toc518027156 \h 122.8.Pathophysiology PAGEREF _Toc518027157 \h 142.9.Risk Factors for RA PAGEREF _Toc518027158 \h 152.10.Summary PAGEREF _Toc518027159 \h 17Chapter 3 : Rheumatoid Arthritis Clinical Trial Design and Endpoints PAGEREF _Toc518027160 \h 193.1.Introduction PAGEREF _Toc518027161 \h 203.2.Aims PAGEREF _Toc518027162 \h 203.3.Assessing disease activity PAGEREF _Toc518027163 \h 203.3.1. American College of Rheumatology (ACR) criteria PAGEREF _Toc518027164 \h 213.3.2. Disease Activity Score (DAS28) PAGEREF _Toc518027165 \h 243.3.3. European League Against Rheumatism (EULAR) response criteria PAGEREF _Toc518027166 \h 253.3.4. Radiographic evaluation in clinical development PAGEREF _Toc518027167 \h 253.3.5. Health Assessment Questionnaire Disability Index (HAQ–DI). PAGEREF _Toc518027168 \h 273.3.6. Functional Assessment of Chronic Illness–Fatigue. (FACIT-Fatigue) PAGEREF _Toc518027169 \h 273.4.Regulatory pathway for approval of new therapies PAGEREF _Toc518027170 \h 283.5.Current treatment strategies for rheumatoid arthritis PAGEREF _Toc518027171 \h 293.6.Phase III Clinical Trial Design PAGEREF _Toc518027172 \h 313.7.Summary PAGEREF _Toc518027173 \h 33Chapter 4 : Data Description PAGEREF _Toc518027174 \h 354.1Introduction PAGEREF _Toc518027175 \h 364.2.Aims PAGEREF _Toc518027176 \h 364.3.Projects and Studies PAGEREF _Toc518027177 \h 364.3.1. Project A PAGEREF _Toc518027178 \h 374.3.2. Project M PAGEREF _Toc518027179 \h 414.3.3. Project O PAGEREF _Toc518027180 \h 454.3.4. Project T PAGEREF _Toc518027181 \h 484.4.Operational challenges in database creation PAGEREF _Toc518027182 \h 524.5Dependent and Explanatory Variables PAGEREF _Toc518027183 \h 564.6.Response & Demography and Disease History PAGEREF _Toc518027184 \h 594.7.Available and Missing Data PAGEREF _Toc518027185 \h 614.8.Response Data PAGEREF _Toc518027186 \h 654.9.Demography and Disease History Data PAGEREF _Toc518027187 \h 654.10. Baseline ACR components Data PAGEREF _Toc518027188 \h 664.11.Vital Signs and Radiography Data PAGEREF _Toc518027189 \h 664.12.Laboratory Data PAGEREF _Toc518027190 \h 694.13.SF36 Data PAGEREF _Toc518027191 \h 704.14.Correlations and Associations with ACR20 PAGEREF _Toc518027192 \h 714.15. Limitations PAGEREF _Toc518027193 \h 754.16.Criteria for method selection PAGEREF _Toc518027194 \h 764.16.1. Mixed Data PAGEREF _Toc518027195 \h 764.16.2. Missing Values PAGEREF _Toc518027196 \h 774.16.3. Outliers PAGEREF _Toc518027197 \h 774.16.4. Scalability PAGEREF _Toc518027198 \h 774.16.5. High Dimensions PAGEREF _Toc518027199 \h 774.16.6. Interpretability PAGEREF _Toc518027200 \h 774.17. Summary PAGEREF _Toc518027201 \h 78Chapter 5 : Data Mining Methodologies PAGEREF _Toc518027202 \h 815.1. Introduction PAGEREF _Toc518027203 \h 825.2. Aims PAGEREF _Toc518027204 \h 835.3.CART PAGEREF _Toc518027205 \h 845.3.1. Bagging PAGEREF _Toc518027206 \h 875.3.2. Boosting PAGEREF _Toc518027207 \h 885.3.3. Random Forest PAGEREF _Toc518027208 \h 895.4. k-Nearest Neighbour Classification PAGEREF _Toc518027209 \h 915.5.Support Vector Machines PAGEREF _Toc518027210 \h 925.6.Linear Regression and Logistic Regression PAGEREF _Toc518027211 \h 945.6.1. Ridge Regression PAGEREF _Toc518027212 \h 955.6.2. Lasso PAGEREF _Toc518027213 \h 955.7.Principal Components PAGEREF _Toc518027214 \h 975.8. Cross Validation PAGEREF _Toc518027215 \h 975.9.Missing Data PAGEREF _Toc518027216 \h 985.10.Summary PAGEREF _Toc518027217 \h 99Chapter 6 : Multivariate Methods Systematic Review PAGEREF _Toc518027218 \h 1016.1.Introduction PAGEREF _Toc518027219 \h 1026.2.Aims PAGEREF _Toc518027220 \h 1026.3.Systematic review of multivariate methods applied in RA research PAGEREF _Toc518027221 \h 1036.3.1. Methods PAGEREF _Toc518027222 \h 1036.3.2. Results PAGEREF _Toc518027223 \h 1046.4.Systematic Review of Data Mining Methodologies PAGEREF _Toc518027224 \h 1086.4.1. Methods PAGEREF _Toc518027225 \h 1086.4.2. Results PAGEREF _Toc518027226 \h 1086.4.3. CART PAGEREF _Toc518027227 \h 1086.4.4. Random Forest PAGEREF _Toc518027228 \h 1096.4.5. Nearest Neighbour PAGEREF _Toc518027229 \h 1096.4.6. Kernel PAGEREF _Toc518027230 \h 1096.4.7. Support Vector Machines PAGEREF _Toc518027231 \h 1106.4.8. Ridge Regression and Lasso PAGEREF _Toc518027232 \h 1106.4.9. Principal Components PAGEREF _Toc518027233 \h 1106.4.10. Cross Validation PAGEREF _Toc518027234 \h 1106.5.Systematic Review Summary PAGEREF _Toc518027235 \h 1116.6.Systematic Review comparing CART and Logistic Regression PAGEREF _Toc518027236 \h 1126.7.Summary PAGEREF _Toc518027237 \h 113Chapter 7 : Simulations PAGEREF _Toc518027238 \h 1157.1.Introduction PAGEREF _Toc518027239 \h 1167.2.Aims PAGEREF _Toc518027240 \h 1177.3.Methods for Generating Simulations PAGEREF _Toc518027241 \h 1177.4.Simulations with continuous explanatory variables and no missing data PAGEREF _Toc518027242 \h 1197.5.Simulations with continuous explanatory variables and no missing data PAGEREF _Toc518027243 \h 1207.5.1. Initial Simulations Introduction PAGEREF _Toc518027244 \h 1207.5.2. Methods PAGEREF _Toc518027245 \h 1207.5.3. Results of Comparisons of CART versus Logistic Regression PAGEREF _Toc518027246 \h 1237.5.4. Improving the performance of Logistic Regression PAGEREF _Toc518027247 \h 1277.5.5. Initial Simulations Conclusions PAGEREF _Toc518027248 \h 1307.6.Simulations with continuous and categorical explanatory variables and no missing data PAGEREF _Toc518027249 \h 1317.6.1. Categorical and Continuous Simulations Introduction PAGEREF _Toc518027250 \h 1317.6.2. Methods PAGEREF _Toc518027251 \h 1317.6.3. Results PAGEREF _Toc518027252 \h 1317.6.4.1.Categorical and Continuous Simulations Conclusions PAGEREF _Toc518027253 \h 1337.7.Simulations with continuous explanatory variables and missing data PAGEREF _Toc518027254 \h 1347.7.1. Missing Data Simulations Introduction PAGEREF _Toc518027255 \h 1347.7.2. Methods PAGEREF _Toc518027256 \h 1347.7.3. Results PAGEREF _Toc518027257 \h 1347.7.4. Missing Data Simulations Conclusions PAGEREF _Toc518027258 \h 1377.8.Summary PAGEREF _Toc518027259 \h 1377.9.Limitations PAGEREF _Toc518027260 \h 1407.10Recommendations PAGEREF _Toc518027261 \h 141Chapter 8 : Application of CART using Clinical Trial Data PAGEREF _Toc518027262 \h 1438.1.Introduction PAGEREF _Toc518027263 \h 1448.2.Aims PAGEREF _Toc518027264 \h 1458.3.Methods PAGEREF _Toc518027265 \h 1458.4.Primary Analysis Results PAGEREF _Toc518027266 \h 1468.5.Additional analyses on primary endpoint PAGEREF _Toc518027267 \h 1508.5.1. CART analysis with boosting & bagging PAGEREF _Toc518027268 \h 1508.5.2. CART analysis with Random Forest PAGEREF _Toc518027269 \h 1528.5.3. k-fold cross validation PAGEREF _Toc518027270 \h 1538.5.4. Logistic regression analysis PAGEREF _Toc518027271 \h 1548.5.5. CART analysis on completers PAGEREF _Toc518027272 \h 1558.6.Exploratory Analysis PAGEREF _Toc518027273 \h 1568.6.1. CART analysis on subgroups PAGEREF _Toc518027274 \h 1568.6.2. CART analysis exploring study specific variables PAGEREF _Toc518027275 \h 1598.6.3. CART analysis on ACR components PAGEREF _Toc518027276 \h 1618.6.4. Logistic regression with multiple imputation PAGEREF _Toc518027277 \h 1678.6.5. CART analysis with multiple imputation PAGEREF _Toc518027278 \h 1698.7.Summary PAGEREF _Toc518027279 \h 170Chapter 9 : Discussion and Conclusions PAGEREF _Toc518027280 \h 1739.1.Introduction PAGEREF _Toc518027281 \h 1749.2.Main findings PAGEREF _Toc518027282 \h 1759.3.Objective 1: Choice of statistical or data mining method that best identifies biomarkers PAGEREF _Toc518027283 \h 1769.3.1. Systematic Reviews PAGEREF _Toc518027284 \h 1779.3.2. Simulations PAGEREF _Toc518027285 \h 1789.4.Objective 2: Identification of biomarkers and baseline patient assessments that predict patient outcome PAGEREF _Toc518027286 \h 1809.5.Objective 3: Improvement in study design in RA clinical trials through patient selection and collection of most relevant data points PAGEREF _Toc518027287 \h 1839.6.Main Thesis Strengths and Achievements PAGEREF _Toc518027288 \h 1859.7.Key Limitations PAGEREF _Toc518027289 \h 1869.8.Future Related Research PAGEREF _Toc518027290 \h 1879.9.Overall Conclusions PAGEREF _Toc518027291 \h 187References PAGEREF _Toc518027292 \h 190Appendices PAGEREF _Toc518027293 \h 197Appendix A: Systematic Reviews PAGEREF _Toc518027294 \h 198A.1.Multivariate Methods Systematic Reviews PAGEREF _Toc518027295 \h 199A.1.1.Medline Systematic Review PAGEREF _Toc518027296 \h 199A.1.2.Embase Systematic Review PAGEREF _Toc518027297 \h 204A.1.3.Medline – revised criteria PAGEREF _Toc518027298 \h 210A.1.4.Systematic Review of Data Mining Methodologies PAGEREF _Toc518027299 \h 211A.1.5.Systematic Review of logistic regression and CART in the context of variable selection PAGEREF _Toc518027300 \h 212Appendix B: Clinical Trial Dataset PAGEREF _Toc518027301 \h 215B.1.Detail of variables PAGEREF _Toc518027302 \h 216B.2.Summary statistics PAGEREF _Toc518027303 \h 219Appendix C: Simulations PAGEREF _Toc518027304 \h 239C.1.Simulation Protocol PAGEREF _Toc518027305 \h 240C.2.Simulation code PAGEREF _Toc518027306 \h 243C.2.1.R Code simrpart.R PAGEREF _Toc518027307 \h 244C.2.2.SAS Code simmacro.sas PAGEREF _Toc518027308 \h 245C.2.3.SAS Code simmacro.sas PAGEREF _Toc518027309 \h 246C.3.Simulation Scenarios PAGEREF _Toc518027310 \h 249C.4.Simulation Poster PAGEREF _Toc518027311 \h 250Appendix D: Analyses PAGEREF _Toc518027312 \h 251D.1.Primary CART Analysis PAGEREF _Toc518027313 \h 252Table of Figures TOC \h \z \c "Figure" Figure 2.1 Inflamed Synovium (Pannus) Causes Joint Destruction PAGEREF _Toc518391747 \h 13Figure 3.1 Joints measured for swelling and tenderness PAGEREF _Toc518391748 \h 23Figure 3.2 Typical RA Phase III Trials at Roche PAGEREF _Toc518391749 \h 32Figure 4.1 ACR20 Response by Treatment Type and Gender PAGEREF _Toc518391750 \h 61Figure 4.2 Missing Data across all variables PAGEREF _Toc518391751 \h 62Figure 4.3 Variables with more than 50% missing data PAGEREF _Toc518391752 \h 63Figure 4.4 Variables with 10% to 50% missing data PAGEREF _Toc518391753 \h 64Figure 4.5 Percentage of patients with an ACR Response assessment PAGEREF _Toc518391754 \h 65Figure 4.6 Percentage of patients with Demographic data PAGEREF _Toc518391755 \h 66Figure 4.7 Percentage of patients with baseline ACR component data PAGEREF _Toc518391756 \h 66Figure 4.8 Percentage of patients with Vital Sign and X-ray data PAGEREF _Toc518391757 \h 67Figure 4.9 Percentage of patients with Vital Sign and X-ray data by project PAGEREF _Toc518391758 \h 68Figure 4.10 Percentage of patients with laboratory data PAGEREF _Toc518391759 \h 69Figure 4.11 Percentage of patients with laboratory data in Project A PAGEREF _Toc518391760 \h 69Figure 4.12 Percentage of patients with laboratory data in Project M PAGEREF _Toc518391761 \h 70Figure 4.13 Percentage of patients with laboratory data in Project O PAGEREF _Toc518391762 \h 70Figure 4.14 Percentage of patients with laboratory data in Project T PAGEREF _Toc518391763 \h 70Figure 4.15 Percentage of patients with SF36 Component data PAGEREF _Toc518391764 \h 71Figure 4.16 Percentage of patients with SF36 Component scores PAGEREF _Toc518391765 \h 71Figure 5.1 Planned analysis method refinement PAGEREF _Toc518391766 \h 83Figure 5.2 Example of CART PAGEREF _Toc518391767 \h 86Figure 5.3 Example ROC Curve PAGEREF _Toc518391768 \h 87Figure 5.4 Bagging PAGEREF _Toc518391769 \h 88Figure 5.5 Boosting PAGEREF _Toc518391770 \h 89Figure 5.6 Random Forest PAGEREF _Toc518391771 \h 90Figure 5.7 Example Support Vector Machine PAGEREF _Toc518391772 \h 93Figure 5.8 Comparison of Ridge Regression and Lasso PAGEREF _Toc518391773 \h 96Figure 6.1 Flowchart of planned analysis method refinement PAGEREF _Toc518391774 \h 114Figure 7.1 Flowchart of planned analysis method refinement PAGEREF _Toc518391775 \h 116Figure 7.2 Example VaSe plot PAGEREF _Toc518391776 \h 120Figure 7.3 Generation of Simulated Datasets PAGEREF _Toc518391777 \h 122Figure 7.4 VaSe plot of logistic regression and CART for 50% response rate and zero correlation PAGEREF _Toc518391778 \h 125Figure 7.5 VaSe plot of logistic regression and CART for 50% response rate and low correlation at 0.1 PAGEREF _Toc518391779 \h 125Figure 7.6 VaSe plot of logistic regression and CART for 50% response rate and low correlation at 0.2 PAGEREF _Toc518391780 \h 126Figure 7.7 VaSe plot of logistic regression and CART for 50% response rate and low correlation at 0.3 PAGEREF _Toc518391781 \h 126Figure 7.8 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with zero correlation PAGEREF _Toc518391782 \h 128Figure 7.9 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with low correlation at 0.1 PAGEREF _Toc518391783 \h 129Figure 7.10 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with low correlation at 0.2 PAGEREF _Toc518391784 \h 129Figure 7.11 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with low correlation at 0.3 PAGEREF _Toc518391785 \h 130Figure 7.12 Effect of dichotomising explanatory variables in Logistic Regression vs CART for 50% Response PAGEREF _Toc518391786 \h 133Figure 7.13 VaSe plot of logistic regression and CART by rate of missing values none, 0.1% and 1% (missing-completely-at-random) PAGEREF _Toc518391787 \h 136Figure 7.14 VaSe plot of logistic regression and CART by rate of missing values none, 0.001 and 0.01 (missing not at random) PAGEREF _Toc518391788 \h 137Figure 7.15 Flowchart of analysis method refinement PAGEREF _Toc518391789 \h 142Figure 8.1 CART primary analysis PAGEREF _Toc518391790 \h 147Figure 8.2 Loss function Operating Characteristic Curve PAGEREF _Toc518391791 \h 149Figure 8.3 Boosting Variable Importance PAGEREF _Toc518391792 \h 151Figure 8.4 Bagging Variable Importance PAGEREF _Toc518391793 \h 152Figure 8.5 Random Forest Variable Importance PAGEREF _Toc518391794 \h 153Figure 8.6 Receiver operating characteristic curve for logistic regression PAGEREF _Toc518391795 \h 155Figure 8.7 CART completers analysis PAGEREF _Toc518391796 \h 156Figure 8.8 CART including study parameters PAGEREF _Toc518391797 \h 159Figure 8.9 CART excluding Project T PAGEREF _Toc518391798 \h 160Figure 8.10 CART 20% Improvement in Tender Joint Count PAGEREF _Toc518391799 \h 163Figure 8.11 CART 20% Improvement in Swollen Joint Count PAGEREF _Toc518391800 \h 163Figure 8.12 CART 20% Improvement in CRP PAGEREF _Toc518391801 \h 164Figure 8.13 CART 20% Improvement in ESR PAGEREF _Toc518391802 \h 164Figure 8.14 CART 20% Improvement in HAQ PAGEREF _Toc518391803 \h 165Figure 8.15 CART 20% Improvement in Patient's Assessment of Pain PAGEREF _Toc518391804 \h 165Figure 8.16 CART 20% Improvement in Physician's Global Assessment of Disease Activity PAGEREF _Toc518391805 \h 166Figure 8.17 CART 20% Improvement in Patient's Global Assessment of Disease Activity PAGEREF _Toc518391806 \h 166Figure 8.18 CART analysis with multiple imputation PAGEREF _Toc518391807 \h 170Figure 9.1 Example VaSe plot PAGEREF _Toc518391808 \h 179Figure A1 Results of Medline Review PAGEREF _Toc518391809 \h 200Figure A2 Results of Revised Medline Review PAGEREF _Toc518391810 \h 209Figure A3 Results of logistic regression and CART variable selection systematic review PAGEREF _Toc518391811 \h 211Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables PAGEREF _Toc518391812 \h 218Figure B2 Categorical variables PAGEREF _Toc518391813 \h 229Figure B3 Medical History variables PAGEREF _Toc518391814 \h 231Figure C1 Generation of Simulated Datasets PAGEREF _Toc518391815 \h 242Table of Tables TOC \h \z \c "Table" Table 2.1 Prevalence of RA worldwide (cases per 100 inhabitants) (2004) PAGEREF _Toc518391816 \h 9Table 2.2 Incidence rates of RA worldwide (2004) PAGEREF _Toc518391817 \h 10Table 2.3 Association between rheumatoid arthritis and cigarette smoking PAGEREF _Toc518391818 \h 16Table 3.1 American College of Rheumatology definition of ACR20 measure PAGEREF _Toc518391819 \h 22Table 3.2 EULAR response PAGEREF _Toc518391820 \h 25Table 4.1 Summary of Project A studies PAGEREF _Toc518391821 \h 38Table 4.2 Summary of Project M studies PAGEREF _Toc518391822 \h 42Table 4.3 Summary of Project O studies PAGEREF _Toc518391823 \h 46Table 4.4 Summary of Project T studies PAGEREF _Toc518391824 \h 49Table 4.5 Summary of RA database available for analysis PAGEREF _Toc518391825 \h 51Table 4.6 Summary of ACR20 response rates by study PAGEREF _Toc518391826 \h 52Table 4.7 Variables available for analysis in RA Database PAGEREF _Toc518391827 \h 57Table 4.8 Summary of ACR20 and Demographic Data PAGEREF _Toc518391828 \h 60Table 4.9 Correlations between ordinal categorical variables and ACR20 PAGEREF _Toc518391829 \h 72Table 4.10 Associations between non ordinal categorical variables and ACR20 response rates PAGEREF _Toc518391830 \h 73Table 4.11 Correlations between continuous variables and ACR20 PAGEREF _Toc518391831 \h 75Table 4.12 Methodology Selection Criteria PAGEREF _Toc518391832 \h 79Table 5.1 Methodology Selection Criteria for CART PAGEREF _Toc518391833 \h 91Table 5.2 Methodology Selection Criteria for k-Nearest Neighbour PAGEREF _Toc518391834 \h 92Table 5.3 Methodology Selection Criteria for Support Vector Machines PAGEREF _Toc518391835 \h 94Table 5.4 Methodology Selection Criteria for Logistic Regression PAGEREF _Toc518391836 \h 96Table 5.5 Methodology Selection Criteria for Principal Components PAGEREF _Toc518391837 \h 97Table 5.6 Some characteristics of classification and data mining methods PAGEREF _Toc518391838 \h 99Table 6.1 Summary of Medline and Embase Systematic Reviews PAGEREF _Toc518391839 \h 105Table 6.2 Data Mining Systematic Review PAGEREF _Toc518391840 \h 108Table 6.3 Some characteristics of classification and data mining methods PAGEREF _Toc518391841 \h 112Table 7.1 Some characteristics of CART and Logistic Regression PAGEREF _Toc518391842 \h 141Table 8.1 CART prediction results PAGEREF _Toc518391843 \h 147Table 8.2 CART prediction results PAGEREF _Toc518391844 \h 148Table 8.3 Impact on accuracy of adapting loss function PAGEREF _Toc518391845 \h 149Table 8.4 Boosting prediction results PAGEREF _Toc518391846 \h 150Table 8.5 Bagging prediction results PAGEREF _Toc518391847 \h 151Table 8.6 Random Forest prediction results on all data PAGEREF _Toc518391848 \h 152Table 8.7 Correlations between X-Ray assessments PAGEREF _Toc518391849 \h 153Table 8.8 Logistic Regression prediction results on CART final model PAGEREF _Toc518391850 \h 155Table 8.9 CART completers prediction results PAGEREF _Toc518391851 \h 156Table 8.10 CART and Random Forest Subgroup Analyses PAGEREF _Toc518391852 \h 157Table 8.11 CART prediction results including study parameters PAGEREF _Toc518391853 \h 159Table 8.12 CART prediction excluding Project T PAGEREF _Toc518391854 \h 160Table 8.13 CART summary of ACR components PAGEREF _Toc518391855 \h 161Table 8.14 Tree nodes from CART analysis of ACR components PAGEREF _Toc518391856 \h 167Table 8.15 Logistic Regression prediction results with multiple imputation PAGEREF _Toc518391857 \h 169Table 8.16 CART prediction results with multiple imputation PAGEREF _Toc518391858 \h 169Table 8.17 Summary of CART and Logistic regression methods on ACR20 PAGEREF _Toc518391859 \h 172Table 9.1 Selection assessment for method selection PAGEREF _Toc518391860 \h 177Table A0.1 Medline Search for statistical methods used to analyse risk factors in RA PAGEREF _Toc518391861 \h 199Table A0.2 Embase Search for statistical methods used to analyse risk factors in RA PAGEREF _Toc518391862 \h 203Table A0.3 Summary of Medline and Embase Systematic Reviews PAGEREF _Toc518391863 \h 206Table A0.4 Data Mining Systematic Review PAGEREF _Toc518391864 \h 210Table C0.1 Summary of number of datasets created for main simulations PAGEREF _Toc518391865 \h 248Table C0.2 Summary of number of data points generated in simulations PAGEREF _Toc518391866 \h 248Table of AbbreviationsAnti-TNFAnti-Tumour Necrosis FactorACRAmerican College of RheumatologyCART Classification and Regression TreeCCP Citrulline Containing ProteinsCDISCClinical Data Interchange Standards ConsortiumCRPC - reactive proteinDAS28 Disease Activity Score (28 Joints)DMARDDisease-Modifying Anti-Rheumatic DrugEMAEuropean Medicines AgencyESRErythrocyte Sedimentation RateEULAREuropean League Against RheumatismFACIT-Fatigue Functional Assessment of Chronic Illness–FatigueFDA Food and Drug Administration GEEGeneralised Estimating EquationsGH Global HealthHAQ-DIHealth Assessment Questionnaire Disability IndexHLA Human Leukocyte AntigenIL-1Interleukin-1IL-6RInterleukin 6 receptor LnNatural LogarithmLRLogistic RegressionMANOVAMultivariate Analysis of VarianceNICENational Institute for Health and Clinical ExcellenceMCPMetacarpophalangealMTXMethotrexatePCAPrincipal Components AnalysisPIPProximal InterphalangealRA Rheumatoid ArthritisRFRheumatoid FactorROCReceiver Operating CharacteristicROWRest of WorldSDStandard DeviationSJCSwollen Joint CountSLESignificance Level for Entry in the modelSLSSignificance Level for Staying in the modelsJIAsystemic Juvenile Idiopathic ArthritisTJC Tender Joint CountVaSeVariable SelectionWHO World Health Organisation: IntroductionBackground When developing a new medicine, a question a clinical development team may wish to explore is what lessons can be learnt from past clinical development experiences. In particular, when there is in-house experience in the disease area, the research question is whether this valuable information could be used for the development of a new asset in the most efficient way. Prognostic markers can have utility in segmentation of patient populations, for example in setting the inclusion criteria for early stages of clinical development ADDIN EN.CITE <EndNote><Cite><Author>Jenkins</Author><Year>2011</Year><RecNum>182</RecNum><DisplayText>(Jenkins, 2011)</DisplayText><record><rec-number>182</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1500565065">182</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Jenkins, M., Flynn, A., Smart, T., Harbron, C., Sabin, T.., Ratnayake, J., Delmar, P., Herath, A., Jarvis, P., Matcham, J.</author></authors></contributors><titles><title>A statistician’s perspective on biomarkers in drug development</title><secondary-title>Pharmaceutical Statistics</secondary-title></titles><periodical><full-title>Pharmaceutical Statistics</full-title></periodical><pages>494-507</pages><volume>10</volume><dates><year>2011</year></dates><urls></urls></record></Cite></EndNote>(Jenkins, 2011). In oncology, for example, sometimes there is sufficient evidence to suggest that potential benefit is limited to a specific biomarker defined subgroup of patients. In these situations the efficiency of targeted clinical trials can be dramatically increased ADDIN EN.CITE <EndNote><Cite><Author>Simon</Author><Year>2004</Year><RecNum>180</RecNum><DisplayText>(Simon & Maitoutnam, 2004)</DisplayText><record><rec-number>180</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">180</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Simon, R. </author><author>Maitoutnam, A.</author></authors></contributors><titles><title>Evaluating the Efficiency of Targeted Designs for Randomized Clinical Trials</title><secondary-title>Clinical Cancer Research</secondary-title></titles><pages>6759-6763</pages><volume>10</volume><dates><year>2004</year></dates><urls></urls></record></Cite></EndNote>(Simon & Maitoutnam, 2004). In 2012, the US Food and Drug Administration (FDA) published a draft guidance on enrichment strategies for clinical trials ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2012</Year><RecNum>186</RecNum><DisplayText>(FDA, 2012)</DisplayText><record><rec-number>186</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502704051">186</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry: Enrichment Strategies for Clinical Trials to Support Approval of Human Drugs and Biological Products</title></titles><dates><year>2012</year></dates><urls></urls></record></Cite></EndNote>(FDA, 2012). The guidance defines enrichment as “the prospective use of any patient characteristic to select a study population in which detection of a drug effect (if one is in fact present) is more likely than it would be in an unselected population”, where biomarker characteristics can be drawn from demographic, pathophysiologic, historical, genetic or proteomic, clinical, and psychological sources. By recruiting patients with a higher likelihood of having a certain event can lead to a higher effect size, reduce heterogeneity and lead to smaller more efficient clinical trials.In this research we have access to a large clinical Rheumatoid Arthritis (RA) database. The data has accumulated from 4 previous clinical programs covering more than 10 years of RA clinical development. We were able to explore this historical data to inform new clinical development. We were interested to understand how selection of patients might influence patient outcome, specifically which patient characteristics measurable at screening and baseline would be important either in a univariate or multivariate fashion. Before we can utilise the database however, we will explore the best statistical methods for identifying the risk factors. This research will describe simulations to compare the performance of those methods most expected to be able to select risk factors correlated with response under differing scenarios with known data relationships. The method that performs better will be applied to the clinical database.Research question and rationaleWe are interested to understand how selection of patients might influence patient outcome in clinical trials. Specifically, which patient characteristics or biomarkers measurable at enrolment in clinical trials would be important in predicting response? With the available data we will attempt to answer the question; which factors identify risk for progression of RA in clinical trials, where progression is measured by the dichotomous response outcome ACR20. Specific ObjectivesThe specific objectives to address the research question are: Identify the statistical or data mining method that best identifies biomarkers;Identify biomarkers and baseline patient assessments that predict patient outcome for disease progression;Improve study design in RA clinical trials through patient selection and collection of most relevant data points.In this thesis, a biomarker is a biological characteristic recorded at enrolment to a clinical trial. This characteristic may be demographic, pathophysiologic, historical, genetic or proteomic, clinical, or psychological, and is a way to identify subgroups of the patient population either by a categorisation of the marker or by a defined cut-point of a quantitative measure. The data utilised later in this thesis will include demographic, pathophysiologic, historical, and clinical characteristics. To achieve these objectives this research will: Review and describe data mining methodologies for classification;Systematically review available literature for multivariate methods identifying risk factors for RA;Through the application of simulations, identify the best classification method for selecting risk factors;Describe the epidemiology, aetiology and diagnosis of RA, as well as risk factors and assessments of disease activity;Collate and describe the available clinical data;Apply the best method from simulations to clinical database;Review current designs of clinical trials in RA and strategies for approval.Scope of the researchThe thesis shall focus on identifying multivariate and data mining methods for predicting patient response in simulated RA data. The best performing methods will be used to identify clinical biomarkers to predict individual patient response in RA disease progression in a RA patient dataset. The data used will be from phase III clinical trials in 4 investigational drugs developed between 1998 and 2010 at Roche Products Ltd.Thesis RoadmapFirst of all, to answer the objectives for this research, this thesis will in Chapter 2, describe RA as a disease, present a definition of the disease, as well as the postulated causes of RA, risk factors believed to affect occurrence, severity, epidemiology of RA, and comorbidities and pathophysiology. Chapter 3 will explain how RA is diagnosed in clinical practice and to discuss the various measures used in clinical practice and clinical trials to assess disease activity. The regulatory pathways for registration and approval of new treatments for RA, and current treatment strategies in the clinic will also be discussed.Chapter 4 will introduce the clinical RA database. Here the data will be described in detail in in terms of: projects and studies, dependent and explanatory variables, missing data, observed correlations, and potential limitations. Building on the description of the data in Chapter 4, Chapter 5 will describe some statistical and data mining methodologies that may be appropriate for application to this data, and through reflection, reduce the number of methods that will be explored in simulations later in the thesis. Systematic reviews of literature will be performed in Chapter 6 to achieve multiple aims. Firstly, this chapter will explore the use of multivariate methods and specific data mining tools for identifying risk factors for RA. These methods will be described as well other data mining classification methodologies and reflect on their usefulness in the context of the available RA clinical trial data. Later in this chapter literature specifically comparing Classification and Regression Trees (CART) and Logistic Regression in the context of variable selection will be identified.Following findings in previous chapters, Chapter 7 will describe simulations conducted to explore how CART and logistic regression perform under differing scenarios with known data relationships with the intention of reducing the number of methods to apply to the real patient database as well as to indicate which analysis parameters to apply.Chapter 8 brings in the learnings from the simulations in Chapter 7 and applies the optimal methodology to the data described in Chapter 4. In conclusion, Chapter 9 completes the thesis with a discussion, recommendations for best practice for future RA drug development, limitations, and overall conclusionsSummaryRA is a multi-factorial disease with many postulated factors influencing onset, severity and outcome. There is a considerable variation of disease expression among different populations and groups of patients. Many measures of outcome rely on multiple subjective and objective assessments, leading to clinical trial designs with large numbers of patients and recorded explanatory variables. A better understanding of data collected at enrolment into RA clinical trials could lead to improvements in the design and conduct of randomised clinical trials in drug development, a potential to predict clinical trial outcomes from baseline data, and an increased focus on personalised healthcare leading to a reduction on the burden on patient volunteers.: Background to Rheumatoid ArthritisIntroductionRheumatoid arthritis (RA) is a systemic, multi-faceted disease that is not yet fully understood. Methods for diagnosing the disease have evolved over time as more is learnt through clinical observation ADDIN EN.CITE <EndNote><Cite><Author>Aletaha</Author><Year>2010</Year><RecNum>92</RecNum><DisplayText>(Aletaha, 2010)</DisplayText><record><rec-number>92</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">92</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Aletaha, D.</author></authors></contributors><titles><title>2010 Rheumatoid Arthritis Classification Criteria</title><secondary-title>ARTHRITIS & RHEUMATISM
An American College of Rheumatology/European League Against Rheumatism Collaborative Initiative</secondary-title></titles><pages>2569-2581</pages><volume>62</volume><number>9</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>classification</keyword><keyword>Rheumatology</keyword><keyword>RA</keyword><keyword>Disease</keyword><keyword>diagnosis</keyword><keyword>Joints</keyword></keywords><dates><year>2010</year></dates><label>162</label><urls></urls></record></Cite></EndNote>(Aletaha, 2010). The symptoms manifested by the disease are many, from joint pain to fatigue and restrictions in activities of daily living, thus making a single objective measure that summarises the status of RA very difficult ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2008</Year><RecNum>134</RecNum><DisplayText>(Smolen et al., 2008)</DisplayText><record><rec-number>134</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">134</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Smolen, J. </author><author>Aletaha, D.</author><author>Grisar, J. </author><author>Redlich, K .</author><author>Steiner, G.</author><author>Wagner, O.</author></authors></contributors><titles><title>The need for prognosticators in rheumatoid arthritis. Biological and clinical markers: where are we now? </title><secondary-title>Arthritis Res Ther</secondary-title></titles><periodical><full-title>Arthritis Res Ther</full-title></periodical><pages>208-219</pages><volume>10</volume><number>3</number><dates><year>2008</year></dates><urls></urls></record></Cite></EndNote>(Smolen et al., 2008). In addition to describing RA, this chapter will explain the diagnosis of RA, detail some of the commonly used measures of disease activity, describe some of the known risk factors for prognosis and prediction of disease course and outline a common strategy for registration of a new molecule for clinical use.The discussion from this chapter will lend context to the research question, describe the endpoint of interest and identify potential important factors influencing the endpoint of interest.AimsThe objectives of this chapter are: To describe RA as a disease;To present a definition of the disease and the postulated causes of RA;To summarise the risk factors and biomarkers believed to affect occurrence, severity, outcome, and epidemiology of RA, as well as comorbidities and pathophysiology. Rheumatoid ArthritisRheumatoid arthritis is a chronic, destructive, inflammatory, autoimmune disease that has both articular and systemic manifestations ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003). Estimates from the World Health Organisation ADDIN EN.CITE <EndNote><Cite><Author>WHO</Author><Year>2008</Year><RecNum>43</RecNum><DisplayText>(WHO, 2008)</DisplayText><record><rec-number>43</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">43</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>WHO</author></authors></contributors><titles><title><style face="italic" font="default" size="100%">The global burden of disease (updated 2008):</style></title></titles><reprint-edition>Not in File</reprint-edition><dates><year>2008</year><pub-dates><date>2008</date></pub-dates></dates><label>44</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(WHO, 2008) suggest RA affects 23.7 million people worldwide, with a prevalence of 0.8–1.0% of the adult population ( REF _Ref490046042 \h \* MERGEFORMAT Table 2.1). Around twice as many women as men are affected and can start at any age, with a peak incidence between the fourth and sixth decade of life ADDIN EN.CITE <EndNote><Cite><Author>woolf</Author><Year>2010</Year><RecNum>44</RecNum><DisplayText>(Woolf & Pfleger, 2010)</DisplayText><record><rec-number>44</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">44</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Woolf, A.</author><author>Pfleger, B</author></authors></contributors><titles><title>Burden of major musculoskeletal conditions</title><secondary-title>Bulletin of the World Health Organization</secondary-title></titles><reprint-edition>Not in File</reprint-edition><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>45</label><urls></urls></record></Cite></EndNote>(Woolf & Pfleger, 2010). As the disease progresses, patients with RA suffer significant disability and a marked reduction in their quality of life ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003). Approximately one third of patients cease work because of the disease within two years of onset, and this prevalence increases thereafter ADDIN EN.CITE <EndNote><Cite><Author>NCCfCC</Author><Year>2015</Year><RecNum>135</RecNum><DisplayText>(NCCfCC, 2015)</DisplayText><record><rec-number>135</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473700982">135</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>NCCfCC</author></authors></contributors><titles><title>Rheumatoid arthritis: national clinical guideline for management and treatment in adults</title></titles><dates><year>2015</year></dates><publisher>London: Royal College of Physicians</publisher><urls></urls></record></Cite></EndNote>(NCCfCC, 2015). RA directly affects patients’ ability to carry out normal daily tasks, has a significant impact on patient quality of life and has economic implications for patients, their families and the healthcare system ADDIN EN.CITE <EndNote><Cite><Author>Cooper</Author><Year>2000</Year><RecNum>84</RecNum><DisplayText>(Cooper, 2000)</DisplayText><record><rec-number>84</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">84</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cooper, N.J.</author></authors></contributors><auth-address>School of Health Policy and Practice, Elizabeth Fry Building, University of East Anglia, Norwich NR4 7TJ, UK</auth-address><titles><title>Economic burden of rheumatoid arthritis: a systematic review</title><secondary-title>Rheumatology. (Oxford)</secondary-title></titles><pages>28-33</pages><volume>39</volume><number>1</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>Canada</keyword><keyword>Cost of Illness</keyword><keyword>economics</keyword><keyword>Health Care Costs</keyword><keyword>HUMANS</keyword><keyword>methods</keyword><keyword>Netherlands</keyword><keyword>RA</keyword><keyword>Research Design</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Sensitivity and Specificity</keyword><keyword>Sweden</keyword><keyword>United States</keyword></keywords><dates><year>2000</year><pub-dates><date>1/2000</date></pub-dates></dates><isbn>1462-0324</isbn><label>154</label><urls><related-urls><url>;(Cooper, 2000).Descriptive EpidemiologyIncidence and prevalence are two common epidemiologic measures used to describe the frequency of a disease or condition. Prevalence is a measure of how commonly a disease occurs in a population. It measures how much of some disease there is in a population at a particular point in time. The prevalence is calculated by dividing the number of persons with the disease at a particular time point by the number of individuals examined ADDIN EN.CITE <EndNote><Cite><Author>WHO</Author><Year>2008</Year><RecNum>43</RecNum><DisplayText>(WHO, 2008)</DisplayText><record><rec-number>43</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">43</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>WHO</author></authors></contributors><titles><title><style face="italic" font="default" size="100%">The global burden of disease (updated 2008):</style></title></titles><reprint-edition>Not in File</reprint-edition><dates><year>2008</year><pub-dates><date>2008</date></pub-dates></dates><label>44</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(WHO, 2008).Incidence is a measure of the rate of occurrence of new cases of a disease. It is calculated as the number of new cases of a disease in a specified time period (usually a year) divided by the size of the population under consideration who are initially disease free ADDIN EN.CITE <EndNote><Cite><Author>WHO</Author><Year>2008</Year><RecNum>43</RecNum><DisplayText>(WHO, 2008)</DisplayText><record><rec-number>43</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">43</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>WHO</author></authors></contributors><titles><title><style face="italic" font="default" size="100%">The global burden of disease (updated 2008):</style></title></titles><reprint-edition>Not in File</reprint-edition><dates><year>2008</year><pub-dates><date>2008</date></pub-dates></dates><label>44</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(WHO, 2008).Prevalence and incidence are usually expressed as a percentage, if values are low, they are often expressed as the number of cases per 100,000 of the population.There are relatively limited data on trends on incidence and prevalence of RA over time; however studies from North America, Europe and Japan suggest a decline in prevalence and incidence of RA since the 1960’s ADDIN EN.CITE <EndNote><Cite><Author>WHO</Author><Year>2008</Year><RecNum>43</RecNum><DisplayText>(Cross et al., 2014; WHO, 2008)</DisplayText><record><rec-number>43</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">43</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>WHO</author></authors></contributors><titles><title><style face="italic" font="default" size="100%">The global burden of disease (updated 2008):</style></title></titles><reprint-edition>Not in File</reprint-edition><dates><year>2008</year><pub-dates><date>2008</date></pub-dates></dates><label>44</label><urls><related-urls><url><style face="underline" font="default" size="100%"> app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1479063059">177</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cross, M.</author><author>Smith, E.</author><author>Hoy, D.</author><author>Carmona, L.</author><author>Wolfe, F.</author><author>Vos, T.</author><author>Williams, B.</author><author>Gabriel, S.</author><author>Lassere, M.</author><author>Johns, N.</author><author>Buchbinder, R.</author><author>Woolf, A.</author><author>March, L.</author></authors></contributors><titles><title>The global burden of rheumatoid arthritis: estimates from the Global Burden of Disease 2010 study</title><secondary-title>Ann Rheum Dis</secondary-title></titles><periodical><full-title>Ann Rheum Dis</full-title></periodical><pages>1-7</pages><volume>0</volume><dates><year>2014</year></dates><urls></urls></record></Cite></EndNote>(Cross et al., 2014; WHO, 2008). REF _Ref490046042 \h \* MERGEFORMAT Table 2.1 and REF _Ref497232199 \h Table 2.2 show some examples of worldwide prevalence and incidence respectively. In Europe and North America the prevalence ranges from about 0.14–0.65% of the adult population. In the UK this equates to approximately 400,000 people with this condition ADDIN EN.CITE <EndNote><Cite><Author>NCCfCC</Author><Year>2015</Year><RecNum>135</RecNum><DisplayText>(NCCfCC, 2015; NICE, 2009)</DisplayText><record><rec-number>135</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473700982">135</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>NCCfCC</author></authors></contributors><titles><title>Rheumatoid arthritis: national clinical guideline for management and treatment in adults</title></titles><dates><year>2015</year></dates><publisher>London: Royal College of Physicians</publisher><urls></urls></record></Cite><Cite><Author>NICE</Author><Year>2009</Year><RecNum>45</RecNum><record><rec-number>45</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">45</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>NICE</author></authors></contributors><titles><title>Rheumatoid Arthritis National clinical guideline for management and treatment in adults</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2009</year><pub-dates><date>2009</date></pub-dates></dates><label>46</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(NCCfCC, 2015; NICE, 2009). The incidence of the disease is low, a rate of 0.02 to 0.09 per 100 of the general adult population per year. This translates into approximately 12,000 people being diagnosed with new RA each year in the UK ADDIN EN.CITE <EndNote><Cite><Author>NCCfCC</Author><Year>2015</Year><RecNum>135</RecNum><DisplayText>(NCCfCC, 2015; NICE, 2009)</DisplayText><record><rec-number>135</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473700982">135</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>NCCfCC</author></authors></contributors><titles><title>Rheumatoid arthritis: national clinical guideline for management and treatment in adults</title></titles><dates><year>2015</year></dates><publisher>London: Royal College of Physicians</publisher><urls></urls></record></Cite><Cite><Author>NICE</Author><Year>2009</Year><RecNum>45</RecNum><record><rec-number>45</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">45</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>NICE</author></authors></contributors><titles><title>Rheumatoid Arthritis National clinical guideline for management and treatment in adults</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2009</year><pub-dates><date>2009</date></pub-dates></dates><label>46</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(NCCfCC, 2015; NICE, 2009). Table STYLEREF 1 \s 2. SEQ Table \* ARABIC \s 1 1 Prevalence of RA worldwide (cases per 100 inhabitants) (2004)Region Prevalence (95% CI)MaleFemaleGlobalAsia, central Asia, east Asia Pacific, high income Asia, south Asia, southeast Australasia Caribbean Europe, central Europe, eastern Europe, western Latin America, andean Latin America, central Latin America, southern Latin America, tropical North Africa Middle East North America, high incomeOceania Sub-Saharan Africa, centralSub-Saharan Africa, east Sub-Saharan Africa, southSub-Saharan Africa, west 0.13 (0.12, 0.13)0.16 (0.12, 0.21)0.08 (0.08, 0.09)0.22 (0.18, 0.26)0.08 (0.07, 0.09)0.08 (0.07, 0.09)0.26 (0.15, 0.46)0.15 (0.12, 0.18)0.15 (0.11, 0.19)0.14 (0.08, 0.22)0.24 (0.21, 0.28)0.15 (0.10, 0.22)0.14 (0.12, 0.17)0.20 (0.13, 0.30)0.14 (0.13, 0.15)0.09 (0.08, 0.11)0.24 (0.22, 0.27)0.09 (0.05, 0.14)0.12 (0.07, 0.18)0.11 (0.08, 0.14)0.10 (0.09, 0.12)0.10 (0.09, 0.12)0.35 (0.34, 0.37)0.39 (0.30, 0.53)0.24 (0.22, 0.27)0.57 (0.48, 0.69) 0.26 (0.24, 0.29)0.23 (0.21, 0.26)0.66 (0.37, 1.11)0.39 (0.32, 0.49)0.41 (0.31, 0.52)0.38 (0.24, 0.57)0.63 (0.55, 0.75)0.39 (0.25, 0.59)0.40 (0.34, 0.49)0.51 (0.33, 0.78)0.38 (0.35, 0.41)0.24 (0.20, 0.28)0.63 (0.58, 0.70)0.25 (0.15, 0.41)0.30 (0.19, 0.47)0.29 (0.23, 0.37)0.30 (0.26, 0.34)0.28 (0.25, 0.32)Source: ADDIN EN.CITE <EndNote><Cite><Author>WHO</Author><Year>2008</Year><RecNum>43</RecNum><DisplayText>(Cross et al., 2014; WHO, 2008)</DisplayText><record><rec-number>43</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">43</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>WHO</author></authors></contributors><titles><title><style face="italic" font="default" size="100%">The global burden of disease (updated 2008):</style></title></titles><reprint-edition>Not in File</reprint-edition><dates><year>2008</year><pub-dates><date>2008</date></pub-dates></dates><label>44</label><urls><related-urls><url><style face="underline" font="default" size="100%"> app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1479063059">177</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cross, M.</author><author>Smith, E.</author><author>Hoy, D.</author><author>Carmona, L.</author><author>Wolfe, F.</author><author>Vos, T.</author><author>Williams, B.</author><author>Gabriel, S.</author><author>Lassere, M.</author><author>Johns, N.</author><author>Buchbinder, R.</author><author>Woolf, A.</author><author>March, L.</author></authors></contributors><titles><title>The global burden of rheumatoid arthritis: estimates from the Global Burden of Disease 2010 study</title><secondary-title>Ann Rheum Dis</secondary-title></titles><periodical><full-title>Ann Rheum Dis</full-title></periodical><pages>1-7</pages><volume>0</volume><dates><year>2014</year></dates><urls></urls></record></Cite></EndNote>(Cross et al., 2014; WHO, 2008)Table STYLEREF 1 \s 2. SEQ Table \* ARABIC \s 1 2 Incidence rates of RA worldwide (2004) Population Incidence(Annual rate per 100 population) North America USA (Caucasian)USA (native-Americans)0.024–0.0750.09–0.89Northern Europe UK Finland Norway Netherlands 0.02–0.330.040-0.0420.020-0.0250.045Southern Europe France (Lorraine) 0.009Source: ADDIN EN.CITE <EndNote><Cite><Author>Kvien</Author><Year>2004</Year><RecNum>178</RecNum><DisplayText>(Abdel Nasser et al., 1997; Kvien, 2004)</DisplayText><record><rec-number>178</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">178</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kvien, T.K.</author></authors></contributors><titles><title>Epidemiology and Burden of Illness of Rheumatoid Arthritis</title><secondary-title>PharmacoEconomics</secondary-title></titles><pages>1-12</pages><volume>22 Suppl</volume><number>1</number><dates><year>2004</year></dates><urls></urls></record></Cite><Cite><Author>Abdel Nasser</Author><Year>1997</Year><RecNum>401</RecNum><record><rec-number>401</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1523639811">401</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Abdel Nasser, A.</author><author>Rasker, J.</author><author>Valkenberg, H.</author></authors></contributors><titles><title>Epidemiological and Clinical Aspects Relating to the Variability of Rheumatoid Arthritis</title><secondary-title>Seminars in Arthritis and Rheumatism</secondary-title></titles><periodical><full-title>Seminars in Arthritis and Rheumatism</full-title></periodical><pages>123-140</pages><volume>27</volume><number>2</number><dates><year>1997</year></dates><urls></urls></record></Cite></EndNote>(Abdel Nasser et al., 1997; Kvien, 2004)Causes of Rheumatoid ArthritisRA is a heterogeneous autoimmune disease of unknown cause and with a variable clinical expression; however certain antigens are associated with poorer outcomes PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TaWxtYW48L0F1dGhvcj48WWVhcj4yMDAyPC9ZZWFyPjxS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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TaWxtYW48L0F1dGhvcj48WWVhcj4yMDAyPC9ZZWFyPjxS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ADDIN EN.CITE.DATA (Choy & Panayi, 2001; Silman & Pearson, 2002). There appears to be a genetic component to the aetiology of RA with the HLA (human leukocyte antigen) region a major genetic risk area ADDIN EN.CITE <EndNote><Cite><Author>van der Helm</Author><Year>2007</Year><RecNum>46</RecNum><DisplayText>(van der Helm, 2007)</DisplayText><record><rec-number>46</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">46</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>van der Helm, A</author></authors></contributors><titles><title><style face="normal" font="Dutch801BT-Roman" size="100%">Emerging Patterns of Risk Factor Make-Up Enable Subclassification of Rheumatoid Arthritis</style></title><secondary-title>ARTHRITIS & RHEUMATISM</secondary-title></titles><pages>1728-1735</pages><volume>56</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>RA</keyword></keywords><dates><year>2007</year><pub-dates><date>2007</date></pub-dates></dates><label>47</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(van der Helm, 2007). A number of environmental factors have been implicated in the aetiology of RA, but smoking is the only well-established environmental risk factor PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TdWdpeWFtYTwvQXV0aG9yPjxZZWFyPjIwMTA8L1llYXI+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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TdWdpeWFtYTwvQXV0aG9yPjxZZWFyPjIwMTA8L1llYXI+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ADDIN EN.CITE.DATA (Pedersen et al., 2006; Sugiyama et al., 2010). Investigations have failed to associate RA with any specific infectious agent, although it is possible that infection per se may trigger RA in individual cases ADDIN EN.CITE <EndNote><Cite><Author>Silman</Author><Year>2002</Year><RecNum>83</RecNum><DisplayText>(Silman & Pearson, 2002)</DisplayText><record><rec-number>83</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">83</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Silman, A.J.</author><author>Pearson, J.E.</author></authors></contributors><auth-address>ARC Epidemiology Unit, School of Epidemiology & Health Sciences, University of Manchester, UK. alan.silman@man.ac.uk</auth-address><titles><title>Epidemiology and genetics of rheumatoid arthritis</title><secondary-title>Arthritis Res</secondary-title></titles><pages>S265-S272</pages><volume>4 Suppl 3</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>Environment</keyword><keyword>epidemiology</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>Genetic Predisposition to Disease</keyword><keyword>genetics</keyword><keyword>HUMANS</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2002</year><pub-dates><date>2002</date></pub-dates></dates><isbn>1465-9905</isbn><label>153</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(Silman & Pearson, 2002).DiagnosisAletaha ADDIN EN.CITE <EndNote><Cite><Author>Aletaha</Author><Year>2010</Year><RecNum>92</RecNum><DisplayText>(Aletaha, 2010)</DisplayText><record><rec-number>92</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">92</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Aletaha, D.</author></authors></contributors><titles><title>2010 Rheumatoid Arthritis Classification Criteria</title><secondary-title>ARTHRITIS & RHEUMATISM
An American College of Rheumatology/European League Against Rheumatism Collaborative Initiative</secondary-title></titles><pages>2569-2581</pages><volume>62</volume><number>9</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>classification</keyword><keyword>Rheumatology</keyword><keyword>RA</keyword><keyword>Disease</keyword><keyword>diagnosis</keyword><keyword>Joints</keyword></keywords><dates><year>2010</year></dates><label>162</label><urls></urls></record></Cite></EndNote>(Aletaha, 2010) discussed how there is no single laboratory test or procedure that is used to diagnose RA. A set of criteria were developed in 1956 by a committee of the American Rheumatism Association. These criteria were developed by 5 committee members based on their clinical experiences, as well as a review of the then current epidemiologic data and a review of 332 cases of RA. Eleven criteria and 9 exclusions were proposed which diagnosed three grades of RA: definite, probable and possible. Aletaha et al described how they were revised in 1958 and in 1966, but widespread use diminished due to the cumbersome nature of the revisions. Aletaha et al’s final revision was made in 1987 by 14 experts representing the Rheumatoid Arthritis Criteria Subcommittee of the Diagnostic and Therapeutic Criteria Committee of the American Rheumatism Association, using data from 262 patients and 262 control subjects. These adapted criteria for diagnosing RA achieved around 90% specificity and sensitivity and remained in common use ADDIN EN.CITE <EndNote><Cite><Author>Arnett</Author><Year>1988</Year><RecNum>86</RecNum><DisplayText>(Arnett et al., 1988)</DisplayText><record><rec-number>86</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">86</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Arnett, F.C.</author><author>Edworthy, S.M.</author><author>Bloch, D.A.</author><author>McShane, D.J.</author><author>Fries, J.F.</author><author>Cooper, N.S.</author><author>Healey, L.A.</author><author>Kaplan, S.R.</author><author>Liang, M.H.</author><author>Luthra, H.S.</author><author>.</author></authors></contributors><auth-address>American Rheumatism Association, Atlanta, GA 30329</auth-address><titles><title>The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>315-324</pages><volume>31</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>classification</keyword><keyword>diagnosis</keyword><keyword>HUMANS</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatoid Factor</keyword><keyword>Rheumatoid Nodule</keyword><keyword>RHEUMATOID-NODULE</keyword><keyword>Rheumatology</keyword><keyword>Societies,Medical</keyword><keyword>Terminology as Topic</keyword><keyword>trends</keyword><keyword>United States</keyword></keywords><dates><year>1988</year><pub-dates><date>3/1988</date></pub-dates></dates><isbn>0004-3591</isbn><label>156</label><urls><related-urls><url>;(Arnett et al., 1988) (For further description of specificity and sensitivity see Section 5.3). In 2010, a collaborative initiative between the American College of Rheumatology PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+QWxldGFoYTwvQXV0aG9yPjxZ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ADDIN EN.CITE PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+QWxldGFoYTwvQXV0aG9yPjxZ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ADDIN EN.CITE.DATA (Aletaha et al.) and the European League Against Rheumatism PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+QWxldGFoYTwvQXV0aG9yPjxZ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ADDIN EN.CITE PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+QWxldGFoYTwvQXV0aG9yPjxZ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ADDIN EN.CITE.DATA (Aletaha et al.) developed a new approach to classifying RA with an increased focus on identifying early RA ADDIN EN.CITE <EndNote><Cite><Author>Aletaha</Author><Year>2010</Year><RecNum>92</RecNum><DisplayText>(Aletaha, 2010)</DisplayText><record><rec-number>92</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">92</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Aletaha, D.</author></authors></contributors><titles><title>2010 Rheumatoid Arthritis Classification Criteria</title><secondary-title>ARTHRITIS & RHEUMATISM
An American College of Rheumatology/European League Against Rheumatism Collaborative Initiative</secondary-title></titles><pages>2569-2581</pages><volume>62</volume><number>9</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>classification</keyword><keyword>Rheumatology</keyword><keyword>RA</keyword><keyword>Disease</keyword><keyword>diagnosis</keyword><keyword>Joints</keyword></keywords><dates><year>2010</year></dates><label>162</label><urls></urls></record></Cite></EndNote>(Aletaha, 2010). Since the data for this research predates these new criteria, these are not discussed further here.The 7 criteria identified by the 1987 committee ADDIN EN.CITE <EndNote><Cite><Author>Arnett</Author><Year>1988</Year><RecNum>86</RecNum><DisplayText>(Arnett et al., 1988)</DisplayText><record><rec-number>86</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">86</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Arnett, F.C.</author><author>Edworthy, S.M.</author><author>Bloch, D.A.</author><author>McShane, D.J.</author><author>Fries, J.F.</author><author>Cooper, N.S.</author><author>Healey, L.A.</author><author>Kaplan, S.R.</author><author>Liang, M.H.</author><author>Luthra, H.S.</author><author>.</author></authors></contributors><auth-address>American Rheumatism Association, Atlanta, GA 30329</auth-address><titles><title>The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>315-324</pages><volume>31</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>classification</keyword><keyword>diagnosis</keyword><keyword>HUMANS</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatoid Factor</keyword><keyword>Rheumatoid Nodule</keyword><keyword>RHEUMATOID-NODULE</keyword><keyword>Rheumatology</keyword><keyword>Societies,Medical</keyword><keyword>Terminology as Topic</keyword><keyword>trends</keyword><keyword>United States</keyword></keywords><dates><year>1988</year><pub-dates><date>3/1988</date></pub-dates></dates><isbn>0004-3591</isbn><label>156</label><urls><related-urls><url>;(Arnett et al., 1988) are:morning stiffness in and around the joints lasting for at least 1 hour before maximal improvement, arthritis of three or more joint areas observed by a physician, arthritis of hand joints with at least one joint area swollen (in wrist, metacarpophalangeal (MCP) or proximal interphalangeal (PIP) joint), symmetric arthritis, rheumatoid nodules (subcutaneous nodules over bony prominences that are observed by a physician), presence of rheumatoid factor (RF) in serum, and radiographic changes (including erosions or bony decalcification localised in or adjacent to the involved joints).The committee ADDIN EN.CITE <EndNote><Cite><Author>Arnett</Author><Year>1988</Year><RecNum>86</RecNum><DisplayText>(Arnett et al., 1988)</DisplayText><record><rec-number>86</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">86</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Arnett, F.C.</author><author>Edworthy, S.M.</author><author>Bloch, D.A.</author><author>McShane, D.J.</author><author>Fries, J.F.</author><author>Cooper, N.S.</author><author>Healey, L.A.</author><author>Kaplan, S.R.</author><author>Liang, M.H.</author><author>Luthra, H.S.</author><author>.</author></authors></contributors><auth-address>American Rheumatism Association, Atlanta, GA 30329</auth-address><titles><title>The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>315-324</pages><volume>31</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>classification</keyword><keyword>diagnosis</keyword><keyword>HUMANS</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatoid Factor</keyword><keyword>Rheumatoid Nodule</keyword><keyword>RHEUMATOID-NODULE</keyword><keyword>Rheumatology</keyword><keyword>Societies,Medical</keyword><keyword>Terminology as Topic</keyword><keyword>trends</keyword><keyword>United States</keyword></keywords><dates><year>1988</year><pub-dates><date>3/1988</date></pub-dates></dates><isbn>0004-3591</isbn><label>156</label><urls><related-urls><url>;(Arnett et al., 1988) stated that for classification purposes, a patient is said to have rheumatoid arthritis if he or she satisfied at least four of the above criteria. The first four must have been present for at least six weeks. The physical effects of early disease are often subtle, and may overlap with symptoms of other conditions, making the diagnosis of RA challenging ADDIN EN.CITE <EndNote><Cite><Author>O'Dell</Author><Year>2004</Year><RecNum>63</RecNum><DisplayText>(O'Dell, 2004)</DisplayText><record><rec-number>63</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">63</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>O'Dell, J.</author></authors></contributors><auth-address>Department of Internal Medicine, University of Nebraska Medical Center, Omaha 68198-3025, USA. jrodell@unmc.edu</auth-address><titles><title>Therapeutic strategies for rheumatoid arthritis</title><secondary-title>The New England journal of medicine</secondary-title></titles><pages>2591-2602</pages><volume>350</volume><number>25</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Adrenal-Cortex-Hormones)</keyword><keyword>0 (Anti-Inflammatory-Agents-Non-Steroidal)</keyword><keyword>0 (Antirheumatic-Agents)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>372-75-8 (Citrulline)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ADRENAL-CORTEX-HORMONES/TU (therapeutic use)</keyword><keyword>ANTI-INFLAMMATORY-AGENTS-NON-STEROIDAL/TU (therapeutic use)</keyword><keyword>ANTIRHEUMATIC-AGENTS/*TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/DI (diagnosis),DT (drug therapy),*TH (therapy)</keyword><keyword>AUTOANTIBODIES/BL (blood)</keyword><keyword>CITRULLINE/IM (immunology)</keyword><keyword>COMORBIDITY</keyword><keyword>DRUG-THERAPY-COMBINATION</keyword><keyword>HUMANS</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/BL (blood)</keyword></keywords><dates><year>2004</year><pub-dates><date>6/17/2004</date></pub-dates></dates><label>64</label><urls></urls><access-date>2006</access-date></record></Cite></EndNote>(O'Dell, 2004). However, early diagnosis and treatment are important, as potentially irreversible joint damage occurs early in the course of disease PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MaW5kcXZpc3Q8L0F1dGhvcj48WWVhcj4yMDAzPC9ZZWFy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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MaW5kcXZpc3Q8L0F1dGhvcj48WWVhcj4yMDAzPC9ZZWFy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ADDIN EN.CITE.DATA (Lindqvist et al., 2003). Typically the onset of RA emerges insidiously over several weeks or months ADDIN EN.CITE <EndNote><Cite><Author>Ruffing</Author><Year>2016</Year><RecNum>136</RecNum><DisplayText>(Ruffing & Clifton, 2016)</DisplayText><record><rec-number>136</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">136</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Ruffing, V.</author><author>Clifton, O.B. </author></authors></contributors><titles><title>Rheumatoid arthritis – clinical presentation </title></titles><dates><year>2016</year><pub-dates><date>13 January 2016</date></pub-dates></dates><publisher>The John Hopkins Arthritis Center</publisher><urls><related-urls><url>;(Ruffing & Clifton, 2016). The first signs observed are usually symmetrical joint swelling and increases in RF, Erythrocyte Sedimentation Rate (ESR), C - reactive protein (CRP), anaemia and high platelet counts (thrombocytosis) ADDIN EN.CITE <EndNote><Cite><Author>Ruffing</Author><Year>2016</Year><RecNum>136</RecNum><DisplayText>(Ruffing & Clifton, 2016)</DisplayText><record><rec-number>136</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">136</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Ruffing, V.</author><author>Clifton, O.B. </author></authors></contributors><titles><title>Rheumatoid arthritis – clinical presentation </title></titles><dates><year>2016</year><pub-dates><date>13 January 2016</date></pub-dates></dates><publisher>The John Hopkins Arthritis Center</publisher><urls><related-urls><url>;(Ruffing & Clifton, 2016). The severity and to some degree the clinical course of RA was categorised into ordinal 4 classes by Hochberg ADDIN EN.CITE <EndNote><Cite><Author>Hochberg</Author><Year>1992</Year><RecNum>97</RecNum><DisplayText>(Hochberg et al., 1992)</DisplayText><record><rec-number>97</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">97</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Hochberg, M.C.</author><author>Chang, R.W.</author><author>Dwosh, I.</author><author>Lindsey, S.</author><author>Pincus, T.</author><author>Wolfe, F.</author></authors></contributors><auth-address>American College of Rheumatology, Atlanta, GA 30329</auth-address><titles><title>The American College of Rheumatology 1991 revised criteria for the classification of global functional status in rheumatoid arthritis</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>498-502</pages><volume>35</volume><number>5</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ACTIVITIES-OF-DAILY-LIVING</keyword><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>ARTHRITIS-RHEUMATOID/CL (classification),*PP (physiopathology)</keyword><keyword>classification</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>JOINTS/PP (physiopathology)</keyword><keyword>MALE</keyword><keyword>methods</keyword><keyword>MIDDLE-AGED</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>United States</keyword><keyword>UNITED-STATES</keyword></keywords><dates><year>1992</year><pub-dates><date>5/1992</date></pub-dates></dates><label>167</label><urls></urls><access-date>2006</access-date></record></Cite></EndNote>(Hochberg et al., 1992) ranging from self-limiting to chronic progressive disease. Hochberg et al identified 3 classes of daily activity: self-care (dressing, feeding, bathing, grooming and toileting), vocational (work, school, homemaking) and avocational (recreational, leisure). Class I disease is the mildest disease state where the patient is completely able to perform usual activities of daily living. With Class II disease the patient is able to perform usual self-care and vocational activities but is limited in avocational activities. Patients with Class III disease are able to perform usual self-care activities but are limited in vocational and avocational activities. Class is the most advanced disease state where patients have limited ability in all usual activities of daily living.Manifestations and co-morbidities associated with rheumatoid arthritisRA affects the small joints of the hands and feet, as well as larger joints throughout the body. In a healthy joint, the delicate internal lining of the joint, the synovium, covers all the intra-articular structures except for cartilage and areas of exposed bone ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003). In RA, the synovium is invaded by immune cells, which multiply and express damaging enzymes, forming an aggressive front of inflammatory tissue known as pannus ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003). This digests the extracellular matrix and destroys the surrounding articular structure of the joint, leading to potentially irreversible cartilage damage, joint damage and bone erosion PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TbW9sZW48L0F1dGhvcj48WWVhcj4yMDAzPC9ZZWFyPjxS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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TbW9sZW48L0F1dGhvcj48WWVhcj4yMDAzPC9ZZWFyPjxS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ADDIN EN.CITE.DATA (Aletaha et al., 2006; Firestein, 2003; Smolen & Steiner, 2003) ( REF _Ref490046258 \h \* MERGEFORMAT Figure 2.1). This expresses as joint pain, stiffness and swelling, which can cause severe impairment in the movement of joints and result in significant disability ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003). Figure STYLEREF 1 \s 2. SEQ Figure \* ARABIC \s 1 1 Inflamed Synovium (Pannus) Causes Joint DestructionFigure Downloaded from ( By US gov - US gov, Public Domain.). Last accessed 31/5/2017In addition to destruction of joints, RA is associated with a wide range of extra-articular and systemic manifestations that lead to further functional impairment and a reduction in life expectancy ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003). As the systemic effects of RA progress, patients have a greater risk of dying from urogenital, gastrointestinal, respiratory and cardiovascular diseases, infections and cancers, compared with the general population ADDIN EN.CITE <EndNote><Cite><Author>Sihvonen</Author><Year>2004</Year><RecNum>51</RecNum><DisplayText>(Sihvonen et al., 2004)</DisplayText><record><rec-number>51</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">51</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Sihvonen, S.</author><author>Korpela, M.</author><author>Laippala, P.</author><author>Mustonen, J.</author><author>Pasternack, A.</author></authors></contributors><auth-address>Department of Internal Medicine, Tampere University Hospital, FIN-33521, Finland. susanna.sihvonen@fimnet.fi</auth-address><titles><title>Death rates and causes of death in patients with rheumatoid arthritis: a population-based study</title><secondary-title>Scandinavian journal of rheumatology</secondary-title></titles><periodical><full-title>Scandinavian journal of rheumatology</full-title></periodical><pages>221-227</pages><volume>33</volume><number>4</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>ARTHRITIS-RHEUMATOID/*CO (complications),*MO (mortality)</keyword><keyword>CASE-CONTROL-STUDIES</keyword><keyword>CAUSE-OF-DEATH/*</keyword><keyword>CROSS-SECTIONAL-STUDIES</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>MORTALITY/TD (trends)</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2004</year><pub-dates><date>2004</date></pub-dates></dates><label>52</label><urls></urls><access-date>2009</access-date></record></Cite></EndNote>(Sihvonen et al., 2004).Rindfleisch and Muller ADDIN EN.CITE <EndNote><Cite><Author>Rindfleisch</Author><Year>2005</Year><RecNum>81</RecNum><DisplayText>(Rindfleisch & Muller, 2005)</DisplayText><record><rec-number>81</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">81</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rindfleisch, J.A.</author><author>Muller, D.</author></authors></contributors><auth-address>Department of Family Medicine, University of Wisconsin-Madison, Madison, Wisconsin, USA</auth-address><titles><title>Diagnosis and management of rheumatoid arthritis</title><secondary-title>Am. Fam. Physician</secondary-title></titles><pages>1037-1047</pages><volume>72</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>administration & dosage</keyword><keyword>ADULT</keyword><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>COMORBIDITY</keyword><keyword>diagnosis</keyword><keyword>Diagnosis,Differential</keyword><keyword>drug therapy</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>Middle Aged</keyword><keyword>physiopathology</keyword><keyword>PREVALENCE</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Risk Factors</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2005</year><pub-dates><date>9/15/2005</date></pub-dates></dates><isbn>0002-838X</isbn><label>83</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(Rindfleisch & Muller, 2005) identify a host of the most common clinical complications associated with RA. The authors state that lymphomas and leukaemias are two to three times more common in patients with rheumatoid arthritis. RA patients are at increased risk for various solid tumours, although genitourinary cancer risk is reduced in rheumatoid arthritis, perhaps in the view of the authors, because of the common use of non-steroidal anti-inflammatory drugs. A third of patients may have asymptomatic fluid around the heart (pericardial effusion) at diagnosis. Lung nodules can coexist with cancers and form cavitary lesions, cricoarytenoid joint inflammation can arise with hoarseness and laryngeal pain and lung inflammation (pleuritis) is present in 20 percent of patients at onset of disease ADDIN EN.CITE <EndNote><Cite><Author>Rindfleisch</Author><Year>2005</Year><RecNum>81</RecNum><DisplayText>(Rindfleisch & Muller, 2005)</DisplayText><record><rec-number>81</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">81</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rindfleisch, J.A.</author><author>Muller, D.</author></authors></contributors><auth-address>Department of Family Medicine, University of Wisconsin-Madison, Madison, Wisconsin, USA</auth-address><titles><title>Diagnosis and management of rheumatoid arthritis</title><secondary-title>Am. Fam. Physician</secondary-title></titles><pages>1037-1047</pages><volume>72</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>administration & dosage</keyword><keyword>ADULT</keyword><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>COMORBIDITY</keyword><keyword>diagnosis</keyword><keyword>Diagnosis,Differential</keyword><keyword>drug therapy</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>Middle Aged</keyword><keyword>physiopathology</keyword><keyword>PREVALENCE</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Risk Factors</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2005</year><pub-dates><date>9/15/2005</date></pub-dates></dates><isbn>0002-838X</isbn><label>83</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(Rindfleisch & Muller, 2005). Anaemia correlates with ESR and disease activity, 75 percent of patients have chronic anaemia, although one fourth of patients respond to iron therapy. Males with high rheumatoid factor titres, treatment with steroids, and higher numbers of disease-modifying anti-rheumatic drugs (DMARD) prescribed have an increased risk of developing vasculitis. Vasculitis is associated with increased risk of myocardial infarction, and forms include distal arteritis, pericarditis, peripheral neuropathy, cutaneous lesions, arteritis of viscera, and coronary arteritis ADDIN EN.CITE <EndNote><Cite><Author>Rindfleisch</Author><Year>2005</Year><RecNum>81</RecNum><DisplayText>(Rindfleisch & Muller, 2005)</DisplayText><record><rec-number>81</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">81</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rindfleisch, J.A.</author><author>Muller, D.</author></authors></contributors><auth-address>Department of Family Medicine, University of Wisconsin-Madison, Madison, Wisconsin, USA</auth-address><titles><title>Diagnosis and management of rheumatoid arthritis</title><secondary-title>Am. Fam. Physician</secondary-title></titles><pages>1037-1047</pages><volume>72</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>administration & dosage</keyword><keyword>ADULT</keyword><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>COMORBIDITY</keyword><keyword>diagnosis</keyword><keyword>Diagnosis,Differential</keyword><keyword>drug therapy</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>Middle Aged</keyword><keyword>physiopathology</keyword><keyword>PREVALENCE</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Risk Factors</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2005</year><pub-dates><date>9/15/2005</date></pub-dates></dates><isbn>0002-838X</isbn><label>83</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(Rindfleisch & Muller, 2005). PathophysiologyMultiple pathogenic mechanisms appear to be implicated in RA ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003). As an autoimmune disease, the tissue damage associated with RA can be triggered as a result of the body mounting an immune response to antigens on its own cells ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003). In the case of RA, the targeted antigens are so expressed in the body that they cannot be eliminated, so, instead of reaching resolution, their continual presence produces a chronic and persistent immune response ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003). Many autoantibody systems that could participate in inflammatory joint disease are now recognised in RA. Some, like rheumatoid factor (RF), are produced by rheumatoid synovial B-cells ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003). The presence of RF in serum, however, is still not specific for RA, although patients with active RA have significantly higher serum RF levels compared with those with inactive disease PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5BdGVzPC9BdXRob3I+PFllYXI+MjAwNzwvWWVhcj48UmVj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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5BdGVzPC9BdXRob3I+PFllYXI+MjAwNzwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Ates et al., 2007).The presence of antibodies directed against proteins containing citrulline (CCPs) appears to be more useful to correctly diagnose patients with RA PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TaWx2ZWlyYTwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TaWx2ZWlyYTwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+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ADDIN EN.CITE.DATA (Silveira et al., 2007).Activation of immunity is thought to occur early in the course of RA and may be a key pathogenic mechanism in initiating synovial inflammation PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5DaG95PC9BdXRob3I+PFllYXI+MjAwMTwvWWVhcj48UmVj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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5DaG95PC9BdXRob3I+PFllYXI+MjAwMTwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Choy & Panayi, 2001). Many of the effects of RA are assisted by pro-inflammatory cytokines – a family of proteins that regulate survival, growth, and functions of cells, particularly those in the immune system PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5DaG95PC9BdXRob3I+PFllYXI+MjAwMTwvWWVhcj48UmVj
TnVtPjU0PC9SZWNOdW0+PERpc3BsYXlUZXh0PihDaG95ICZhbXA7IFBhbmF5aSwgMjAwMSk8L0Rp
c3BsYXlUZXh0PjxyZWNvcmQ+PHJlYy1udW1iZXI+NTQ8L3JlYy1udW1iZXI+PGZvcmVpZ24ta2V5
cz48a2V5IGFwcD0iRU4iIGRiLWlkPSI1YTk5d2RhYXp3ZHh0M2V4NXBmcHBhZTNmc3Bzd3Z0ZnJ4
cmQiIHRpbWVzdGFtcD0iMCI+NTQ8L2tleT48L2ZvcmVpZ24ta2V5cz48cmVmLXR5cGUgbmFtZT0i
Sm91cm5hbCBBcnRpY2xlIj4xNzwvcmVmLXR5cGU+PGNvbnRyaWJ1dG9ycz48YXV0aG9ycz48YXV0
aG9yPkNob3ksIEUuSC48L2F1dGhvcj48YXV0aG9yPlBhbmF5aSwgRy5TLjwvYXV0aG9yPjwvYXV0
aG9ycz48L2NvbnRyaWJ1dG9ycz48YXV0aC1hZGRyZXNzPkRlcGFydG1lbnQgb2YgUmhldW1hdG9s
b2d5LCBHdXkmYXBvcztzLCBLaW5nJmFwb3M7cywgYW5kIFN0LiBUaG9tYXMmYXBvczsgSG9zcGl0
YWxzIFNjaG9vbCBvZiBNZWRpY2luZSwgS2luZyZhcG9zO3MgQ29sbGVnZSwgTG9uZG9uLCBVbml0
ZWQgS2luZ2RvbTwvYXV0aC1hZGRyZXNzPjx0aXRsZXM+PHRpdGxlPkN5dG9raW5lIHBhdGh3YXlz
IGFuZCBqb2ludCBpbmZsYW1tYXRpb24gaW4gcmhldW1hdG9pZCBhcnRocml0aXM8L3RpdGxlPjxz
ZWNvbmRhcnktdGl0bGU+VGhlIE5ldyBFbmdsYW5kIEpvdXJuYWwgb2YgTWVkaWNpbmU8L3NlY29u
ZGFyeS10aXRsZT48L3RpdGxlcz48cGFnZXM+OTA3LTkxNjwvcGFnZXM+PHZvbHVtZT4zNDQ8L3Zv
bHVtZT48bnVtYmVyPjEyPC9udW1iZXI+PHJlcHJpbnQtZWRpdGlvbj5Ob3QgaW4gRmlsZTwvcmVw
cmludC1lZGl0aW9uPjxrZXl3b3Jkcz48a2V5d29yZD4wIChBbnRpYm9kaWVzLU1vbm9jbG9uYWwp
PC9rZXl3b3JkPjxrZXl3b3JkPjAgKEFudGlyaGV1bWF0aWMtQWdlbnRzKTwva2V5d29yZD48a2V5
d29yZD4wIChDeXRva2luZXMpPC9rZXl3b3JkPjxrZXl3b3JkPjAgKEltbXVub2dsb2J1bGluLUcp
PC9rZXl3b3JkPjxrZXl3b3JkPjAgKGluZmxpeGltYWIpPC9rZXl3b3JkPjxrZXl3b3JkPjAgKFJl
Y2VwdG9ycy1DeXRva2luZSk8L2tleXdvcmQ+PGtleXdvcmQ+MCAoUmVjZXB0b3JzLVR1bW9yLU5l
Y3Jvc2lzLUZhY3Rvcik8L2tleXdvcmQ+PGtleXdvcmQ+MTg1MjQzLTY5LTAgKFRORlItRmMtZnVz
aW9uLXByb3RlaW4pPC9rZXl3b3JkPjxrZXl3b3JkPkFOVElCT0RJRVMtTU9OT0NMT05BTC9UVSAo
dGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5d29yZD5BTlRJUkhFVU1BVElDLUFHRU5UUy9U
VSAodGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5d29yZD5BUlRIUklUSVMtUkhFVU1BVE9J
RC8qRFQgKGRydWcgdGhlcmFweSksKklNIChpbW11bm9sb2d5KTwva2V5d29yZD48a2V5d29yZD5D
WVRPS0lORVMvKkFJIChhbnRhZ29uaXN0cyAmYW1wOyBpbmhpYml0b3JzKSxNRSAobWV0YWJvbGlz
bSksKlBIIChwaHlzaW9sb2d5KSxUVSAodGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5d29y
ZD5IVU1BTlM8L2tleXdvcmQ+PGtleXdvcmQ+SU1NVU5PR0xPQlVMSU4tRy9UVSAodGhlcmFwZXV0
aWMgdXNlKTwva2V5d29yZD48a2V5d29yZD5JTkZMQU1NQVRJT04vSU0gKGltbXVub2xvZ3kpPC9r
ZXl3b3JkPjxrZXl3b3JkPlJFQ0VQVE9SUy1DWVRPS0lORS9BSSAoYW50YWdvbmlzdHMgJmFtcDsg
aW5oaWJpdG9ycyk8L2tleXdvcmQ+PGtleXdvcmQ+UkVDRVBUT1JTLVRVTU9SLU5FQ1JPU0lTLUZB
Q1RPUi9UVSAodGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5d29yZD5SaGV1bWF0b2lkIGFy
dGhyaXRpczwva2V5d29yZD48a2V5d29yZD5TSUdOQUwtVFJBTlNEVUNUSU9OPC9rZXl3b3JkPjwv
a2V5d29yZHM+PGRhdGVzPjx5ZWFyPjIwMDE8L3llYXI+PHB1Yi1kYXRlcz48ZGF0ZT4zLzIyLzIw
MDE8L2RhdGU+PC9wdWItZGF0ZXM+PC9kYXRlcz48bGFiZWw+NTU8L2xhYmVsPjx1cmxzPjwvdXJs
cz48YWNjZXNzLWRhdGU+MjAwMzwvYWNjZXNzLWRhdGU+PC9yZWNvcmQ+PC9DaXRlPjwvRW5kTm90
ZT5=
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5DaG95PC9BdXRob3I+PFllYXI+MjAwMTwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Choy & Panayi, 2001). Activated T-cells in the synovium stimulate monocytes, macrophages and synovial fibroblasts to produce various pro-inflammatory cytokines, and drive the continued recruitment of immune cells and result in an accumulation of T-cells in the inflamed synovium ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003). Activated T-cells also stimulate B-cells to produce immunoglobulins, including RF, and drive the expression of damaging enzymes that initiate the destructive phase of the disease ADDIN EN.CITE <EndNote><Cite><Author>Firestein</Author><Year>2003</Year><RecNum>49</RecNum><DisplayText>(Firestein, 2003)</DisplayText><record><rec-number>49</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">49</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Firestein, G</author></authors></contributors><auth-address>Division of Rheumatology, Allergy and Immunology, School of Medicine, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0656, USA. gfirestein@ucsd.edu</auth-address><titles><title>Evolving concepts of rheumatoid arthritis</title><secondary-title>Nature</secondary-title></titles><pages>356-361</pages><volume>423</volume><number>6937</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Antigen-Antibody-Complex)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>0 (Cytokines)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ANTIGEN-ANTIBODY-COMPLEX/IM (immunology)</keyword><keyword>ARTHRITIS-RHEUMATOID/*IM (immunology),*PA (pathology)</keyword><keyword>AUTOANTIBODIES/IM (immunology)</keyword><keyword>CYTOKINES/GE (genetics),IM (immunology),ME (metabolism)</keyword><keyword>EUROPE</keyword><keyword>GENE-EXPRESSION-REGULATION</keyword><keyword>HUMANS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/IM (immunology)</keyword><keyword>T-LYMPHOCYTES/IM (immunology)</keyword></keywords><dates><year>2003</year><pub-dates><date>5/15/2003</date></pub-dates></dates><label>50</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Firestein, 2003).Risk Factors for RAA significant amount of research has been performed to identify risk factors for the incidence and severity of RA. Alamanos and Drosos ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005) highlight that there is evidence that the occurrence and severity of RA are related to genetic factors. The author goes on to state that twin and family studies offer a strong suggestion of that risk of the disease is influenced by shared genetic factors, and studies in patients with advanced disease suggest an association with alleles coding to a shared ‘rheumatoid epitope’. The relationship of rheumatoid epitope, tumour necrosis factor alleles or other genetic factor related to the risk or severity of RA still remains unclear, although the differences seen in the various populations seen in REF _Ref490046042 \h \* MERGEFORMAT Table 2.1 for example could be explained partly by genetic variation, ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005).The incidence of RA is higher in women than in men. The ratio varies in studies from 2:1 to 3:1 ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005). A British study by Symmons ADDIN EN.CITE <EndNote><Cite><Author>Symmons</Author><Year>1994</Year><RecNum>95</RecNum><DisplayText>(Symmons et al., 1994)</DisplayText><record><rec-number>95</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">95</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Symmons, D.P.</author><author>Barrett, E.M.</author><author>Bankhead, C.R.</author><author>Scott, D.G.</author><author>Silman, A.J.</author></authors></contributors><auth-address>ARC Epidemiology Research Unit, University of Manchester</auth-address><titles><title>The incidence of rheumatoid arthritis in the United Kingdom: results from the Norfolk Arthritis Register</title><secondary-title>British journal of rheumatology</secondary-title></titles><pages>735-739</pages><volume>33</volume><number>8</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADOLESCENT</keyword><keyword>ADULT</keyword><keyword>AGE-DISTRIBUTION</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology)</keyword><keyword>blood</keyword><keyword>epidemiology</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>GREAT-BRITAIN/EP (epidemiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>PRIMARY-HEALTH-CARE</keyword><keyword>PROSPECTIVE-STUDIES</keyword><keyword>RA</keyword><keyword>REGISTRIES</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>SEX-DISTRIBUTION</keyword></keywords><dates><year>1994</year><pub-dates><date>8/1994</date></pub-dates></dates><label>165</label><urls></urls><access-date>2009</access-date></record></Cite></EndNote>(Symmons et al., 1994) estimated the RA incidence for females of 0.036% compared to a 0.014% rates for males. The age of disease onset reaches a peak in the fifth or sixth decade of life ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005). A study of RA in North America highlights an age and gender interaction in the incidence of RA ADDIN EN.CITE <EndNote><Cite><Author>Doran</Author><Year>2002</Year><RecNum>98</RecNum><DisplayText>(Doran et al., 2002)</DisplayText><record><rec-number>98</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">98</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Doran, Michele</author><author>Pond, Gregory</author><author>Crowson, Cynthia</author><author>Fallon, W.</author><author>Gabriel, Sherine</author></authors></contributors><auth-address>Mayo Clinic, Rochester, Minnesota 55905, USA</auth-address><titles><title>Trends in incidence and mortality in rheumatoid arthritis in Rochester, Minnesota, over a forty-year period</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and rheumatism</full-title></periodical><pages>625-631</pages><volume>46</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGE-DISTRIBUTION</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>analysis</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),MO (mortality)</keyword><keyword>COHORT-STUDIES</keyword><keyword>epidemiology</keyword><keyword>etiology</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>LINEAR-MODELS</keyword><keyword>MALE</keyword><keyword>methods</keyword><keyword>MIDDLE-AGED</keyword><keyword>MINNESOTA/EP (epidemiology)</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>SEX-DISTRIBUTION</keyword><keyword>SURVIVAL-ANALYSIS</keyword><keyword>trends</keyword><keyword>United States</keyword><keyword>UNITED-STATES</keyword></keywords><dates><year>2002</year><pub-dates><date>3/2002</date></pub-dates></dates><label>168</label><urls></urls><access-date>2003</access-date></record></Cite></EndNote>(Doran et al., 2002). Males tend to experience a very low incidence up to 34 years of age, after which there is a steady progression through to 85 years. In females, the incidence of RA rises through ages 55 to 64 after which incidence declines. ADDIN EN.CITE <EndNote><Cite><Author>Doran</Author><Year>2002</Year><RecNum>98</RecNum><DisplayText>(Doran et al., 2002)</DisplayText><record><rec-number>98</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">98</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Doran, Michele</author><author>Pond, Gregory</author><author>Crowson, Cynthia</author><author>Fallon, W.</author><author>Gabriel, Sherine</author></authors></contributors><auth-address>Mayo Clinic, Rochester, Minnesota 55905, USA</auth-address><titles><title>Trends in incidence and mortality in rheumatoid arthritis in Rochester, Minnesota, over a forty-year period</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and rheumatism</full-title></periodical><pages>625-631</pages><volume>46</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGE-DISTRIBUTION</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>analysis</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),MO (mortality)</keyword><keyword>COHORT-STUDIES</keyword><keyword>epidemiology</keyword><keyword>etiology</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>LINEAR-MODELS</keyword><keyword>MALE</keyword><keyword>methods</keyword><keyword>MIDDLE-AGED</keyword><keyword>MINNESOTA/EP (epidemiology)</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>SEX-DISTRIBUTION</keyword><keyword>SURVIVAL-ANALYSIS</keyword><keyword>trends</keyword><keyword>United States</keyword><keyword>UNITED-STATES</keyword></keywords><dates><year>2002</year><pub-dates><date>3/2002</date></pub-dates></dates><label>168</label><urls></urls><access-date>2003</access-date></record></Cite></EndNote>(Doran et al., 2002).Smoking appears to influence both the risk of developing the disease as well as the severity although there is some variation in the extent ( REF _Ref460428922 \h \* MERGEFORMAT Table 2.3). The postulated association is most clear for heavy smokers ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005).Table STYLEREF 1 \s 2. SEQ Table \* ARABIC \s 1 3 Association between rheumatoid arthritis and cigarette smokingStudyCases ControlsSmoking GroupOdds Ratio (95% CI)Hutchinson et al23923941-50 Pack yearsEver smokedNever smoked13.5 (2.9–63.4)1.8 (1.2–2.2)1.0Symmons et al165178Current smokerEx smokerNever smoked0.95 (0.6-1.6)1.7 (0.95-3.1)1.0Karlson et al7697370,000Current smokerEx smokerNever smoked1.2 (1.1-1.3)1.0 (0.95-1.1)1.0Uhlig, Hagen and Kvien3615851Current smoker (male)Current smoker (female)Never smoked2.4 (1.5-3.9)1.1 (0.8-1.6)1.0 Sources: PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IdXRjaGluc29uPC9BdXRob3I+PFllYXI+MjAwMTwvWWVh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==
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IdXRjaGluc29uPC9BdXRob3I+PFllYXI+MjAwMTwvWWVh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==
ADDIN EN.CITE.DATA (Hutchinson et al., 2001), PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TeW1tb25zPC9BdXRob3I+PFllYXI+MTk5NzwvWWVhcj48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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5TeW1tb25zPC9BdXRob3I+PFllYXI+MTk5NzwvWWVhcj48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ADDIN EN.CITE.DATA (Symmons et al., 1997), ADDIN EN.CITE <EndNote><Cite><Author>Karlson</Author><Year>1999</Year><RecNum>102</RecNum><DisplayText>(Karlson et al., 1999)</DisplayText><record><rec-number>102</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">102</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Karlson, E.W.</author><author>Lee, I.M.</author><author>Cook, N.R.</author><author>Manson, J.E.</author><author>Buring, J.E.</author><author>Hennekens, C.H.</author></authors></contributors><auth-address>Harvard Medical School, and Brigham and Women's Hospital, Boston, Massachusetts, USA</auth-address><titles><title>A retrospective cohort study of cigarette smoking and risk of rheumatoid arthritis in female health professionals</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>910-917</pages><volume>42</volume><number>5</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Estrogens)</keyword><keyword>ARTHRITIS-RHEUMATOID/BL (blood),*EP (epidemiology)</keyword><keyword>COHORT-STUDIES</keyword><keyword>diagnosis</keyword><keyword>DOUBLE-BLIND-METHOD</keyword><keyword>ESTROGENS/PH (physiology)</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MENARCHE</keyword><keyword>methods</keyword><keyword>QUESTIONNAIRES</keyword><keyword>RA</keyword><keyword>RETROSPECTIVE-STUDIES</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword><keyword>SMOKING/*AE (adverse effects),BL (blood)</keyword><keyword>United States</keyword><keyword>UNITED-STATES</keyword></keywords><dates><year>1999</year><pub-dates><date>5/1999</date></pub-dates></dates><label>172</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Karlson et al., 1999), ADDIN EN.CITE <EndNote><Cite><Author>Uhlig</Author><Year>1999</Year><RecNum>100</RecNum><DisplayText>(Uhlig et al., 1999)</DisplayText><record><rec-number>100</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">100</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Uhlig, T.</author><author>Hagen, K.B.</author><author>Kvien, T.K.</author></authors></contributors><auth-address>Oslo City Department of Rheumatology, Diakonhjemmet Hospital, Norway</auth-address><titles><title>Current tobacco smoking, formal education, and the risk of rheumatoid arthritis</title><secondary-title>The Journal of rheumatology</secondary-title></titles><pages>47-54</pages><volume>26</volume><number>1</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGE-FACTORS</keyword><keyword>AGE-OF-ONSET</keyword><keyword>AGED</keyword><keyword>analysis</keyword><keyword>ARTHRITIS-RHEUMATOID/EP (epidemiology),*ET (etiology)</keyword><keyword>COHORT-STUDIES</keyword><keyword>CONFIDENCE-INTERVALS</keyword><keyword>Disease</keyword><keyword>EDUCATIONAL-STATUS</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>MALE</keyword><keyword>methods</keyword><keyword>MIDDLE-AGED</keyword><keyword>NORWAY</keyword><keyword>ODDS-RATIO</keyword><keyword>QUESTIONNAIRES</keyword><keyword>RA</keyword><keyword>REGRESSION-ANALYSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>Risk Factors</keyword><keyword>RISK-FACTORS</keyword><keyword>SEX-FACTORS</keyword><keyword>SMOKING/*AE (adverse effects)</keyword></keywords><dates><year>1999</year><pub-dates><date>1/1999</date></pub-dates></dates><label>170</label><urls></urls><access-date>2003</access-date></record></Cite></EndNote>(Uhlig et al., 1999).Despite these data, Wilson and Goldsmith ADDIN EN.CITE <EndNote><Cite><Author>Wilson</Author><Year>1999</Year><RecNum>93</RecNum><DisplayText>(Wilson & Goldsmith, 1999)</DisplayText><record><rec-number>93</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">93</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Wilson, K.</author><author>Goldsmith, C.H.</author></authors></contributors><titles><title>Does smoking cause rheumatoid arthritis?</title><secondary-title>The Journal of rheumatology</secondary-title></titles><pages>1-3</pages><volume>26</volume><number>1</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ARTHRITIS-RHEUMATOID/*ET (etiology)</keyword><keyword>BIAS-EPIDEMIOLOGY</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>RISK-FACTORS</keyword><keyword>SMOKING/*AE (adverse effects)</keyword></keywords><dates><year>1999</year><pub-dates><date>1/1999</date></pub-dates></dates><label>163</label><urls></urls><access-date>2003</access-date></record></Cite></EndNote>(Wilson & Goldsmith, 1999) found that the association between other conditions such as lung cancer and heart disease have been much more convincing than that for smoking and RA. In RA they found the association only apparent in men.Socioeconomic factors tend to impact the course and outcome of RA rather than the risk of developing the disease. Factors such as education, occupation and marital status appear to be predictors for severity and outcome. Studies are conflicting on the question of how socioeconomic factors influence the onset of RA ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005).Epidemiological studies suggest a protective effect of a regular diet of fish, olive oil and cooked vegetables. A study of RA patients by Skoldstam, ADDIN EN.CITE <EndNote><Cite><Author>Skoldstam</Author><Year>2003</Year><RecNum>103</RecNum><DisplayText>(Skoldstam et al., 2003)</DisplayText><record><rec-number>103</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">103</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Skoldstam, L.</author><author>Hagfors, L.</author><author>Johansson, G.</author></authors></contributors><auth-address>Department of Medicine, Kalmar County Hospital, S-391 85 Kalmar, Sweden</auth-address><titles><title>An experimental study of a Mediterranean diet intervention for patients with rheumatoid arthritis</title><secondary-title>Annals of the rheumatic diseases</secondary-title></titles><periodical><full-title>Annals of the Rheumatic Diseases</full-title></periodical><pages>208-214</pages><volume>62</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Anti-Inflammatory-Agents-Non-Steroidal)</keyword><keyword>0 (Lipids)</keyword><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>ANTI-INFLAMMATORY-AGENTS-NON-STEROIDAL/TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/BL (blood),*DH (diet therapy),DT (drug therapy)</keyword><keyword>BODY-WEIGHT</keyword><keyword>DIET-MEDITERRANEAN/*</keyword><keyword>Disease</keyword><keyword>epidemiology</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>LIPIDS/BL (blood)</keyword><keyword>MALE</keyword><keyword>methods</keyword><keyword>MIDDLE-AGED</keyword><keyword>QUALITY-OF-LIFE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Sweden</keyword></keywords><dates><year>2003</year><pub-dates><date>3/2003</date></pub-dates></dates><label>173</label><urls></urls><access-date>2002</access-date></record></Cite></EndNote>(Skoldstam et al., 2003) compared a Mediterranean diet to a western diet and concluded that a Mediterranean diet did tend to reduce inflammatory activity. Fish oils also appear protective against a severe course of the disease ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005). Differences in the occurrence of RA among different populations have been reported by several studies. As seen in REF _Ref490046042 \h \* MERGEFORMAT Table 2.1 the prevalence of RA in European and North American populations are relatively constant. The geographic variation of the disease occurrence and increased incidence in certain ethnic groups, suggest an association of ethnicity with RA ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005). Notably the Chippewa and Pima Indians in North America have the highest known prevalence of RA at 5.3% and 6.8% respectively. Populations with very low prevalence include African populations in South Africa and Nigeria where no cases were found in studies of 500 and 2000 adults, respectively ADDIN EN.CITE <EndNote><Cite><Author>Silman</Author><Year>2002</Year><RecNum>83</RecNum><DisplayText>(Silman & Pearson, 2002)</DisplayText><record><rec-number>83</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">83</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Silman, A.J.</author><author>Pearson, J.E.</author></authors></contributors><auth-address>ARC Epidemiology Unit, School of Epidemiology & Health Sciences, University of Manchester, UK. alan.silman@man.ac.uk</auth-address><titles><title>Epidemiology and genetics of rheumatoid arthritis</title><secondary-title>Arthritis Res</secondary-title></titles><pages>S265-S272</pages><volume>4 Suppl 3</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>Environment</keyword><keyword>epidemiology</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>Genetic Predisposition to Disease</keyword><keyword>genetics</keyword><keyword>HUMANS</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2002</year><pub-dates><date>2002</date></pub-dates></dates><isbn>1465-9905</isbn><label>153</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(Silman & Pearson, 2002). It is possible that the low rates may be due to low reporting rates. Summary In this chapter we have described RA as a disease, its postulated causes and potential biomarkers for occurrence. RA is a chronic, destructive, inflammatory, autoimmune disease that has both articular and systemic manifestations that affects approximately 1% of the general adult population. The causes of RA are unknown although there are many risk factors believed to affect occurrence and severity, these risk factors include age, gender, smoking, genetics, ethnicity, diet, and socioeconomic status. RA is recognised as a multi-faceted disease and as may be expected, a multitude of measures are used in clinical practice and clinical trials to assess disease activity and progression. Section 2.6 described how RA is diagnosed using criteria developed in 1987. A patient is said to have RA if he or she satisfied at least four of 7 criteria, with the first four present for at least six weeks.In clinical trials, regulatory submissions and real-life settings, there are great advantages in being able to predict patient outcome from early data. In drug development, trials can be made more efficient through targeted recruitment and improved stopping rules. In the clinic, knowing a patient’s likely disease outcome, can lead to a more personalised healthcare strategy. This thesis will look at how well patient outcome can be predicted from baseline biomarkers and patient characteristics. As has been described in this chapter, RA is a complex disease that is not attuned to simple characterisation in diagnosis, treatment or measurement of severity of disease or response. In the next chapter we will see what measures are used in practice to assess disease activity and to measure response. This will identify the variable we will aim to predict.: Rheumatoid Arthritis Clinical Trial Design and Endpoints3.1.IntroductionChapter 2 described RA as a disease, including risk factors and biomarkers believed to affect incidence, severity, outcome, and epidemiology of RA, as well as comorbidities and pathophysiology and how RA is diagnosed in clinical practice. The objective of this research is to look at how well patient outcome can be predicted from baseline biomarkers and patient characteristics, and this chapter will explore the key measures of disease activity and response in a clinical trial setting. From this we will identify and justify an endpoint for prediction. This chapter will describe phase III clinical trial designs and strategies for drug registration and approval, thus setting the context for the clinical trial database available from Roche for this research. In the context of this thesis, some understanding of clinical trial design in RA may inform improvements that could be made following the results obtained from this research.3.2.AimsThe objectives of this chapter are: To discuss the various measures used in clinical practice and clinical trials to assess disease activity;To describe typical phase III clinical trial designs for RA;To describe the regulatory pathways for registration and approval of new treatments for RA in Europe and USA, and current treatment strategies in the clinic.3.3.Assessing disease activityAs previous discussed, RA is a systemic multifactorial disease with considerable variation of disease expression. Symptoms are many and include joint pain, fatigue, restrictions in activities of daily living, and increasing disability. Section REF _Ref461274873 \r \h \* MERGEFORMAT 2.6 described how diagnosis of RA has changed over time, reflecting changing clinician views on definition of the disease and the complexity of its expression. As may be expected, several methods are used to measure RA disease activity and associated impairment of function and quality of life. A number of these are detailed below.3.3.1.American College of Rheumatology (ACR) criteria Originally published in 1995, these criteria are widely used by clinicians, drug makers and regulators to measure response to therapy ADDIN EN.CITE <EndNote><Cite><Author>Felson</Author><Year>1995</Year><RecNum>65</RecNum><DisplayText>(Felson et al., 1995)</DisplayText><record><rec-number>65</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">65</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Felson, D.T.</author><author>Anderson, J.J.</author><author>Boers, M.</author><author>Bombardier, C.</author><author>Furst, D.</author><author>Goldsmith, C.</author><author>Katz, L.M.</author><author>Lightfoot, R., Jr.</author><author>Paulus, H.</author><author>Strand, V.</author></authors></contributors><auth-address>Boston University Arthritis Center, Massachusetts, USA</auth-address><titles><title>American College of Rheumatology Preliminary definition of improvement in rheumatoid arthritis</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>727-735</pages><volume>38</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Anti-Inflammatory-Agents)</keyword><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>ANTI-INFLAMMATORY-AGENTS/*ST (standards),TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/*DT (drug therapy),PP (physiopathology)</keyword><keyword>CLINICAL-TRIALS-AS-TOPIC</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>JOINTS/PP (physiopathology)</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>OUTCOME-ASSESSMENT-HEALTH-CARE/*</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX</keyword><keyword>SOCIETIES-MEDICAL</keyword></keywords><dates><year>1995</year><pub-dates><date>6/1995</date></pub-dates></dates><label>66</label><urls></urls><access-date>2007</access-date></record></Cite></EndNote>(Felson et al., 1995). The ACR criteria, measure an improvement to treatment as opposed to a measurement of disease state. The ACR committee led by Felson pointed out that in 1992, of the 15 trials published for RA treatment at that time, only 6 used improvement criteria and each of these used different criteria. In order to develop a de facto improvement measure, the committee surveyed the rheumatology community for which improvement criteria were used in practice. The committee asked rheumatologists to review fictional patient data. The focus then turned to those fictional patients on which 80% of the physicians agreed had improved. This reduced the potential criteria to 20 variations. Clinical trial data was then used to identify criteria which showed the clearest differentiation of treatment difference between a control and an effective treatment. The final 8 definitions were then reviewed for ease of use and accordance with rheumatologists’ impressions of improvement and then combined into a composite score ADDIN EN.CITE <EndNote><Cite><Author>Felson</Author><Year>1995</Year><RecNum>65</RecNum><DisplayText>(Felson et al., 1995)</DisplayText><record><rec-number>65</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">65</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Felson, D.T.</author><author>Anderson, J.J.</author><author>Boers, M.</author><author>Bombardier, C.</author><author>Furst, D.</author><author>Goldsmith, C.</author><author>Katz, L.M.</author><author>Lightfoot, R., Jr.</author><author>Paulus, H.</author><author>Strand, V.</author></authors></contributors><auth-address>Boston University Arthritis Center, Massachusetts, USA</auth-address><titles><title>American College of Rheumatology Preliminary definition of improvement in rheumatoid arthritis</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>727-735</pages><volume>38</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Anti-Inflammatory-Agents)</keyword><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>ANTI-INFLAMMATORY-AGENTS/*ST (standards),TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/*DT (drug therapy),PP (physiopathology)</keyword><keyword>CLINICAL-TRIALS-AS-TOPIC</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>JOINTS/PP (physiopathology)</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>OUTCOME-ASSESSMENT-HEALTH-CARE/*</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX</keyword><keyword>SOCIETIES-MEDICAL</keyword></keywords><dates><year>1995</year><pub-dates><date>6/1995</date></pub-dates></dates><label>66</label><urls></urls><access-date>2007</access-date></record></Cite></EndNote>(Felson et al., 1995).A 20% improvement in these criteria, a result that is known as ACR20, requires a 20% reduction in: the number of swollen joints on a 66-joint count, the number of tender joints on a 68-joint count and in three of five other criteria ADDIN EN.CITE <EndNote><Cite><Author>Felson</Author><Year>1995</Year><RecNum>65</RecNum><DisplayText>(Felson et al., 1995)</DisplayText><record><rec-number>65</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">65</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Felson, D.T.</author><author>Anderson, J.J.</author><author>Boers, M.</author><author>Bombardier, C.</author><author>Furst, D.</author><author>Goldsmith, C.</author><author>Katz, L.M.</author><author>Lightfoot, R., Jr.</author><author>Paulus, H.</author><author>Strand, V.</author></authors></contributors><auth-address>Boston University Arthritis Center, Massachusetts, USA</auth-address><titles><title>American College of Rheumatology Preliminary definition of improvement in rheumatoid arthritis</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>727-735</pages><volume>38</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Anti-Inflammatory-Agents)</keyword><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>ANTI-INFLAMMATORY-AGENTS/*ST (standards),TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/*DT (drug therapy),PP (physiopathology)</keyword><keyword>CLINICAL-TRIALS-AS-TOPIC</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>JOINTS/PP (physiopathology)</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>OUTCOME-ASSESSMENT-HEALTH-CARE/*</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX</keyword><keyword>SOCIETIES-MEDICAL</keyword></keywords><dates><year>1995</year><pub-dates><date>6/1995</date></pub-dates></dates><label>66</label><urls></urls><access-date>2007</access-date></record></Cite></EndNote>(Felson et al., 1995) ( REF _Ref490046510 \h \* MERGEFORMAT Table 3.1). The extra 2 joints in tender count being the hip joints, where swelling is considered difficult to assess ADDIN EN.CITE <EndNote><Cite><Author>Sokka</Author><Year>2005</Year><RecNum>402</RecNum><DisplayText>(Sokka & Pincus, 2005)</DisplayText><record><rec-number>402</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1524052849">402</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Sokka, T.</author><author>Pincus, T.</author></authors></contributors><titles><title>Quantitative joint assessment in rheumatoid arthritis</title><secondary-title>Clinical and Experimental Rheumatology</secondary-title></titles><periodical><full-title>Clinical and Experimental Rheumatology</full-title></periodical><pages>S58-S62</pages><volume>23</volume><number>S39</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(Sokka & Pincus, 2005). See REF _Ref490046523 \h \* MERGEFORMAT Figure 3.1 for a diagram of joints assessed for swelling and tenderness.Table STYLEREF 1 \s 3. SEQ Table \* ARABIC \s 1 1 American College of Rheumatology definition of ACR20 measure≥20% improvement in tender 68-joint countAND≥20% improvement in swollen 66-joint countAND≥20% improvement in three of the following measures? Patient’s assessment of pain? Patient’s global assessment of disease activity? Physician’s global assessment of disease activity? Patient’s assessment of physical function? Markers of inflammation (e.g. ESR or CRP) ADDIN EN.CITE <EndNote><Cite><Author>Felson</Author><Year>1995</Year><RecNum>65</RecNum><DisplayText>(Felson et al., 1995)</DisplayText><record><rec-number>65</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">65</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Felson, D.T.</author><author>Anderson, J.J.</author><author>Boers, M.</author><author>Bombardier, C.</author><author>Furst, D.</author><author>Goldsmith, C.</author><author>Katz, L.M.</author><author>Lightfoot, R., Jr.</author><author>Paulus, H.</author><author>Strand, V.</author></authors></contributors><auth-address>Boston University Arthritis Center, Massachusetts, USA</auth-address><titles><title>American College of Rheumatology Preliminary definition of improvement in rheumatoid arthritis</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>727-735</pages><volume>38</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Anti-Inflammatory-Agents)</keyword><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>ANTI-INFLAMMATORY-AGENTS/*ST (standards),TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/*DT (drug therapy),PP (physiopathology)</keyword><keyword>CLINICAL-TRIALS-AS-TOPIC</keyword><keyword>FEMALE</keyword><keyword>First</keyword><keyword>HUMANS</keyword><keyword>JOINTS/PP (physiopathology)</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>OUTCOME-ASSESSMENT-HEALTH-CARE/*</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX</keyword><keyword>SOCIETIES-MEDICAL</keyword></keywords><dates><year>1995</year><pub-dates><date>6/1995</date></pub-dates></dates><label>66</label><urls></urls><access-date>2007</access-date></record></Cite></EndNote>(Felson et al., 1995)Figure STYLEREF 1 \s 3. SEQ Figure \* ARABIC \s 1 1 Joints measured for swelling and tendernessFigure adapted from ( By LadyofHats Mariana Ruiz Villarreal [Public domain]. Wikipedia Commons is a licence free media repository.) Last accessed 31/5/2017In order to assess greater responses to therapy, ACR50 and ACR70 are often calculated using the same methodology as ACR20 except they require 50% and 70% reductions, respectively ADDIN EN.CITE <EndNote><Cite><Author>Rindfleisch</Author><Year>2005</Year><RecNum>81</RecNum><DisplayText>(Rindfleisch & Muller, 2005)</DisplayText><record><rec-number>81</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">81</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rindfleisch, J.A.</author><author>Muller, D.</author></authors></contributors><auth-address>Department of Family Medicine, University of Wisconsin-Madison, Madison, Wisconsin, USA</auth-address><titles><title>Diagnosis and management of rheumatoid arthritis</title><secondary-title>Am. Fam. Physician</secondary-title></titles><pages>1037-1047</pages><volume>72</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>administration & dosage</keyword><keyword>ADULT</keyword><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>COMORBIDITY</keyword><keyword>diagnosis</keyword><keyword>Diagnosis,Differential</keyword><keyword>drug therapy</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>Middle Aged</keyword><keyword>physiopathology</keyword><keyword>PREVALENCE</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Risk Factors</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2005</year><pub-dates><date>9/15/2005</date></pub-dates></dates><isbn>0002-838X</isbn><label>83</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(Rindfleisch & Muller, 2005). At a patient level, these ACR thresholds define a patient as a responder, each measure then is a dichotomous outcome: responder or non-responder. In addition, a further variable ACRn can be derived. ACRn is defined as each patient’s lowest percent improvement from baseline in 3 measures: Tender joint count (TJC) (68 joints), Swollen joint count (SJC) (66 joints), and the improved score achieved in at least 3 of the 5 remaining ACR core set parameters. So that a patient with an ACRn value of 20 is equivalent to ACR20 response.3.3.2.Disease Activity Score (DAS28)In contrast to the ACR criteria, which measure an improvement from a baseline in RA symptoms, the Disease Activity Score 28 (DAS28) measures the level of disease state PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IZWlqZGU8L0F1dGhvcj48WWVhcj4xOTkwPC9ZZWFyPjxS
ZWNOdW0+OTY8L1JlY051bT48RGlzcGxheVRleHQ+KHZhbiBkZXIgSGVpamRlIGV0IGFsLiwgMTk5
MCk8L0Rpc3BsYXlUZXh0PjxyZWNvcmQ+PHJlYy1udW1iZXI+OTY8L3JlYy1udW1iZXI+PGZvcmVp
Z24ta2V5cz48a2V5IGFwcD0iRU4iIGRiLWlkPSI1YTk5d2RhYXp3ZHh0M2V4NXBmcHBhZTNmc3Bz
d3Z0ZnJ4cmQiIHRpbWVzdGFtcD0iMCI+OTY8L2tleT48L2ZvcmVpZ24ta2V5cz48cmVmLXR5cGUg
bmFtZT0iSm91cm5hbCBBcnRpY2xlIj4xNzwvcmVmLXR5cGU+PGNvbnRyaWJ1dG9ycz48YXV0aG9y
cz48YXV0aG9yPnZhbiBkZXIgSGVpamRlLCBELjwvYXV0aG9yPjxhdXRob3I+SG9mLCBNLkEuPC9h
dXRob3I+PGF1dGhvcj52YW4gUmllbC4gUC48L2F1dGhvcj48YXV0aG9yPlRoZXVuaXNzZSwgTC5B
LjwvYXV0aG9yPjxhdXRob3I+THViYmVydHMsIEUuVy48L2F1dGhvcj48YXV0aG9yPnZhbiBMZWV1
d2VuLCBNLkEuPC9hdXRob3I+PGF1dGhvcj52YW4gUmlqc3dpamssIE0uPC9hdXRob3I+PGF1dGhv
cj52YW4gZGUgUHV0dGUsIEwuPC9hdXRob3I+PC9hdXRob3JzPjwvY29udHJpYnV0b3JzPjxhdXRo
LWFkZHJlc3M+RGVwYXJ0bWVudCBvZiBSaGV1bWF0b2xvZ3ksIFVuaXZlcnNpdHkgSG9zcGl0YWws
IE5pam1lZ2VuLCBUaGUgTmV0aGVybGFuZHM8L2F1dGgtYWRkcmVzcz48dGl0bGVzPjx0aXRsZT5K
dWRnaW5nIGRpc2Vhc2UgYWN0aXZpdHkgaW4gY2xpbmljYWwgcHJhY3RpY2UgaW4gcmhldW1hdG9p
ZCBhcnRocml0aXM6IGZpcnN0IHN0ZXAgaW4gdGhlIGRldmVsb3BtZW50IG9mIGEgZGlzZWFzZSBh
Y3Rpdml0eSBzY29yZTwvdGl0bGU+PHNlY29uZGFyeS10aXRsZT5Bbm5hbHMgb2YgdGhlIHJoZXVt
YXRpYyBkaXNlYXNlczwvc2Vjb25kYXJ5LXRpdGxlPjwvdGl0bGVzPjxwZXJpb2RpY2FsPjxmdWxs
LXRpdGxlPkFubmFscyBvZiB0aGUgUmhldW1hdGljIERpc2Vhc2VzPC9mdWxsLXRpdGxlPjwvcGVy
aW9kaWNhbD48cGFnZXM+OTE2LTkyMDwvcGFnZXM+PHZvbHVtZT40OTwvdm9sdW1lPjxudW1iZXI+
MTE8L251bWJlcj48cmVwcmludC1lZGl0aW9uPk5vdCBpbiBGaWxlPC9yZXByaW50LWVkaXRpb24+
PGtleXdvcmRzPjxrZXl3b3JkPjAgKEFudGktSW5mbGFtbWF0b3J5LUFnZW50cyk8L2tleXdvcmQ+
PGtleXdvcmQ+QURPTEVTQ0VOVDwva2V5d29yZD48a2V5d29yZD5BRFVMVDwva2V5d29yZD48a2V5
d29yZD5BR0VEPC9rZXl3b3JkPjxrZXl3b3JkPmFuYWx5c2lzPC9rZXl3b3JkPjxrZXl3b3JkPkFO
VEktSU5GTEFNTUFUT1JZLUFHRU5UUy9UVSAodGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5
d29yZD5BUlRIUklUSVMtUkhFVU1BVE9JRC9CTCAoYmxvb2QpLERUIChkcnVnIHRoZXJhcHkpLCpQ
QSAocGF0aG9sb2d5KTwva2V5d29yZD48a2V5d29yZD5ESVNDUklNSU5BTlQtQU5BTFlTSVM8L2tl
eXdvcmQ+PGtleXdvcmQ+RGlzZWFzZTwva2V5d29yZD48a2V5d29yZD5GRU1BTEU8L2tleXdvcmQ+
PGtleXdvcmQ+Rmlyc3Q8L2tleXdvcmQ+PGtleXdvcmQ+SFVNQU5TPC9rZXl3b3JkPjxrZXl3b3Jk
PkpvaW50czwva2V5d29yZD48a2V5d29yZD5NQUxFPC9rZXl3b3JkPjxrZXl3b3JkPk1JRERMRS1B
R0VEPC9rZXl3b3JkPjxrZXl3b3JkPk5ldGhlcmxhbmRzPC9rZXl3b3JkPjxrZXl3b3JkPlByb3Nw
ZWN0aXZlIFN0dWRpZXM8L2tleXdvcmQ+PGtleXdvcmQ+UFJPU1BFQ1RJVkUtU1RVRElFUzwva2V5
d29yZD48a2V5d29yZD5SRUdSRVNTSU9OLUFOQUxZU0lTPC9rZXl3b3JkPjxrZXl3b3JkPlJFTUlT
U0lPTi1JTkRVQ1RJT048L2tleXdvcmQ+PGtleXdvcmQ+UmhldW1hdG9pZCBhcnRocml0aXM8L2tl
eXdvcmQ+PGtleXdvcmQ+UmhldW1hdG9sb2d5PC9rZXl3b3JkPjxrZXl3b3JkPlNUQVRJU1RJQ1Mt
QVMtVE9QSUM8L2tleXdvcmQ+PGtleXdvcmQ+VElNRS1GQUNUT1JTPC9rZXl3b3JkPjwva2V5d29y
ZHM+PGRhdGVzPjx5ZWFyPjE5OTA8L3llYXI+PHB1Yi1kYXRlcz48ZGF0ZT4xMS8xOTkwPC9kYXRl
PjwvcHViLWRhdGVzPjwvZGF0ZXM+PGxhYmVsPjE2NjwvbGFiZWw+PHVybHM+PC91cmxzPjxhY2Nl
c3MtZGF0ZT4yMDAxPC9hY2Nlc3MtZGF0ZT48L3JlY29yZD48L0NpdGU+PC9FbmROb3RlPgB=
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IZWlqZGU8L0F1dGhvcj48WWVhcj4xOTkwPC9ZZWFyPjxS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=
ADDIN EN.CITE.DATA (van der Heijde et al., 1990). The DAS28 was developed and validated by a group of Dutch investigators in 1990 from 113 RA patients PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IZWlqZGU8L0F1dGhvcj48WWVhcj4xOTkwPC9ZZWFyPjxS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=
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IZWlqZGU8L0F1dGhvcj48WWVhcj4xOTkwPC9ZZWFyPjxS
ZWNOdW0+OTY8L1JlY051bT48RGlzcGxheVRleHQ+KHZhbiBkZXIgSGVpamRlIGV0IGFsLiwgMTk5
MCk8L0Rpc3BsYXlUZXh0PjxyZWNvcmQ+PHJlYy1udW1iZXI+OTY8L3JlYy1udW1iZXI+PGZvcmVp
Z24ta2V5cz48a2V5IGFwcD0iRU4iIGRiLWlkPSI1YTk5d2RhYXp3ZHh0M2V4NXBmcHBhZTNmc3Bz
d3Z0ZnJ4cmQiIHRpbWVzdGFtcD0iMCI+OTY8L2tleT48L2ZvcmVpZ24ta2V5cz48cmVmLXR5cGUg
bmFtZT0iSm91cm5hbCBBcnRpY2xlIj4xNzwvcmVmLXR5cGU+PGNvbnRyaWJ1dG9ycz48YXV0aG9y
cz48YXV0aG9yPnZhbiBkZXIgSGVpamRlLCBELjwvYXV0aG9yPjxhdXRob3I+SG9mLCBNLkEuPC9h
dXRob3I+PGF1dGhvcj52YW4gUmllbC4gUC48L2F1dGhvcj48YXV0aG9yPlRoZXVuaXNzZSwgTC5B
LjwvYXV0aG9yPjxhdXRob3I+THViYmVydHMsIEUuVy48L2F1dGhvcj48YXV0aG9yPnZhbiBMZWV1
d2VuLCBNLkEuPC9hdXRob3I+PGF1dGhvcj52YW4gUmlqc3dpamssIE0uPC9hdXRob3I+PGF1dGhv
cj52YW4gZGUgUHV0dGUsIEwuPC9hdXRob3I+PC9hdXRob3JzPjwvY29udHJpYnV0b3JzPjxhdXRo
LWFkZHJlc3M+RGVwYXJ0bWVudCBvZiBSaGV1bWF0b2xvZ3ksIFVuaXZlcnNpdHkgSG9zcGl0YWws
IE5pam1lZ2VuLCBUaGUgTmV0aGVybGFuZHM8L2F1dGgtYWRkcmVzcz48dGl0bGVzPjx0aXRsZT5K
dWRnaW5nIGRpc2Vhc2UgYWN0aXZpdHkgaW4gY2xpbmljYWwgcHJhY3RpY2UgaW4gcmhldW1hdG9p
ZCBhcnRocml0aXM6IGZpcnN0IHN0ZXAgaW4gdGhlIGRldmVsb3BtZW50IG9mIGEgZGlzZWFzZSBh
Y3Rpdml0eSBzY29yZTwvdGl0bGU+PHNlY29uZGFyeS10aXRsZT5Bbm5hbHMgb2YgdGhlIHJoZXVt
YXRpYyBkaXNlYXNlczwvc2Vjb25kYXJ5LXRpdGxlPjwvdGl0bGVzPjxwZXJpb2RpY2FsPjxmdWxs
LXRpdGxlPkFubmFscyBvZiB0aGUgUmhldW1hdGljIERpc2Vhc2VzPC9mdWxsLXRpdGxlPjwvcGVy
aW9kaWNhbD48cGFnZXM+OTE2LTkyMDwvcGFnZXM+PHZvbHVtZT40OTwvdm9sdW1lPjxudW1iZXI+
MTE8L251bWJlcj48cmVwcmludC1lZGl0aW9uPk5vdCBpbiBGaWxlPC9yZXByaW50LWVkaXRpb24+
PGtleXdvcmRzPjxrZXl3b3JkPjAgKEFudGktSW5mbGFtbWF0b3J5LUFnZW50cyk8L2tleXdvcmQ+
PGtleXdvcmQ+QURPTEVTQ0VOVDwva2V5d29yZD48a2V5d29yZD5BRFVMVDwva2V5d29yZD48a2V5
d29yZD5BR0VEPC9rZXl3b3JkPjxrZXl3b3JkPmFuYWx5c2lzPC9rZXl3b3JkPjxrZXl3b3JkPkFO
VEktSU5GTEFNTUFUT1JZLUFHRU5UUy9UVSAodGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5
d29yZD5BUlRIUklUSVMtUkhFVU1BVE9JRC9CTCAoYmxvb2QpLERUIChkcnVnIHRoZXJhcHkpLCpQ
QSAocGF0aG9sb2d5KTwva2V5d29yZD48a2V5d29yZD5ESVNDUklNSU5BTlQtQU5BTFlTSVM8L2tl
eXdvcmQ+PGtleXdvcmQ+RGlzZWFzZTwva2V5d29yZD48a2V5d29yZD5GRU1BTEU8L2tleXdvcmQ+
PGtleXdvcmQ+Rmlyc3Q8L2tleXdvcmQ+PGtleXdvcmQ+SFVNQU5TPC9rZXl3b3JkPjxrZXl3b3Jk
PkpvaW50czwva2V5d29yZD48a2V5d29yZD5NQUxFPC9rZXl3b3JkPjxrZXl3b3JkPk1JRERMRS1B
R0VEPC9rZXl3b3JkPjxrZXl3b3JkPk5ldGhlcmxhbmRzPC9rZXl3b3JkPjxrZXl3b3JkPlByb3Nw
ZWN0aXZlIFN0dWRpZXM8L2tleXdvcmQ+PGtleXdvcmQ+UFJPU1BFQ1RJVkUtU1RVRElFUzwva2V5
d29yZD48a2V5d29yZD5SRUdSRVNTSU9OLUFOQUxZU0lTPC9rZXl3b3JkPjxrZXl3b3JkPlJFTUlT
U0lPTi1JTkRVQ1RJT048L2tleXdvcmQ+PGtleXdvcmQ+UmhldW1hdG9pZCBhcnRocml0aXM8L2tl
eXdvcmQ+PGtleXdvcmQ+UmhldW1hdG9sb2d5PC9rZXl3b3JkPjxrZXl3b3JkPlNUQVRJU1RJQ1Mt
QVMtVE9QSUM8L2tleXdvcmQ+PGtleXdvcmQ+VElNRS1GQUNUT1JTPC9rZXl3b3JkPjwva2V5d29y
ZHM+PGRhdGVzPjx5ZWFyPjE5OTA8L3llYXI+PHB1Yi1kYXRlcz48ZGF0ZT4xMS8xOTkwPC9kYXRl
PjwvcHViLWRhdGVzPjwvZGF0ZXM+PGxhYmVsPjE2NjwvbGFiZWw+PHVybHM+PC91cmxzPjxhY2Nl
c3MtZGF0ZT4yMDAxPC9hY2Nlc3MtZGF0ZT48L3JlY29yZD48L0NpdGU+PC9FbmROb3RlPgB=
ADDIN EN.CITE.DATA (van der Heijde et al., 1990). These patients were divided into low and high disease activity groups. Factor analysis was applied using a high number of clinical and laboratory values leading to the selection of 5 important factors. Discriminant analysis was then performed to ascertain to what extent each factor contributed to discrimination between high and low disease activity.The resulting index, using a number of measures, including a 28 tender/swollen joint count, ESR and general health status, is calculated as follows:DAS28 = 0.56 x sqrt (TJC28) + 0.28 x sqrt (SJC28) + 0.70 x ln (ESR) +0.014 x GHAbbreviations: ESR, erythrocyte sedimentation rate;GH, global health (visual analogue scale 0 to 10); SJC28, swollen joint count based on a count of 28 joints; TJC28, tender joint count based on a count of 28 joints.Ln, Natural LogIn determining DAS28 joint counts and ESR are considered more important. In clinical practice a 3 item DAS28 can be calculated where general health assessment is omitted ADDIN EN.CITE <EndNote><Cite><Author>Fransen</Author><Year>2009</Year><RecNum>115</RecNum><DisplayText>(Fransen & van Riel, 2009)</DisplayText><record><rec-number>115</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">115</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Fransen, J.</author><author>van Riel, P.</author></authors></contributors><auth-address>Department of Rheumatology, Radboud University Nijmegen Medical Centre, PO Box 9101, NL-6500HB Nijmegen, The Netherlands. J.Fransen @reuma.umcn.nl</auth-address><titles><title>The Disease Activity Score and the EULAR response criteria</title><secondary-title>Rheumatic diseases clinics of North America</secondary-title></titles><pages>745-757</pages><volume>35</volume><number>4</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ARTHRITIS-RHEUMATOID/*DI (diagnosis),*PP (physiopathology)</keyword><keyword>ARTICLE</keyword><keyword>CLINICAL-TRIAL</keyword><keyword>DISABILITY-EVALUATION/*</keyword><keyword>Disease</keyword><keyword>DISEASE-ACTIVITY</keyword><keyword>HEALTH-STATUS/*</keyword><keyword>HUMANS</keyword><keyword>Joints</keyword><keyword>Netherlands</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX/*</keyword></keywords><dates><year>2009</year><pub-dates><date>11/2009</date></pub-dates></dates><label>185</label><urls></urls><access-date>2012</access-date></record></Cite></EndNote>(Fransen & van Riel, 2009).The score can range on a continuous spectrum from 0 to 10, with a score >5.1 indicating severe disease. A DAS28 of <2.6 is commonly used to define remission, and is associated with a lower level of disease activity (fewer active joints, lower ESR) than an ACR70 response PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IZWlqZGU8L0F1dGhvcj48WWVhcj4xOTkwPC9ZZWFyPjxS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=
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5IZWlqZGU8L0F1dGhvcj48WWVhcj4xOTkwPC9ZZWFyPjxS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=
ADDIN EN.CITE.DATA (van der Heijde et al., 1990).In the context of a clinical trial the DAS28 and change from baseline in DAS28 endpoints are usually treated as Normally distributed continuous variables ADDIN EN.CITE <EndNote><Cite><Author>Fransen</Author><Year>2009</Year><RecNum>115</RecNum><DisplayText>(Fransen & van Riel, 2009)</DisplayText><record><rec-number>115</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">115</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Fransen, J.</author><author>van Riel, P.</author></authors></contributors><auth-address>Department of Rheumatology, Radboud University Nijmegen Medical Centre, PO Box 9101, NL-6500HB Nijmegen, The Netherlands. J.Fransen @reuma.umcn.nl</auth-address><titles><title>The Disease Activity Score and the EULAR response criteria</title><secondary-title>Rheumatic diseases clinics of North America</secondary-title></titles><pages>745-757</pages><volume>35</volume><number>4</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ARTHRITIS-RHEUMATOID/*DI (diagnosis),*PP (physiopathology)</keyword><keyword>ARTICLE</keyword><keyword>CLINICAL-TRIAL</keyword><keyword>DISABILITY-EVALUATION/*</keyword><keyword>Disease</keyword><keyword>DISEASE-ACTIVITY</keyword><keyword>HEALTH-STATUS/*</keyword><keyword>HUMANS</keyword><keyword>Joints</keyword><keyword>Netherlands</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatology</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX/*</keyword></keywords><dates><year>2009</year><pub-dates><date>11/2009</date></pub-dates></dates><label>185</label><urls></urls><access-date>2012</access-date></record></Cite></EndNote>(Fransen & van Riel, 2009). 3.3.3.European League Against Rheumatism (EULAR) response criteriaIn an effort to convert DAS28 back into an improvement score this measure combines the DAS28 at the time of evaluation with the change in DAS28 between two time points. Response categories include good, moderate and no response ( REF _Ref490046608 \h \* MERGEFORMAT Table 3.2) PEVuZE5vdGU+PENpdGU+PEF1dGhvcj52YW4gR2VzdGVsPC9BdXRob3I+PFllYXI+MTk5NjwvWWVh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=
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj52YW4gR2VzdGVsPC9BdXRob3I+PFllYXI+MTk5NjwvWWVh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=
ADDIN EN.CITE.DATA (van Gestel et al., 1996).Table STYLEREF 1 \s 3. SEQ Table \* ARABIC \s 1 2 EULAR responseReduction from Baseline in DAS28 scoreDAS28>1.2 >0.6 and ≤ 1.2 ≤ 0.6 ≤ 3.2 Good Moderate No Response > 3.2 to ≤ 5.1 Moderate Moderate No Response >5.1ModerateNo ResponseNo ResponseThe EULAR response is can be analysed as a dichotomous variable by combining good and moderate categories.3.3.4.Radiographic evaluation in clinical development Radiography provides an objective measure of the extent of anatomical joint damage. A range of radiographic scores are currently used, many of which combine assessments of erosions and joint space narrowing PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5CYXJvbjwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+PFJl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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5CYXJvbjwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+PFJl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ADDIN EN.CITE.DATA (Baron et al., 2007).The European Medicines Agency (EMA) ADDIN EN.CITE <EndNote><Cite><Author>EMA</Author><Year>2003</Year><RecNum>137</RecNum><DisplayText>(EMA, 2003)</DisplayText><record><rec-number>137</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">137</key></foreign-keys><ref-type name="Government Document">46</ref-type><contributors><authors><author>EMA</author></authors></contributors><titles><title>Points to Consider on the Clinical Investigation of Medicinal Products other than Nsaids in Rheumatoid Arthritis</title></titles><dates><year>2003</year></dates><urls></urls></record></Cite></EndNote>(EMA, 2003) recommends that in order to demonstrate radiological differences of hands and feet on the basis of before/after comparisons, the radiographs should be taken not less than a year apart and ideally for two years using full randomization and pre-agreed criteria ADDIN EN.CITE <EndNote><Cite><Author>EMA</Author><Year>2003</Year><RecNum>137</RecNum><DisplayText>(EMA, 2003)</DisplayText><record><rec-number>137</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">137</key></foreign-keys><ref-type name="Government Document">46</ref-type><contributors><authors><author>EMA</author></authors></contributors><titles><title>Points to Consider on the Clinical Investigation of Medicinal Products other than Nsaids in Rheumatoid Arthritis</title></titles><dates><year>2003</year></dates><urls></urls></record></Cite></EndNote>(EMA, 2003). The FDA ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010) similarly recommends prevention of structural damage as an important goal of RA therapy. The duration of such trials are expected to be at least one-year duration. Claims that can be pursued include slowing X-ray progression, using either the Larsen, modified Sharp or another validated radiographic index, or prevention of new x-ray erosions – maintaining an erosion-free state or preventing new erosions ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010).Larsen ADDIN EN.CITE <EndNote><Cite><Author>Larsen</Author><Year>1977</Year><RecNum>89</RecNum><DisplayText>(Larsen et al., 1977)</DisplayText><record><rec-number>89</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">89</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Larsen, A.</author><author>Dale, K.</author><author>Eek, M.</author></authors></contributors><titles><title>Radiographic evaluation of rheumatoid arthritis and related conditions by standard reference films</title><secondary-title>Acta Radiol. Diagn. (Stockh)</secondary-title></titles><pages>481-491</pages><volume>18</volume><number>4</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>Evaluation Studies as Topic</keyword><keyword>HUMANS</keyword><keyword>methods</keyword><keyword>radiography</keyword><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>1977</year><pub-dates><date>7/1977</date></pub-dates></dates><isbn>0567-8056</isbn><label>159</label><urls><related-urls><url>;(Larsen et al., 1977) or Modified Sharp Scores ADDIN EN.CITE <EndNote><Cite><Author>Sharp</Author><Year>1985</Year><RecNum>106</RecNum><DisplayText>(Sharp et al., 1985)</DisplayText><record><rec-number>106</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">106</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Sharp, J.T.</author><author>Young, D.Y.</author><author>Bluhm, G.B.</author><author>Brook, A.</author><author>Brower, A.C.</author><author>Corbett, M.</author><author>Decker, J.L.</author><author>Genant, H.K.</author><author>Gofton, J.P.</author><author>Goodman, N.</author></authors></contributors><titles><title>How many joints in the hands and wrists should be included in a score of radiologic abnormalities used to assess rheumatoid arthritis?</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>1326-1335</pages><volume>28</volume><number>12</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ARTHRITIS-RHEUMATOID/PA (pathology),*RA (radiography)</keyword><keyword>ARTHROGRAPHY/*MT (methods)</keyword><keyword>Computers</keyword><keyword>FINGER-JOINT/RA (radiography)</keyword><keyword>Hand</keyword><keyword>HAND/*</keyword><keyword>HUMANS</keyword><keyword>INFORMATION-SYSTEMS</keyword><keyword>Joints</keyword><keyword>Mathematics</keyword><keyword>methods</keyword><keyword>Rheumatoid arthritis</keyword><keyword>WRIST-JOINT/*RA (radiography)</keyword></keywords><dates><year>1985</year><pub-dates><date>12/1985</date></pub-dates></dates><label>176</label><urls></urls><access-date>2002</access-date></record></Cite></EndNote>(Sharp et al., 1985) are common validated methods used to assess radiographs. In the Larsen method, the pooled information is captured in grading whereas the Sharp score captures information on erosion and joint space narrowing and combines into a total score. In practice, the Sharp Score is modified in one or two ways: Genant ADDIN EN.CITE <EndNote><Cite><Author>Genant</Author><Year>1983</Year><RecNum>105</RecNum><DisplayText>(Genant, 1983)</DisplayText><record><rec-number>105</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">105</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Genant, H.K.</author></authors></contributors><titles><title>Methods of assessing radiographic change in rheumatoid arthritis</title><secondary-title>The American journal of medicine</secondary-title></titles><pages>35-47</pages><volume>75</volume><number>6A</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>7440-57-5 (Gold)</keyword><keyword>analysis</keyword><keyword>ARTHRITIS-RHEUMATOID/DT (drug therapy),*RA (radiography)</keyword><keyword>First</keyword><keyword>FOOT/RA (radiography)</keyword><keyword>GOLD/TU (therapeutic use)</keyword><keyword>HAND/RA (radiography)</keyword><keyword>HUMANS</keyword><keyword>methods</keyword><keyword>PROGNOSIS</keyword><keyword>RADIOGRAPHIC-MAGNIFICATION/MT (methods)</keyword><keyword>radiography</keyword><keyword>Rheumatoid arthritis</keyword><keyword>standards</keyword><keyword>X-RAY-FILM</keyword></keywords><dates><year>1983</year><pub-dates><date>12/30/1983</date></pub-dates></dates><label>175</label><urls></urls><access-date>2002</access-date></record></Cite></EndNote>(Genant, 1983) or van der Heijde ADDIN EN.CITE <EndNote><Cite><Author>Heijde</Author><Year>2002</Year><RecNum>107</RecNum><DisplayText>(Heijde, 2002)</DisplayText><record><rec-number>107</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">107</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>van der Heijde</author></authors></contributors><auth-address>University Hospital Maastricht, Maastricht, The Netherlands. dhe @sint.azm.nl</auth-address><titles><title>Structural damage in rheumatoid arthritis as visualized through radiographs</title><secondary-title>Arthritis research</secondary-title></titles><pages>S29-S33</pages><volume>4</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>ARTHRITIS-RHEUMATOID/*RA (radiography)</keyword><keyword>ARTHROGRAPHY/*MT (methods),*ST (standards)</keyword><keyword>HUMANS</keyword><keyword>Joints</keyword><keyword>Netherlands</keyword><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2002</year><pub-dates><date>2002</date></pub-dates></dates><label>177</label><urls></urls><access-date>2007</access-date></record></Cite></EndNote>(Heijde, 2002). For the van der Heijde modification, in each hand there are 16 areas measured for erosions and 15 areas measured for joint space narrowing. In the foot, 6 areas are measured for erosions and 6 for joint space narrowing. The erosion score for each joint in the hand is rated on a scale from 0 to 5. Zero represents no erosion and 5 a complete collapse of the joint or the full surface of the joint is affected by erosion. Thus the maximum score for erosions in the hand is 80. Joint space narrowing is measured on a scale from 0 to 4, where normal joint spacing is 0 and 4 can represent a fusing of the joint or complete dislocation. Thus the maximum score for joint space narrowing in the hand is 60. Similar ratings are used for the joints in the feet. The total maximum scores combining both hands and feet are 280 for erosions and 168 for joint space narrowing. The total Modified Sharp Score is obtained by totalling erosion and joint space narrowing scores, leading to a maximum possible score of 448. Baron PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5CYXJvbjwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+PFJl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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5CYXJvbjwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+PFJl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ADDIN EN.CITE.DATA (Baron et al., 2007) summarised radiographic methods reported in clinical trials. In 65.2% of the 46 assessable articles more than 2 radiographic scores were used (e.g. erosions and joint space narrowing). Among the scores combining erosions and joint space narrowing the Sharp Score and Larsen Score were the most used (47.8% and 34.7% respectively).Due to the nature of joint damage, radiographic data generally has a non-Normal distribution ADDIN EN.CITE <EndNote><Cite><Author>Landewe</Author><Year>2005</Year><RecNum>116</RecNum><DisplayText>(Landewe & van der Heijde, 2005)</DisplayText><record><rec-number>116</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">116</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Landewe, R.</author><author>van der Heijde, D.</author></authors></contributors><auth-address>University Hospital Maastricht, Department of Internal Medicine /Rheumatology, PO Box 5800, 6202AZ Maastricht, the Netherlands. rlan @sint.azm.nl</auth-address><titles><title>Presentation and analysis of radiographic data in clinical trials and observational studies</title><secondary-title>Annals of the rheumatic diseases</secondary-title></titles><periodical><full-title>Annals of the rheumatic diseases</full-title></periodical><pages>iv48-iv51</pages><volume>64</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>analysis</keyword><keyword>ARTHRITIS-RHEUMATOID/DT (drug therapy),*RA (radiography)</keyword><keyword>CLINICAL-TRIAL</keyword><keyword>CLINICAL-TRIALS-AS-TOPIC</keyword><keyword>DATA-INTERPRETATION-STATISTICAL</keyword><keyword>Disease</keyword><keyword>DISEASE-PROGRESSION</keyword><keyword>HUMANS</keyword><keyword>Internal-Medicine</keyword><keyword>Netherlands</keyword><keyword>PROBABILITY</keyword><keyword>radiography</keyword><keyword>TREATMENT-OUTCOME</keyword></keywords><dates><year>2005</year><pub-dates><date>11/2005</date></pub-dates></dates><label>186</label><urls></urls><access-date>2010</access-date></record></Cite></EndNote>(Landewe & van der Heijde, 2005). 3.3.5.Health Assessment Questionnaire Disability Index (HAQ–DI). The Stanford Health Assessment Questionnaire disability index (HAQ-DI) is a patient completed questionnaire specific for rheumatoid arthritis ADDIN EN.CITE <EndNote><Cite><Author>Fries</Author><Year>1980</Year><RecNum>88</RecNum><DisplayText>(Fries et al., 1980)</DisplayText><record><rec-number>88</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">88</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Fries, J.F.</author><author>Spitz, P.</author><author>Kraines, R.G.</author><author>Holman, H.R.</author></authors></contributors><titles><title>Measurement of patient outcome in arthritis</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>137-145</pages><volume>23</volume><number>2</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Disability Evaluation</keyword><keyword>Evaluation Studies as Topic</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>Interviews as Topic</keyword><keyword>MALE</keyword><keyword>methods</keyword><keyword>Middle Aged</keyword><keyword>Outcome and Process Assessment (Health Care)</keyword><keyword>QUESTIONNAIRES</keyword><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>1980</year><pub-dates><date>2/1980</date></pub-dates></dates><isbn>0004-3591</isbn><label>158</label><urls><related-urls><url>;(Fries et al., 1980). It consists of 20 questions referring to 8 domains: dressing/grooming, arising, eating, walking, hygiene, reach, grip, and common daily activities. Each domain has at least two component questions. There are four possible responses for each component.0 = without any difficulty1 = with some difficulty2 = with much difficulty3 = unable to do The HAQ score is a continuous outcome variable that is defines as the mean of the highest scores across the eight domains and ranges from 0 to 3, with higher levels reflecting greater disability.3.3.6.Functional Assessment of Chronic Illness–Fatigue. (FACIT-Fatigue)This brief, validated measure is used to monitor fatigue and the effects of this important symptom on the quality of life of patients with RA ADDIN EN.CITE <EndNote><Cite><Author>Cella</Author><Year>2005</Year><RecNum>12</RecNum><DisplayText>(Cella et al., 2005)</DisplayText><record><rec-number>12</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">12</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cella, D.</author><author>Yount, S.</author><author>Sorensen, M.</author><author>Chartash, E.</author><author>Sengupta, N.</author><author>Grober, J.</author></authors></contributors><auth-address>Center on Outcomes, Research and Education, Evanston Northwestern Healthcare and Northwestern University, Evanston, Illinois 60201, USA. d-cella@northwestern.edu</auth-address><titles><title>Validation of the Functional Assessment of Chronic Illness Therapy Fatigue Scale relative to other instrumentation in patients with rheumatoid arthritis</title><secondary-title>The Journal of rheumatology</secondary-title></titles><pages>811-819</pages><volume>32</volume><number>5</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>ARTHRITIS-RHEUMATOID/CO (complications),*PP (physiopathology)</keyword><keyword>CHRONIC-DISEASE</keyword><keyword>FATIGUE/ET (etiology),*PP (physiopathology)</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>PSYCHOMETRICS/*ST (standards)</keyword><keyword>QUESTIONNAIRES/*ST (standards)</keyword><keyword>RA</keyword><keyword>REPRODUCIBILITY-OF-RESULTS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX/*</keyword></keywords><dates><year>2005</year><pub-dates><date>5/2005</date></pub-dates></dates><label>13</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Cella et al., 2005). It was previously validated in patients with cancer and patients with anaemia ADDIN EN.CITE <EndNote><Cite><Author>Yellen</Author><Year>1997</Year><RecNum>117</RecNum><DisplayText>(Yellen et al., 1997)</DisplayText><record><rec-number>117</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">117</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Yellen, S.B.</author><author>Cella, D.F.</author><author>Webster, K.</author><author>Blendowski, C.</author><author>Kaplan, E.</author></authors></contributors><auth-address>Department of Psychology and Social Sciences, Rush-Presbyterian-St. Luke's Medical Center, Chicago, IL 60612, USA</auth-address><titles><title>Measuring fatigue and other anemia-related symptoms with the Functional Assessment of Cancer Therapy FACT measurement system</title><secondary-title>Journal of pain and symptom management</secondary-title></titles><pages>63-74</pages><volume>13</volume><number>2</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>ANEMIA/*CO (complications)</keyword><keyword>FATIGUE/*DI (diagnosis),ET (etiology)</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>NEOPLASMS/*TH (therapy)</keyword><keyword>PREDICTIVE-VALUE-OF-TESTS</keyword><keyword>QUALITY-OF-LIFE</keyword><keyword>QUESTIONNAIRES</keyword><keyword>REPRODUCIBILITY-OF-RESULTS</keyword></keywords><dates><year>1997</year><pub-dates><date>2/1997</date></pub-dates></dates><label>187</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Yellen et al., 1997). Patients are asked to rate 13 items on a scale of 0 (Not at all) to 4 (very much). Items include “I feel fatigued”, “I have trouble finishing things because I am tired”, and “I need to sleep during the day”. Possible FACIT-fatigue scores range from 0 to 52, with higher scores indicating more fatigue ADDIN EN.CITE <EndNote><Cite><Author>Cella</Author><Year>2005</Year><RecNum>12</RecNum><DisplayText>(Cella et al., 2005)</DisplayText><record><rec-number>12</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">12</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cella, D.</author><author>Yount, S.</author><author>Sorensen, M.</author><author>Chartash, E.</author><author>Sengupta, N.</author><author>Grober, J.</author></authors></contributors><auth-address>Center on Outcomes, Research and Education, Evanston Northwestern Healthcare and Northwestern University, Evanston, Illinois 60201, USA. d-cella@northwestern.edu</auth-address><titles><title>Validation of the Functional Assessment of Chronic Illness Therapy Fatigue Scale relative to other instrumentation in patients with rheumatoid arthritis</title><secondary-title>The Journal of rheumatology</secondary-title></titles><pages>811-819</pages><volume>32</volume><number>5</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>AGED-80-AND-OVER</keyword><keyword>ARTHRITIS-RHEUMATOID/CO (complications),*PP (physiopathology)</keyword><keyword>CHRONIC-DISEASE</keyword><keyword>FATIGUE/ET (etiology),*PP (physiopathology)</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>PSYCHOMETRICS/*ST (standards)</keyword><keyword>QUESTIONNAIRES/*ST (standards)</keyword><keyword>RA</keyword><keyword>REPRODUCIBILITY-OF-RESULTS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX/*</keyword></keywords><dates><year>2005</year><pub-dates><date>5/2005</date></pub-dates></dates><label>13</label><urls></urls><access-date>2005</access-date></record></Cite></EndNote>(Cella et al., 2005). Although the FACIT-fatigue can take integer values, it is usually treated as a continuous outcome.3.4.Regulatory pathway for approval of new therapiesIn the context of approval of novel therapies for RA there are 3 main regulatory claims which are often sought in a stepwise fashion: improvement of signs and symptoms, prevention of structural damage to joints, and improvement in physical function and disability ADDIN EN.CITE <EndNote><Cite><Author>EMA</Author><Year>2003</Year><RecNum>137</RecNum><DisplayText>(EMA, 2003; FDA, 2010)</DisplayText><record><rec-number>137</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">137</key></foreign-keys><ref-type name="Government Document">46</ref-type><contributors><authors><author>EMA</author></authors></contributors><titles><title>Points to Consider on the Clinical Investigation of Medicinal Products other than Nsaids in Rheumatoid Arthritis</title></titles><dates><year>2003</year></dates><urls></urls></record></Cite><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(EMA, 2003; FDA, 2010).The FDA state the claim for improvement of signs and symptoms is intended to reflect the demonstration of symptomatic benefit that includes improvement in the signs of the disease as well as symptoms ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010). This claim is usually the initial claim granted for marketing approval, and is based on following patients for a minimum of 24 weeks (6 months) ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010). The acceptable outcome measure that supports this claim is the American College of Rheumatology PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+QWxldGFoYTwvQXV0aG9yPjxZ
ZWFyPjIwMDg8L1llYXI+PFJlY051bT4yMDA8L1JlY051bT48RGlzcGxheVRleHQ+KEFsZXRhaGEg
ZXQgYWwuKTwvRGlzcGxheVRleHQ+PHJlY29yZD48cmVjLW51bWJlcj4yMDA8L3JlYy1udW1iZXI+
PGZvcmVpZ24ta2V5cz48a2V5IGFwcD0iRU4iIGRiLWlkPSI1YTk5d2RhYXp3ZHh0M2V4NXBmcHBh
ZTNmc3Bzd3Z0ZnJ4cmQiIHRpbWVzdGFtcD0iMTUxOTkwMzk5NCI+MjAwPC9rZXk+PGtleSBhcHA9
IkVOV2ViIiBkYi1pZD0iIj4wPC9rZXk+PC9mb3JlaWduLWtleXM+PHJlZi10eXBlIG5hbWU9Ikpv
dXJuYWwgQXJ0aWNsZSI+MTc8L3JlZi10eXBlPjxjb250cmlidXRvcnM+PGF1dGhvcnM+PGF1dGhv
cj5BbGV0YWhhLCBELjwvYXV0aG9yPjxhdXRob3I+TGFuZGV3ZSwgUi48L2F1dGhvcj48YXV0aG9y
Pkthcm9uaXRzY2gsIFQuPC9hdXRob3I+PGF1dGhvcj5CYXRob24sIEouPC9hdXRob3I+PGF1dGhv
cj5Cb2VycywgTS48L2F1dGhvcj48YXV0aG9yPkJvbWJhcmRpZXIsIEMuPC9hdXRob3I+PGF1dGhv
cj5Cb21iYXJkaWVyaSwgUy48L2F1dGhvcj48YXV0aG9yPkNob2ksIEguPC9hdXRob3I+PGF1dGhv
cj5Db21iZSwgQi48L2F1dGhvcj48YXV0aG9yPkRvdWdhZG9zLCBNLjwvYXV0aG9yPjxhdXRob3I+
RW1lcnksIFAuPC9hdXRob3I+PGF1dGhvcj5Hb21lei1SZWlubywgSi48L2F1dGhvcj48YXV0aG9y
PktleXN0b25lLCBFLjwvYXV0aG9yPjxhdXRob3I+S29jaCwgRy48L2F1dGhvcj48YXV0aG9yPkt2
aWVuLCBULiBLLjwvYXV0aG9yPjxhdXRob3I+TWFydGluLU1vbGEsIEUuPC9hdXRob3I+PGF1dGhv
cj5NYXR1Y2NpLUNlcmluaWMsIE0uPC9hdXRob3I+PGF1dGhvcj5NaWNoYXVkLCBLLjwvYXV0aG9y
PjxhdXRob3I+TyZhcG9zO0RlbGwsIEouPC9hdXRob3I+PGF1dGhvcj5QYXVsdXMsIEguPC9hdXRo
b3I+PGF1dGhvcj5QaW5jdXMsIFQuPC9hdXRob3I+PGF1dGhvcj5SaWNoYXJkcywgUC48L2F1dGhv
cj48YXV0aG9yPlNpbW9uLCBMLjwvYXV0aG9yPjxhdXRob3I+U2llZ2VsLCBKLjwvYXV0aG9yPjxh
dXRob3I+U21vbGVuLCBKLiBTLjwvYXV0aG9yPjxhdXRob3I+U29ra2EsIFQuPC9hdXRob3I+PGF1
dGhvcj5TdHJhbmQsIFYuPC9hdXRob3I+PGF1dGhvcj5UdWd3ZWxsLCBQLjwvYXV0aG9yPjxhdXRo
b3I+dmFuIGRlciBIZWlqZGUsIEQuPC9hdXRob3I+PGF1dGhvcj52YW4gUmllbCwgUC48L2F1dGhv
cj48YXV0aG9yPlZsYWQsIFMuPC9hdXRob3I+PGF1dGhvcj52YW4gVm9sbGVuaG92ZW4sIFIuPC9h
dXRob3I+PGF1dGhvcj5XYXJkLCBNLjwvYXV0aG9yPjxhdXRob3I+V2VpbmJsYXR0LCBNLjwvYXV0
aG9yPjxhdXRob3I+V2VsbHMsIEcuPC9hdXRob3I+PGF1dGhvcj5XaGl0ZSwgQi48L2F1dGhvcj48
YXV0aG9yPldvbGZlLCBGLjwvYXV0aG9yPjxhdXRob3I+WmhhbmcsIEIuPC9hdXRob3I+PGF1dGhv
cj5aaW5rLCBBLjwvYXV0aG9yPjxhdXRob3I+RmVsc29uLCBELjwvYXV0aG9yPjxhdXRob3I+RXVs
YXIsPC9hdXRob3I+PGF1dGhvcj5BY3IsPC9hdXRob3I+PC9hdXRob3JzPjwvY29udHJpYnV0b3Jz
PjxhdXRoLWFkZHJlc3M+TWVkaWNhbCBVbml2ZXJzaXR5IG9mIFZpZW5uYSwgVmllbm5hLCBBdXN0
cmlhLiBkYW5pZWwuYWxldGFoYUBtZWR1bml3aWVuLmFjLmF0PC9hdXRoLWFkZHJlc3M+PHRpdGxl
cz48dGl0bGU+UmVwb3J0aW5nIGRpc2Vhc2UgYWN0aXZpdHkgaW4gY2xpbmljYWwgdHJpYWxzIG9m
IHBhdGllbnRzIHdpdGggcmhldW1hdG9pZCBhcnRocml0aXM6IEVVTEFSL0FDUiBjb2xsYWJvcmF0
aXZlIHJlY29tbWVuZGF0aW9uczwvdGl0bGU+PHNlY29uZGFyeS10aXRsZT5BcnRocml0aXMgUmhl
dW08L3NlY29uZGFyeS10aXRsZT48L3RpdGxlcz48cGVyaW9kaWNhbD48ZnVsbC10aXRsZT5BcnRo
cml0aXMgUmhldW08L2Z1bGwtdGl0bGU+PC9wZXJpb2RpY2FsPjxwYWdlcz4xMzcxLTc8L3BhZ2Vz
Pjx2b2x1bWU+NTk8L3ZvbHVtZT48bnVtYmVyPjEwPC9udW1iZXI+PGVkaXRpb24+MjAwOC8wOS8z
MDwvZWRpdGlvbj48a2V5d29yZHM+PGtleXdvcmQ+QW50aXJoZXVtYXRpYyBBZ2VudHMvKnRoZXJh
cGV1dGljIHVzZTwva2V5d29yZD48a2V5d29yZD5BcnRocml0aXMsIFJoZXVtYXRvaWQvKmRydWcg
dGhlcmFweTwva2V5d29yZD48a2V5d29yZD5DbGluaWNhbCBUcmlhbHMgYXMgVG9waWMvKnN0YW5k
YXJkczwva2V5d29yZD48a2V5d29yZD5Db29wZXJhdGl2ZSBCZWhhdmlvcjwva2V5d29yZD48a2V5
d29yZD5FdmlkZW5jZS1CYXNlZCBNZWRpY2luZS8qbWV0aG9kczwva2V5d29yZD48a2V5d29yZD5I
dW1hbnM8L2tleXdvcmQ+PC9rZXl3b3Jkcz48ZGF0ZXM+PHllYXI+MjAwODwveWVhcj48cHViLWRh
dGVzPjxkYXRlPk9jdCAxNTwvZGF0ZT48L3B1Yi1kYXRlcz48L2RhdGVzPjxpc2JuPjAwMDQtMzU5
MSAoUHJpbnQpJiN4RDswMDA0LTM1OTEgKExpbmtpbmcpPC9pc2JuPjxhY2Nlc3Npb24tbnVtPjE4
ODIxNjQ4PC9hY2Nlc3Npb24tbnVtPjx1cmxzPjxyZWxhdGVkLXVybHM+PHVybD5odHRwczovL3d3
dy5uY2JpLm5sbS5uaWguZ292L3B1Ym1lZC8xODgyMTY0ODwvdXJsPjwvcmVsYXRlZC11cmxzPjwv
dXJscz48ZWxlY3Ryb25pYy1yZXNvdXJjZS1udW0+MTAuMTAwMi9hcnQuMjQxMjM8L2VsZWN0cm9u
aWMtcmVzb3VyY2UtbnVtPjwvcmVjb3JkPjwvQ2l0ZT48L0VuZE5vdGU+AG==
ADDIN EN.CITE PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+QWxldGFoYTwvQXV0aG9yPjxZ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ADDIN EN.CITE.DATA (Aletaha et al.) definition of improvement ACR20 (see Section 3.3.1 for detailed definition). Regulatory authorities’ submission requirements currently mandate the use of ACR20 as the minimum primary efficacy endpoint in marketing authorisation applications for RA treatments however, with better treatments now being initiated earlier in the course of the disease, treatment goals for RA have shifted towards achieving remission ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(EMA, 2003; FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%"> app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">137</key></foreign-keys><ref-type name="Government Document">46</ref-type><contributors><authors><author>EMA</author></authors></contributors><titles><title>Points to Consider on the Clinical Investigation of Medicinal Products other than Nsaids in Rheumatoid Arthritis</title></titles><dates><year>2003</year></dates><urls></urls></record></Cite></EndNote>(EMA, 2003; FDA, 2010).The FDA indicate that a claim for prevention of structural damage to joints is an important goal of RA therapy and that trials evaluating this outcome should be at least 1 year in duration ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010). They recommend that outcome measures to support the prevention of structural damage include x-ray evidence showing slowing of progression and prevention of new erosions.The FDA allows claims for improvement in physical function and disability to encourage long term trials in RA ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010). The primary outcome measure for this claim is the Health Assessment Questionnaire Disability Index (HAQ-DI), this is a validated measure and for this claim is reported at 2 to 5 years from study start.The US Food and Drug Administration ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010) in their 2010 guidance extend the ACR20 criteria to evaluate longer term and higher hurdle benefits for new treatments. They define major clinical response as a six-month period of success by the ACR 70.The definition of an ACR remission requires that 5 of the following criteria are met for at least two consecutive months: morning stiffness of less than 15 minutes, no fatigue (defined as a score of 52 on the Functional Assessment of Chronic Illness–Fatigue (FACIT-Fatigue) questionnaire), no joint pain, no joint tenderness or pain on motion, no soft tissue swelling in joints or tendon sheaths, and an ESR (Westergren method) of less than 30 mm/hr for females or 20 mm/hr for males ADDIN EN.CITE <EndNote><Cite><Author>Pinals</Author><Year>1981</Year><RecNum>91</RecNum><DisplayText>(Pinals et al., 1981)</DisplayText><record><rec-number>91</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">91</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Pinals, R.S.</author><author>Masi, A.T.</author><author>Larsen, R.A.</author></authors></contributors><titles><title>Preliminary criteria for clinical remission in rheumatoid arthritis</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>1308-1315</pages><volume>24</volume><number>10</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>analysis</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Blood Sedimentation</keyword><keyword>Demography</keyword><keyword>diagnosis</keyword><keyword>Disease</keyword><keyword>drug therapy</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>Joints</keyword><keyword>MALE</keyword><keyword>Middle Aged</keyword><keyword>Outcome and Process Assessment (Health Care)</keyword><keyword>physiopathology</keyword><keyword>Prospective Studies</keyword><keyword>RA</keyword><keyword>radiography</keyword><keyword>Remission,Spontaneous</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatoid Factor</keyword></keywords><dates><year>1981</year><pub-dates><date>10/1981</date></pub-dates></dates><isbn>0004-3591</isbn><label>161</label><urls><related-urls><url>;(Pinals et al., 1981).The FDA ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010) also define complete clinical response. Complete clinical response is intended to describe a therapeutic benefit of greater magnitude than the major clinical response. Complete clinical response is defined as a continuous 6 month period of both remission by ACR criteria and radiographic arrest (no radiographic progression ADDIN EN.CITE <EndNote><Cite><Author>Larsen</Author><Year>1977</Year><RecNum>89</RecNum><DisplayText>(Larsen et al., 1977)</DisplayText><record><rec-number>89</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">89</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Larsen, A.</author><author>Dale, K.</author><author>Eek, M.</author></authors></contributors><titles><title>Radiographic evaluation of rheumatoid arthritis and related conditions by standard reference films</title><secondary-title>Acta Radiol. Diagn. (Stockh)</secondary-title></titles><pages>481-491</pages><volume>18</volume><number>4</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Arthritis,Rheumatoid</keyword><keyword>Evaluation Studies as Topic</keyword><keyword>HUMANS</keyword><keyword>methods</keyword><keyword>radiography</keyword><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>1977</year><pub-dates><date>7/1977</date></pub-dates></dates><isbn>0567-8056</isbn><label>159</label><urls><related-urls><url>;(Larsen et al., 1977) or modified Sharp methods ADDIN EN.CITE <EndNote><Cite><Author>Sharp</Author><Year>1985</Year><RecNum>106</RecNum><DisplayText>(Sharp et al., 1985)</DisplayText><record><rec-number>106</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">106</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Sharp, J.T.</author><author>Young, D.Y.</author><author>Bluhm, G.B.</author><author>Brook, A.</author><author>Brower, A.C.</author><author>Corbett, M.</author><author>Decker, J.L.</author><author>Genant, H.K.</author><author>Gofton, J.P.</author><author>Goodman, N.</author></authors></contributors><titles><title>How many joints in the hands and wrists should be included in a score of radiologic abnormalities used to assess rheumatoid arthritis?</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>1326-1335</pages><volume>28</volume><number>12</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ARTHRITIS-RHEUMATOID/PA (pathology),*RA (radiography)</keyword><keyword>ARTHROGRAPHY/*MT (methods)</keyword><keyword>Computers</keyword><keyword>FINGER-JOINT/RA (radiography)</keyword><keyword>Hand</keyword><keyword>HAND/*</keyword><keyword>HUMANS</keyword><keyword>INFORMATION-SYSTEMS</keyword><keyword>Joints</keyword><keyword>Mathematics</keyword><keyword>methods</keyword><keyword>Rheumatoid arthritis</keyword><keyword>WRIST-JOINT/*RA (radiography)</keyword></keywords><dates><year>1985</year><pub-dates><date>12/1985</date></pub-dates></dates><label>176</label><urls></urls><access-date>2002</access-date></record></Cite></EndNote>(Sharp et al., 1985). Complete clinical response implies a benefit requiring ongoing drug therapy; remission is defined by the same result while off all anti-rheumatic drugs. The 1981 ACR remission criteria ADDIN EN.CITE <EndNote><Cite><Author>Pinals</Author><Year>1981</Year><RecNum>91</RecNum><DisplayText>(Pinals et al., 1981)</DisplayText><record><rec-number>91</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">91</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Pinals, R.S.</author><author>Masi, A.T.</author><author>Larsen, R.A.</author></authors></contributors><titles><title>Preliminary criteria for clinical remission in rheumatoid arthritis</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>1308-1315</pages><volume>24</volume><number>10</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>analysis</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Blood Sedimentation</keyword><keyword>Demography</keyword><keyword>diagnosis</keyword><keyword>Disease</keyword><keyword>drug therapy</keyword><keyword>FEMALE</keyword><keyword>HUMANS</keyword><keyword>Joints</keyword><keyword>MALE</keyword><keyword>Middle Aged</keyword><keyword>Outcome and Process Assessment (Health Care)</keyword><keyword>physiopathology</keyword><keyword>Prospective Studies</keyword><keyword>RA</keyword><keyword>radiography</keyword><keyword>Remission,Spontaneous</keyword><keyword>Rheumatoid arthritis</keyword><keyword>Rheumatoid Factor</keyword></keywords><dates><year>1981</year><pub-dates><date>10/1981</date></pub-dates></dates><isbn>0004-3591</isbn><label>161</label><urls><related-urls><url>;(Pinals et al., 1981) require at least five of the following: morning stiffness less than 15 minutes, no fatigue, no joint pain by history, no joint tenderness or pain on motion, no swelling of joints or tendon sheaths, and ESR less than 20 for males or less than 30 for females ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2010</Year><RecNum>41</RecNum><DisplayText>(FDA, 2010)</DisplayText><record><rec-number>41</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">41</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA)</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword><keyword>Biological Products</keyword><keyword>RA</keyword></keywords><dates><year>2010</year><pub-dates><date>2010</date></pub-dates></dates><label>42</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(FDA, 2010).The ACR20, ACR50, ACR70, the various methods of remission and response reduce multiple components to a dichotomous outcome PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5GZWxzb248L0F1dGhvcj48WWVhcj4xOTk1PC9ZZWFyPjxS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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5GZWxzb248L0F1dGhvcj48WWVhcj4xOTk1PC9ZZWFyPjxS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ADDIN EN.CITE.DATA (FDA, 2010; Felson et al., 1995).3.5.Current treatment strategies for rheumatoid arthritisThe early recognition and treatment of RA is considered important for minimising disability and maximising quality of life PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5OZWxsPC9BdXRob3I+PFllYXI+MjAwNDwvWWVhcj48UmVj
TnVtPjExODwvUmVjTnVtPjxEaXNwbGF5VGV4dD4oTmVsbCBldCBhbC4sIDIwMDQpPC9EaXNwbGF5
VGV4dD48cmVjb3JkPjxyZWMtbnVtYmVyPjExODwvcmVjLW51bWJlcj48Zm9yZWlnbi1rZXlzPjxr
ZXkgYXBwPSJFTiIgZGItaWQ9IjA5cHY1YXpwamV2dmRoZTVhczF2dnN6eXdmZnNyZnd0ZTVlNSIg
dGltZXN0YW1wPSIxNDQzNTM0OTgwIj4xMTg8L2tleT48L2ZvcmVpZ24ta2V5cz48cmVmLXR5cGUg
bmFtZT0iSm91cm5hbCBBcnRpY2xlIj4xNzwvcmVmLXR5cGU+PGNvbnRyaWJ1dG9ycz48YXV0aG9y
cz48YXV0aG9yPk5lbGwsIFYuUC5LLjwvYXV0aG9yPjxhdXRob3I+TWFjaG9sZCwgSy5QLjwvYXV0
aG9yPjxhdXRob3I+RWJlcmwsIEcuPC9hdXRob3I+PGF1dGhvcj5TdGFtbSwgVC5BLjwvYXV0aG9y
PjxhdXRob3I+VWZmbWFubiwgTS48L2F1dGhvcj48YXV0aG9yPlNtb2xlbiwgSi5TLjwvYXV0aG9y
PjwvYXV0aG9ycz48L2NvbnRyaWJ1dG9ycz48YXV0aC1hZGRyZXNzPkRlcGFydG1lbnQgb2YgUmhl
dW1hdG9sb2d5LCBJbnRlcm5hbCBNZWRpY2luZSBJSUksIFVuaXZlcnNpdHkgb2YgVmllbm5hLCBX
YWVocmluZ2VyIEd1ZXJ0ZWwgMTgtMjAsIEEtMTA5MCBWaWVubmEsIEF1c3RyaWE8L2F1dGgtYWRk
cmVzcz48dGl0bGVzPjx0aXRsZT5CZW5lZml0IG9mIHZlcnkgZWFybHkgcmVmZXJyYWwgYW5kIHZl
cnkgZWFybHkgdGhlcmFweSB3aXRoIGRpc2Vhc2UtIG1vZGlmeWluZyBhbnRpLXJoZXVtYXRpYyBk
cnVncyBpbiBwYXRpZW50cyB3aXRoIGVhcmx5IHJoZXVtYXRvaWQgYXJ0aHJpdGlzPC90aXRsZT48
c2Vjb25kYXJ5LXRpdGxlPlJoZXVtYXRvbG9neSAoT3hmb3JkIEVuZ2xhbmQpPC9zZWNvbmRhcnkt
dGl0bGU+PC90aXRsZXM+PHBlcmlvZGljYWw+PGZ1bGwtdGl0bGU+UmhldW1hdG9sb2d5IChPeGZv
cmQgRW5nbGFuZCk8L2Z1bGwtdGl0bGU+PC9wZXJpb2RpY2FsPjxwYWdlcz45MDYtOTE0PC9wYWdl
cz48dm9sdW1lPjQzPC92b2x1bWU+PG51bWJlcj43PC9udW1iZXI+PHJlcHJpbnQtZWRpdGlvbj5O
b3QgaW4gRmlsZTwvcmVwcmludC1lZGl0aW9uPjxrZXl3b3Jkcz48a2V5d29yZD4wIChBbnRpcmhl
dW1hdGljLUFnZW50cyk8L2tleXdvcmQ+PGtleXdvcmQ+QUNVVEUtRElTRUFTRTwva2V5d29yZD48
a2V5d29yZD5BRFVMVDwva2V5d29yZD48a2V5d29yZD5BR0VEPC9rZXl3b3JkPjxrZXl3b3JkPkFO
VElSSEVVTUFUSUMtQUdFTlRTLypUVSAodGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5d29y
ZD5BUlRIUklUSVMtUkhFVU1BVE9JRC8qRFQgKGRydWcgdGhlcmFweSk8L2tleXdvcmQ+PGtleXdv
cmQ+Q0FTRS1DT05UUk9MLVNUVURJRVM8L2tleXdvcmQ+PGtleXdvcmQ+ZGlhZ25vc2lzPC9rZXl3
b3JkPjxrZXl3b3JkPkRpc2Vhc2U8L2tleXdvcmQ+PGtleXdvcmQ+RGlzZWFzZSBQcm9ncmVzc2lv
bjwva2V5d29yZD48a2V5d29yZD5ESVNFQVNFLUFDVElWSVRZPC9rZXl3b3JkPjxrZXl3b3JkPkRJ
U0VBU0UtUFJPR1JFU1NJT048L2tleXdvcmQ+PGtleXdvcmQ+RUFSTFktRElBR05PU0lTPC9rZXl3
b3JkPjxrZXl3b3JkPkZFTUFMRTwva2V5d29yZD48a2V5d29yZD5GaXJzdDwva2V5d29yZD48a2V5
d29yZD5IVU1BTlM8L2tleXdvcmQ+PGtleXdvcmQ+SW50ZXJuYWwtTWVkaWNpbmU8L2tleXdvcmQ+
PGtleXdvcmQ+TUFMRTwva2V5d29yZD48a2V5d29yZD5tZXRob2RzPC9rZXl3b3JkPjxrZXl3b3Jk
Pk1JRERMRS1BR0VEPC9rZXl3b3JkPjxrZXl3b3JkPlBST0dOT1NJUzwva2V5d29yZD48a2V5d29y
ZD5QUk9TUEVDVElWRS1TVFVESUVTPC9rZXl3b3JkPjxrZXl3b3JkPlJBPC9rZXl3b3JkPjxrZXl3
b3JkPlJoZXVtYXRvaWQgYXJ0aHJpdGlzPC9rZXl3b3JkPjxrZXl3b3JkPlJoZXVtYXRvbG9neTwv
a2V5d29yZD48a2V5d29yZD5UUkVBVE1FTlQtT1VUQ09NRTwva2V5d29yZD48L2tleXdvcmRzPjxk
YXRlcz48eWVhcj4yMDA0PC95ZWFyPjxwdWItZGF0ZXM+PGRhdGU+Ny8yMDA0PC9kYXRlPjwvcHVi
LWRhdGVzPjwvZGF0ZXM+PGxhYmVsPjE4ODwvbGFiZWw+PHVybHM+PC91cmxzPjxhY2Nlc3MtZGF0
ZT4yMDA2PC9hY2Nlc3MtZGF0ZT48L3JlY29yZD48L0NpdGU+PC9FbmROb3RlPgB=
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5OZWxsPC9BdXRob3I+PFllYXI+MjAwNDwvWWVhcj48UmVj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=
ADDIN EN.CITE.DATA (Nell et al., 2004). A delay in treatment of just a few months can result in significantly more joint damage subsequently in the course of disease PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MYXJkPC9BdXRob3I+PFllYXI+MjAwMTwvWWVhcj48UmVj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==
ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MYXJkPC9BdXRob3I+PFllYXI+MjAwMTwvWWVhcj48UmVj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==
ADDIN EN.CITE.DATA (Lard et al., 2001).The optimal control of RA is best achieved by regularly assessing disease progression (every 2–4 months) and subsequently modifying treatment if appropriate ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003).Therapies for RA generally fall into two broad categories ADDIN EN.CITE <EndNote><Cite><Author>Smolen</Author><Year>2003</Year><RecNum>42</RecNum><DisplayText>(Smolen & Steiner, 2003)</DisplayText><record><rec-number>42</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">42</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>Smolen, J.</author><author>Steiner, G.</author></authors></contributors><titles><title>Therapeutic Strategies for Rheumatoid Arthritis</title><secondary-title>Nature Reviews Drug Discovery</secondary-title></titles><pages>473-488</pages><volume>2</volume><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2003</year><pub-dates><date>6 AD</date></pub-dates></dates><label>43</label><urls></urls></record></Cite></EndNote>(Smolen & Steiner, 2003):Therapies for symptomatic treatment; Disease-modifying agents ADDIN EN.CITE <EndNote><Cite><Author>Gaffo</Author><Year>2006</Year><RecNum>85</RecNum><DisplayText>(Gaffo et al., 2006)</DisplayText><record><rec-number>85</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">85</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Gaffo, A.</author><author>Saag, K.G.</author><author>Curtis, J.R.</author></authors></contributors><auth-address>Center for Education and Research and Therapeutics of Musculoskeletal Diseases, University of Alabama at Birmingham, Birmingham, AL, USA</auth-address><titles><title>Treatment of rheumatoid arthritis</title><secondary-title>Am. J. Health Syst. Pharm</secondary-title></titles><pages>2451-2465</pages><volume>63</volume><number>24</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Biological Products</keyword><keyword>Drug Delivery Systems</keyword><keyword>drug therapy</keyword><keyword>Drugs,Investigational</keyword><keyword>First</keyword><keyword>Glucocorticoids</keyword><keyword>HUMANS</keyword><keyword>Immunologic Factors</keyword><keyword>Methotrexate</keyword><keyword>physiopathology</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2006</year><pub-dates><date>12/15/2006</date></pub-dates></dates><isbn>1079-2082</isbn><label>155</label><urls><related-urls><url>;(Gaffo et al., 2006). Therapies for symptomatic treatment provide relief from the pain and inflammation characteristic of RA but are unable to prevent progressive joint damage ADDIN EN.CITE <EndNote><Cite><Author>Gaffo</Author><Year>2006</Year><RecNum>85</RecNum><DisplayText>(Gaffo et al., 2006)</DisplayText><record><rec-number>85</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">85</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Gaffo, A.</author><author>Saag, K.G.</author><author>Curtis, J.R.</author></authors></contributors><auth-address>Center for Education and Research and Therapeutics of Musculoskeletal Diseases, University of Alabama at Birmingham, Birmingham, AL, USA</auth-address><titles><title>Treatment of rheumatoid arthritis</title><secondary-title>Am. J. Health Syst. Pharm</secondary-title></titles><pages>2451-2465</pages><volume>63</volume><number>24</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Biological Products</keyword><keyword>Drug Delivery Systems</keyword><keyword>drug therapy</keyword><keyword>Drugs,Investigational</keyword><keyword>First</keyword><keyword>Glucocorticoids</keyword><keyword>HUMANS</keyword><keyword>Immunologic Factors</keyword><keyword>Methotrexate</keyword><keyword>physiopathology</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2006</year><pub-dates><date>12/15/2006</date></pub-dates></dates><isbn>1079-2082</isbn><label>155</label><urls><related-urls><url>;(Gaffo et al., 2006). They include non-steroidal anti-inflammatory drugs, selective cox-2 inhibitors and aspirin, as well as glucocorticoid drugs, such as prednisolone ADDIN EN.CITE <EndNote><Cite><Author>Gaffo</Author><Year>2006</Year><RecNum>85</RecNum><DisplayText>(Gaffo et al., 2006)</DisplayText><record><rec-number>85</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">85</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Gaffo, A.</author><author>Saag, K.G.</author><author>Curtis, J.R.</author></authors></contributors><auth-address>Center for Education and Research and Therapeutics of Musculoskeletal Diseases, University of Alabama at Birmingham, Birmingham, AL, USA</auth-address><titles><title>Treatment of rheumatoid arthritis</title><secondary-title>Am. J. Health Syst. Pharm</secondary-title></titles><pages>2451-2465</pages><volume>63</volume><number>24</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Biological Products</keyword><keyword>Drug Delivery Systems</keyword><keyword>drug therapy</keyword><keyword>Drugs,Investigational</keyword><keyword>First</keyword><keyword>Glucocorticoids</keyword><keyword>HUMANS</keyword><keyword>Immunologic Factors</keyword><keyword>Methotrexate</keyword><keyword>physiopathology</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2006</year><pub-dates><date>12/15/2006</date></pub-dates></dates><isbn>1079-2082</isbn><label>155</label><urls><related-urls><url>;(Gaffo et al., 2006). Glucocorticoids can relieve inflammation and in high doses can have a dramatic effect on the symptoms of RA, however, they are generally used only temporarily and at low doses because of their association with serious adverse effects such as fractures, serious infections and gastrointestinal events PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Hb3R6c2NoZTwvQXV0aG9yPjxZZWFyPjE5OTg8L1llYXI+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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Hb3R6c2NoZTwvQXV0aG9yPjxZZWFyPjE5OTg8L1llYXI+
PFJlY051bT41PC9SZWNOdW0+PERpc3BsYXlUZXh0PihHb3R6c2NoZSAmYW1wOyBKb2hhbnNlbiwg
MTk5ODsgU2FhZyBldCBhbC4sIDE5OTQpPC9EaXNwbGF5VGV4dD48cmVjb3JkPjxyZWMtbnVtYmVy
PjU8L3JlYy1udW1iZXI+PGZvcmVpZ24ta2V5cz48a2V5IGFwcD0iRU4iIGRiLWlkPSI1YTk5d2Rh
YXp3ZHh0M2V4NXBmcHBhZTNmc3Bzd3Z0ZnJ4cmQiIHRpbWVzdGFtcD0iMCI+NTwva2V5PjwvZm9y
ZWlnbi1rZXlzPjxyZWYtdHlwZSBuYW1lPSJKb3VybmFsIEFydGljbGUiPjE3PC9yZWYtdHlwZT48
Y29udHJpYnV0b3JzPjxhdXRob3JzPjxhdXRob3I+R290enNjaGUsIFAuQy48L2F1dGhvcj48YXV0
aG9yPkpvaGFuc2VuLCBILksuPC9hdXRob3I+PC9hdXRob3JzPjwvY29udHJpYnV0b3JzPjxhdXRo
LWFkZHJlc3M+Tm9yZGljIENvY2hyYW5lIENlbnRyZSwgUmlnc2hvc3BpdGFsZXQsIERlcGFydG1l
bnQgNzExMiwgQ29wZW5oYWdlbiwgRGVubWFyazwvYXV0aC1hZGRyZXNzPjx0aXRsZXM+PHRpdGxl
Pk1ldGEtYW5hbHlzaXMgb2Ygc2hvcnQtdGVybSBsb3cgZG9zZSBwcmVkbmlzb2xvbmUgdmVyc3Vz
IHBsYWNlYm8gYW5kIG5vbi1zdGVyb2lkYWwgYW50aS1pbmZsYW1tYXRvcnkgZHJ1Z3MgaW4gcmhl
dW1hdG9pZCBhcnRocml0aXM8L3RpdGxlPjxzZWNvbmRhcnktdGl0bGU+Qk1KIChDbGluaWNhbCBy
ZXNlYXJjaCBlZC4gKTwvc2Vjb25kYXJ5LXRpdGxlPjwvdGl0bGVzPjxwYWdlcz44MTEtODE4PC9w
YWdlcz48dm9sdW1lPjMxNjwvdm9sdW1lPjxudW1iZXI+NzEzNDwvbnVtYmVyPjxyZXByaW50LWVk
aXRpb24+Tm90IGluIEZpbGU8L3JlcHJpbnQtZWRpdGlvbj48a2V5d29yZHM+PGtleXdvcmQ+MCAo
QW50aS1JbmZsYW1tYXRvcnktQWdlbnRzKTwva2V5d29yZD48a2V5d29yZD4wIChBbnRpLUluZmxh
bW1hdG9yeS1BZ2VudHMtTm9uLVN0ZXJvaWRhbCk8L2tleXdvcmQ+PGtleXdvcmQ+NTAtMjQtOCAo
UHJlZG5pc29sb25lKTwva2V5d29yZD48a2V5d29yZD5BRE1JTklTVFJBVElPTi1PUkFMPC9rZXl3
b3JkPjxrZXl3b3JkPkFOVEktSU5GTEFNTUFUT1JZLUFHRU5UUy1OT04tU1RFUk9JREFMLypUVSAo
dGhlcmFwZXV0aWMgdXNlKTwva2V5d29yZD48a2V5d29yZD5BTlRJLUlORkxBTU1BVE9SWS1BR0VO
VFMvKkFEIChhZG1pbmlzdHJhdGlvbiAmYW1wOyBkb3NhZ2UpLEFFIChhZHZlcnNlIGVmZmVjdHMp
PC9rZXl3b3JkPjxrZXl3b3JkPkFSVEhSSVRJUy1SSEVVTUFUT0lELypEVCAoZHJ1ZyB0aGVyYXB5
KTwva2V5d29yZD48a2V5d29yZD5ERU5NQVJLPC9rZXl3b3JkPjxrZXl3b3JkPkhBTkQtU1RSRU5H
VEg8L2tleXdvcmQ+PGtleXdvcmQ+SFVNQU5TPC9rZXl3b3JkPjxrZXl3b3JkPlBBSU4vUEMgKHBy
ZXZlbnRpb24gJmFtcDsgY29udHJvbCk8L2tleXdvcmQ+PGtleXdvcmQ+UFJFRE5JU09MT05FLypB
RCAoYWRtaW5pc3RyYXRpb24gJmFtcDsgZG9zYWdlKSxBRSAoYWR2ZXJzZSBlZmZlY3RzKTwva2V5
d29yZD48a2V5d29yZD5SQU5ET01JWkVELUNPTlRST0xMRUQtVFJJQUxTLUFTLVRPUElDL01UICht
ZXRob2RzKSxTVCAoc3RhbmRhcmRzKTwva2V5d29yZD48a2V5d29yZD5SaGV1bWF0b2lkIGFydGhy
aXRpczwva2V5d29yZD48a2V5d29yZD5UUkVBVE1FTlQtT1VUQ09NRTwva2V5d29yZD48L2tleXdv
cmRzPjxkYXRlcz48eWVhcj4xOTk4PC95ZWFyPjxwdWItZGF0ZXM+PGRhdGU+My8xNC8xOTk4PC9k
YXRlPjwvcHViLWRhdGVzPjwvZGF0ZXM+PGxhYmVsPjY8L2xhYmVsPjx1cmxzPjwvdXJscz48YWNj
ZXNzLWRhdGU+MjAwNDwvYWNjZXNzLWRhdGU+PC9yZWNvcmQ+PC9DaXRlPjxDaXRlPjxBdXRob3I+
U2FhZzwvQXV0aG9yPjxZZWFyPjE5OTQ8L1llYXI+PFJlY051bT43MzwvUmVjTnVtPjxyZWNvcmQ+
PHJlYy1udW1iZXI+NzM8L3JlYy1udW1iZXI+PGZvcmVpZ24ta2V5cz48a2V5IGFwcD0iRU4iIGRi
LWlkPSI1YTk5d2RhYXp3ZHh0M2V4NXBmcHBhZTNmc3Bzd3Z0ZnJ4cmQiIHRpbWVzdGFtcD0iMCI+
NzM8L2tleT48L2ZvcmVpZ24ta2V5cz48cmVmLXR5cGUgbmFtZT0iSm91cm5hbCBBcnRpY2xlIj4x
NzwvcmVmLXR5cGU+PGNvbnRyaWJ1dG9ycz48YXV0aG9ycz48YXV0aG9yPlNhYWcsIEsuRy48L2F1
dGhvcj48YXV0aG9yPktvZWhua2UsIFIuPC9hdXRob3I+PGF1dGhvcj5DYWxkd2VsbCwgSi5SLjwv
YXV0aG9yPjxhdXRob3I+QnJhc2luZ3RvbiwgUi48L2F1dGhvcj48YXV0aG9yPkJ1cm1laXN0ZXIs
IEwuRi48L2F1dGhvcj48YXV0aG9yPlppbW1lcm1hbiwgQi48L2F1dGhvcj48YXV0aG9yPktvaGxl
ciwgSi5BLjwvYXV0aG9yPjxhdXRob3I+RnVyc3QsIEQuRS48L2F1dGhvcj48L2F1dGhvcnM+PC9j
b250cmlidXRvcnM+PGF1dGgtYWRkcmVzcz5EZXBhcnRtZW50IG9mIEludGVybmFsIE1lZGljaW5l
LCBVbml2ZXJzaXR5IG9mIElvd2EsIElvd2EgQ2l0eTwvYXV0aC1hZGRyZXNzPjx0aXRsZXM+PHRp
dGxlPkxvdyBkb3NlIGxvbmctdGVybSBjb3J0aWNvc3Rlcm9pZCB0aGVyYXB5IGluIHJoZXVtYXRv
aWQgYXJ0aHJpdGlzOiBhbiBhbmFseXNpcyBvZiBzZXJpb3VzIGFkdmVyc2UgZXZlbnRzPC90aXRs
ZT48c2Vjb25kYXJ5LXRpdGxlPlRoZSBBbWVyaWNhbiBqb3VybmFsIG9mIG1lZGljaW5lPC9zZWNv
bmRhcnktdGl0bGU+PC90aXRsZXM+PHBhZ2VzPjExNS0xMjM8L3BhZ2VzPjx2b2x1bWU+OTY8L3Zv
bHVtZT48bnVtYmVyPjI8L251bWJlcj48cmVwcmludC1lZGl0aW9uPk5vdCBpbiBGaWxlPC9yZXBy
aW50LWVkaXRpb24+PGtleXdvcmRzPjxrZXl3b3JkPjAgKEFudGktSW5mbGFtbWF0b3J5LUFnZW50
cyk8L2tleXdvcmQ+PGtleXdvcmQ+NTMtMDMtMiAoUHJlZG5pc29uZSk8L2tleXdvcmQ+PGtleXdv
cmQ+QURVTFQ8L2tleXdvcmQ+PGtleXdvcmQ+QU5USS1JTkZMQU1NQVRPUlktQUdFTlRTL0FEIChh
ZG1pbmlzdHJhdGlvbiAmYW1wOyBkb3NhZ2UpLEFFIChhZHZlcnNlIGVmZmVjdHMpPC9rZXl3b3Jk
PjxrZXl3b3JkPkFSVEhSSVRJUy1SSEVVTUFUT0lELypEVCAoZHJ1ZyB0aGVyYXB5KTwva2V5d29y
ZD48a2V5d29yZD5EUlVHLUFETUlOSVNUUkFUSU9OLVNDSEVEVUxFPC9rZXl3b3JkPjxrZXl3b3Jk
PkZFTUFMRTwva2V5d29yZD48a2V5d29yZD5GaXJzdDwva2V5d29yZD48a2V5d29yZD5GT0xMT1ct
VVAtU1RVRElFUzwva2V5d29yZD48a2V5d29yZD5IVU1BTlM8L2tleXdvcmQ+PGtleXdvcmQ+TE9H
SVNUSUMtTU9ERUxTPC9rZXl3b3JkPjxrZXl3b3JkPk1BTEU8L2tleXdvcmQ+PGtleXdvcmQ+TUVE
SUNBTC1SRUNPUkRTPC9rZXl3b3JkPjxrZXl3b3JkPk1JRERMRS1BR0VEPC9rZXl3b3JkPjxrZXl3
b3JkPk9ERFMtUkFUSU88L2tleXdvcmQ+PGtleXdvcmQ+UFJFRE5JU09ORS9BRCAoYWRtaW5pc3Ry
YXRpb24gJmFtcDsgZG9zYWdlKSwqQUUgKGFkdmVyc2UgZWZmZWN0cyk8L2tleXdvcmQ+PGtleXdv
cmQ+UkE8L2tleXdvcmQ+PGtleXdvcmQ+UkVHUkVTU0lPTi1BTkFMWVNJUzwva2V5d29yZD48a2V5
d29yZD5SRVRST1NQRUNUSVZFLVNUVURJRVM8L2tleXdvcmQ+PGtleXdvcmQ+UmhldW1hdG9pZCBh
cnRocml0aXM8L2tleXdvcmQ+PGtleXdvcmQ+UkhFVU1BVE9JRC1OT0RVTEU8L2tleXdvcmQ+PGtl
eXdvcmQ+VElNRS1GQUNUT1JTPC9rZXl3b3JkPjxrZXl3b3JkPlVOSVRFRC1TVEFURVM8L2tleXdv
cmQ+PC9rZXl3b3Jkcz48ZGF0ZXM+PHllYXI+MTk5NDwveWVhcj48cHViLWRhdGVzPjxkYXRlPjIv
MTk5NDwvZGF0ZT48L3B1Yi1kYXRlcz48L2RhdGVzPjxsYWJlbD43NTwvbGFiZWw+PHVybHM+PC91
cmxzPjxhY2Nlc3MtZGF0ZT4yMDAzPC9hY2Nlc3MtZGF0ZT48L3JlY29yZD48L0NpdGU+PC9FbmRO
b3RlPn==
ADDIN EN.CITE.DATA (Gotzsche & Johansen, 1998; Saag et al., 1994).Disease-modifying agents have a demonstrated ability to halt the destructive course of RA and prevent joint damage. This category of drugs can be further subdivided into two groups:Synthetic (non-biologic/traditional/conventional) disease-modifying anti-rheumatic drugs (DMARDs), such as methotrexate, act as general anti-inflammatory and anti-proliferative agents ADDIN EN.CITE <EndNote><Cite><Author>Gaffo</Author><Year>2006</Year><RecNum>85</RecNum><DisplayText>(Gaffo et al., 2006)</DisplayText><record><rec-number>85</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">85</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Gaffo, A.</author><author>Saag, K.G.</author><author>Curtis, J.R.</author></authors></contributors><auth-address>Center for Education and Research and Therapeutics of Musculoskeletal Diseases, University of Alabama at Birmingham, Birmingham, AL, USA</auth-address><titles><title>Treatment of rheumatoid arthritis</title><secondary-title>Am. J. Health Syst. Pharm</secondary-title></titles><pages>2451-2465</pages><volume>63</volume><number>24</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>Anti-Inflammatory Agents,Non-Steroidal</keyword><keyword>Antirheumatic Agents</keyword><keyword>Arthritis,Rheumatoid</keyword><keyword>Biological Products</keyword><keyword>Drug Delivery Systems</keyword><keyword>drug therapy</keyword><keyword>Drugs,Investigational</keyword><keyword>First</keyword><keyword>Glucocorticoids</keyword><keyword>HUMANS</keyword><keyword>Immunologic Factors</keyword><keyword>Methotrexate</keyword><keyword>physiopathology</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>therapeutic use</keyword></keywords><dates><year>2006</year><pub-dates><date>12/15/2006</date></pub-dates></dates><isbn>1079-2082</isbn><label>155</label><urls><related-urls><url>;(Gaffo et al., 2006). These agents tend to have a slower onset of action than corticosteroids and lack a direct pain relief effect ADDIN EN.CITE <EndNote><Cite><Author>O'Dell</Author><Year>2004</Year><RecNum>63</RecNum><DisplayText>(O'Dell, 2004)</DisplayText><record><rec-number>63</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">63</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>O'Dell, J.</author></authors></contributors><auth-address>Department of Internal Medicine, University of Nebraska Medical Center, Omaha 68198-3025, USA. jrodell@unmc.edu</auth-address><titles><title>Therapeutic strategies for rheumatoid arthritis</title><secondary-title>The New England journal of medicine</secondary-title></titles><pages>2591-2602</pages><volume>350</volume><number>25</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Adrenal-Cortex-Hormones)</keyword><keyword>0 (Anti-Inflammatory-Agents-Non-Steroidal)</keyword><keyword>0 (Antirheumatic-Agents)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>372-75-8 (Citrulline)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ADRENAL-CORTEX-HORMONES/TU (therapeutic use)</keyword><keyword>ANTI-INFLAMMATORY-AGENTS-NON-STEROIDAL/TU (therapeutic use)</keyword><keyword>ANTIRHEUMATIC-AGENTS/*TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/DI (diagnosis),DT (drug therapy),*TH (therapy)</keyword><keyword>AUTOANTIBODIES/BL (blood)</keyword><keyword>CITRULLINE/IM (immunology)</keyword><keyword>COMORBIDITY</keyword><keyword>DRUG-THERAPY-COMBINATION</keyword><keyword>HUMANS</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/BL (blood)</keyword></keywords><dates><year>2004</year><pub-dates><date>6/17/2004</date></pub-dates></dates><label>64</label><urls></urls><access-date>2006</access-date></record></Cite></EndNote>(O'Dell, 2004);Biological agents (biologics), such as anti-Tumour Necrosis Factors (anti-TNFs) (adalimumab, etanercept, infliximab), B-cell targeted therapies (rituximab), T-lymphocyte co-stimulation modulators (abatacept) and Inteleukin-1 (IL-1) inhibitors (anakinra), are immune modulators that directly target the intracellular signalling pathways, cytokines, and other mediators contributing to the pathogenesis of RA ADDIN EN.CITE <EndNote><Cite><Author>O'Dell</Author><Year>2004</Year><RecNum>63</RecNum><DisplayText>(O'Dell, 2004)</DisplayText><record><rec-number>63</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">63</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>O'Dell, J.</author></authors></contributors><auth-address>Department of Internal Medicine, University of Nebraska Medical Center, Omaha 68198-3025, USA. jrodell@unmc.edu</auth-address><titles><title>Therapeutic strategies for rheumatoid arthritis</title><secondary-title>The New England journal of medicine</secondary-title></titles><pages>2591-2602</pages><volume>350</volume><number>25</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>0 (Adrenal-Cortex-Hormones)</keyword><keyword>0 (Anti-Inflammatory-Agents-Non-Steroidal)</keyword><keyword>0 (Antirheumatic-Agents)</keyword><keyword>0 (Autoantibodies)</keyword><keyword>372-75-8 (Citrulline)</keyword><keyword>9009-79-4 (Rheumatoid-Factor)</keyword><keyword>ADRENAL-CORTEX-HORMONES/TU (therapeutic use)</keyword><keyword>ANTI-INFLAMMATORY-AGENTS-NON-STEROIDAL/TU (therapeutic use)</keyword><keyword>ANTIRHEUMATIC-AGENTS/*TU (therapeutic use)</keyword><keyword>ARTHRITIS-RHEUMATOID/DI (diagnosis),DT (drug therapy),*TH (therapy)</keyword><keyword>AUTOANTIBODIES/BL (blood)</keyword><keyword>CITRULLINE/IM (immunology)</keyword><keyword>COMORBIDITY</keyword><keyword>DRUG-THERAPY-COMBINATION</keyword><keyword>HUMANS</keyword><keyword>PROGNOSIS</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-FACTOR/BL (blood)</keyword></keywords><dates><year>2004</year><pub-dates><date>6/17/2004</date></pub-dates></dates><label>64</label><urls></urls><access-date>2006</access-date></record></Cite></EndNote>(O'Dell, 2004). They are usually prescribed in combination with methotrexate or in some cases as monotherapy if patients have a history of methotrexate intolerance or if methotrexate is inappropriate.Despite the increasing range of synthetic and biologic agents available for the treatment of RA, evidence suggests that approximately 30–40% of patients fail to respond, or respond inadequately, to therapy PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Wb2xsPC9BdXRob3I+PFllYXI+MjAwNTwvWWVhcj48UmVj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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Wb2xsPC9BdXRob3I+PFllYXI+MjAwNTwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Voll & Kalden, 2005). In addition, many patients are unable to tolerate currently available treatment options. In England and Wales the National Institute for Health and Clinical Excellence (NICE) publish guidelines for management and treatment for RA in adults ADDIN EN.CITE <EndNote><Cite><Author>NICE</Author><Year>2009</Year><RecNum>45</RecNum><DisplayText>(NICE, 2009, 2015)</DisplayText><record><rec-number>45</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">45</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>NICE</author></authors></contributors><titles><title>Rheumatoid Arthritis National clinical guideline for management and treatment in adults</title></titles><reprint-edition>Not in File</reprint-edition><keywords><keyword>Rheumatoid arthritis</keyword></keywords><dates><year>2009</year><pub-dates><date>2009</date></pub-dates></dates><label>46</label><urls><related-urls><url><style face="underline" font="default" size="100%"> app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1512496305">190</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>NICE</author></authors></contributors><titles><title>Rheumatoid arthritis in adults: management</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2015</year><pub-dates><date>December 2015</date></pub-dates></dates><urls><related-urls><url>;(NICE, 2009, 2015). The NICE guidelines recommend a combination of DMARDs (including methotrexate and at least one other DMARD, plus short-term glucocorticoids) as first-line treatment as soon as possible. The guideline also recommends “an annual review to: assess disease activity and damage, and measure functional ability (e.g. Health Assessment Questionnaire)”.3.6.Phase III Clinical Trial DesignThe costs of developing a new medicine are very high. A typical clinical development plan for registration of an RA drug will include multiple phase I, phase II and phase III trials. A typical phase III trial costs an estimated $11.9 million ADDIN EN.CITE <EndNote><Cite><Author>Sertkaya</Author><Year>2016</Year><RecNum>181</RecNum><DisplayText>(Sertkaya, 2016)</DisplayText><record><rec-number>181</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1500394673">181</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Sertkaya, A., Wong, H.,Jessup, A., Beleche, T.</author></authors></contributors><titles><title>Key cost drivers of pharmaceutical clinical trials in the United States</title><secondary-title>Clinical Trials</secondary-title></titles><periodical><full-title>Clinical Trials</full-title></periodical><pages>117-126</pages><volume>13</volume><number>2</number><section>117</section><dates><year>2016</year></dates><urls></urls></record></Cite></EndNote>(Sertkaya, 2016). In the light of regulatory guidance described in Section 3.4, phase III confirmatory trials are usually designed for a patient duration of at least 6 months. The usual primary endpoint is the ACR20 (as described in Section 3.3.1). Typical phase III trials at Roche are randomised, parallel group, double blind (for first 6 or 12 months). REF _Ref490046874 \h \* MERGEFORMAT Figure 3.2 illustrates a typical phase III design. A background therapy of stable methotrexate is included in each, unless the objective is for approval for monotherapy. Due to the availability of effective treatments, in more recent trials, patients have the option to ‘escape’ onto active treatment after 16 weeks based on lack of therapeutic response. Such patients are then considered non responders for ACR20 assessment at 6 months. Figure STYLEREF 1 \s 3. SEQ Figure \* ARABIC \s 1 2 Typical RA Phase III Trials at RocheAssessment of ACR20 at 6 months allows a claim for improvement of signs and symptoms. Trials that continue to 1 year and that collect X-ray data, can be submitted to health authorities for a claim of prevention of structural damage to joints. Furthermore, trials that continue to 2 years in duration can claim for an improvement in physical function and disability based upon improvement in Health Assessment Questionnaire Disability Index (HAQ-DI). This strategy is still followed by pharmaceutical companies currently developing RA treatments. On 28th June 2017 Janssen presented their clinical development program of Sirukumab?, which included 2 pivotal phase III trials. Primary endpoints were ACR20 at week 16 for improvement of signs and symptoms, inhibition of structural damage measured by Sharp score at 1-year and improvement in physical function measured by HAQ-DI at 2 years ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2017</Year><RecNum>187</RecNum><DisplayText>(FDA, 2017)</DisplayText><record><rec-number>187</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1506341386">187</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Briefing Document Arthritis Advisory Committee PLIVENSIA? (sirukumab)</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2017</year></dates><urls><related-urls><url>;(FDA, 2017).The design of clinical trials in RA has the potential to benefit from application of enrichment designs. In 2012, the FDA published a draft guidance on enrichment strategies for clinical trials ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2012</Year><RecNum>186</RecNum><DisplayText>(FDA, 2012)</DisplayText><record><rec-number>186</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502704051">186</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry: Enrichment Strategies for Clinical Trials to Support Approval of Human Drugs and Biological Products</title></titles><dates><year>2012</year></dates><urls></urls></record></Cite></EndNote>(FDA, 2012). The guidance defines enrichment as “the prospective use of any patient characteristic to select a study population in which detection of a drug effect (if one is in fact present) is more likely than it would be in an unselected population”, where characteristics can be drawn from demographic, pathophysiologic, historical, genetic or proteomic, clinical, and psychological sources. By recruiting patients with a higher likelihood of having a certain event can lead to a higher effect size, reduce heterogeneity and lead to smaller more efficient clinical trials. 3.7.SummaryThis chapter described the key disease change and disease status measures: ACR20, DAS28, EULAR Score and radiologic evaluations. In the main, these measures rely on subjective patient or physician reported outcomes at fixed points in time. The exceptions being ESR and CRP components of ACR20 and DAS28 and the radiologic assessments. The measures do not include any acknowledgement of within patient variability, although one would expect high correlation of subjective assessments from visit to visit. E.g. arthritic pain assessed by a patient at month 2 would likely be recorded based on a comparison to the level of pain at month 1. Ward ADDIN EN.CITE <EndNote><Cite><Author>Ward</Author><Year>2004</Year><RecNum>403</RecNum><DisplayText>(Ward, 2004)</DisplayText><record><rec-number>403</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1524058593">403</key></foreign-keys><ref-type name="Book Section">5</ref-type><contributors><authors><author>Ward, M.</author></authors><secondary-authors><author>St Clair, E.W.</author></secondary-authors></contributors><titles><title>Clinical and Laboratory Measures</title><secondary-title>Rheumatoid Arthiritis</secondary-title></titles><dates><year>2004</year></dates><pub-location>USA</pub-location><publisher>Lippincott Williams and Wilkins</publisher><urls></urls></record></Cite></EndNote>(Ward, 2004) discusses clinical and laboratory measures in RA, he notes that serial measures such as tender joint count and pain increase or decrease coincidentally at a patient level, demonstrating longitudinal construct validity. Notably, duration of morning stiffness does not follow these other measures, and this is likely why morning stiffness is not used as a measure of disease status, despite being a measure used in diagnosis. The ACR20 and the EULAR response criteria are both constructs of responder analyses from continuous based measures. One drawback of this type of assessment is that small changes in a patient’s disease can lead to an exaggerated change in status. For example, in a hypothetical case where an improvement of 50% is considered a response, a patient is classified as a responder with 51% improvement and a non-responder with 49% improvement, when in fact the patient’s disease state is essentially the same. Although the ACR20 is a measure of this type, it may be more robust to this phenomenon as it is dependent on 7 measures to determine response. The regulatory pathway for approval of novel therapies is typically achieved in 3 steps. First, a claim is made for improvement of signs and symptoms with 6 months data using ACR20. Secondly a claim for prevention of structural damage to joints with 1 year of radiological data (such as Sharp Score and Joint Space Narrowing). Finally, a claim for an improvement in physical function and disability can be pursued with 2 years of patient data (with HAQ-DI). This chapter has also described a typical Roche phase III clinical trial that is designed to meet the regulatory requirements, in the period that these trials were conducted, the study design was similar to rival competing pharmaceutical companies ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2017</Year><RecNum>187</RecNum><DisplayText>(FDA, 2017)</DisplayText><record><rec-number>187</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1506341386">187</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Briefing Document Arthritis Advisory Committee PLIVENSIA? (sirukumab)</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2017</year></dates><urls><related-urls><url>;(FDA, 2017).In this research, I have selected ACR20 to be used as the primary outcome of interest to define patient response. ACR20 values will be generated in simulations and biomarkers to predict ACR20 will be sought. Rheumatoid Arthritis is a multi-factor, systemic disease with a multitude of symptoms, and as described in Section 3.3.1 the ACR20 reduces the measurement of disease change to a dichotomous response/ no response result. I have chosen ACR20 since it is available from all phase III clinical trials in this research and is the earliest regulatory measure on which to assess efficacy of a medicine for RA. The purpose of this research is to understand how selection of patients might influence patient outcome. Chapter 2 described RA as a disease, and this chapter went on to describe endpoints for assessing disease status and progression as well as clinical trial designs employed in Roche RA development. The ACR20 outcome has usually been the primary endpoint for RA clinical trials; therefore, this research will attempt to predict ACR20 from patient data available at time of recruitment. The ACR20 is a composite endpoint that was developed to represent patient improvement in this multifactorial disease. If trial sponsors understand more about markers predictive of patients, then if may be possible to build this information into design of trials to: guide patient recruitment to those most likely to benefit, or stratify randomisation to optimise sample size estimates. Chapter 4 will describe the RA clinical data available for analysis in this thesis. The data will inform simulations and choice of statistical and data methods that will be applied. : Data DescriptionIntroductionChapter 3 described endpoints and trial designs in RA. This chapter summarises the data set to be used in this dissertation and the analysis strategy. The attributes of this data will direct which statistical and data mining analysis methods will be the most appropriate for analysis. In addition, the structure of the data and the relationships between variables will inform the parameters of the simulations that will be conducted in Chapter 7.4.2.AimsThe objectives of this chapter are to describe the data available to inform simulations that will be conducted in the next chapter. The chapter will describe:projects and studies; dependent and explanatory variables; missing data;observed correlations;potential limitations.4.3.Projects and Studies For this research, a rich and substantial pooled database was created to explore the methods for predicting patient response. These data exist in the form of SAS? datasets that were available from Roche Products Limited on 4 distinct development products for the treatment of RA. The data comprise of 11,705 adult patients diagnosed with RA from 16 Randomised Clinical Trials enrolled between 1998 and 2008 ( REF _Ref494121969 \h Table 4.5). These studies were chosen as they represent phase III registration studies for the adult RA indications of the 4 development products. The database includes data from patients with established RA and a variety of subpopulations including: patients refractory to anti-TNF treatment, patients refractory to other DMARDs and patients na?ve to biologic treatment for their disease. Assembling the database took over a year to complete; details of operational issues to compile the data are described in Section 4.4.4.3.1. Project AProject A concerned the clinical drug development of a recombinant humanized, anti-human monoclonal antibody of the immunoglobulin directed against the soluble and membrane bound interleukin 6 receptor (IL-6). IL-6 has been shown to be involved in such diverse physiological processes as T cell activation, induction of acute phase proteins, stimulation of hemopoietic precursor cell growth and differentiation, proliferation of hepatic, dermal and neural cells, bone metabolism, lipid metabolism and hepatoprotection ADDIN EN.CITE <EndNote><Cite><Author>Pasare</Author><Year>2004</Year><RecNum>189</RecNum><DisplayText>(Pasare & Medzhitov, 2004)</DisplayText><record><rec-number>189</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1507647844">189</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Pasare, C.</author><author>Medzhitov, R.</author></authors></contributors><titles><title>Toll-like receptors: linking innate and adaptive immunity</title><secondary-title>Microbes and Infection</secondary-title></titles><periodical><full-title>Microbes and Infection</full-title></periodical><pages>1382-7</pages><volume>6</volume><number>15</number><dates><year>2004</year></dates><urls></urls></record></Cite></EndNote>(Pasare & Medzhitov, 2004). Elevated tissue and serum levels of IL 6 have been implicated in the disease pathology of several inflammatory and autoimmune disorders including RA. Therefore, inhibition of the biological activity of IL 6 and/or its receptor represented a promising approach for the treatment of RA. Five pivotal Phase III studies from the Project A development program were included in this research. All studies were randomised, double blind parallel group designs.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 1 Summary of Project A studiesPopulation Inclusion CriteriaTreatment armsStudy A1Moderate to severe active RA with inadequate response to MTXSJC ≥ 6 andTJC ≥ 8 andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr2 active arms plus MTX vs placebo plus MTX Study A2Moderate to severe active RA with inadequate response to MTXSJC ≥ 6 andTJC ≥ 8 andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr2 active arms plus MTX vs placebo plus MTXStudy A3Moderate to severe active RA with no MTX treatment in last 6 monthsSJC ≥ 6 andTJC ≥ 8 andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr1 active arm plus MTX vs placebo plus MTXStudy A4Moderate to severe active RA with prior inadequate response to anti-TNF therapySJC ≥ 6 andTJC ≥ 8 andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr2 active arms plus MTX vs placebo plus MTXStudy A5Moderate to severe active RA with prior inadequate response to DMARD therapySJC ≥ 6 andTJC ≥ 8 andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr1 active arm plus MTX vs placebo plus MTXStudy A1 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had moderate to severe active RA and who had previously had an inadequate clinical response to Methotrexate (MTX) therapy. In this study 2 dose levels of active treatment in combination with MTX were compared to a placebo plus MTX arm. Patients who did not achieve a 20% improvement from baseline in both Swollen and Tender Joint Counts at week 16 could, if requested and deemed necessary by the investigator, receive escape therapy. A total of 623 patients were enrolled at 73 centres in 17 countries (Argentina, Australia, Austria, Brazil, Bulgaria, Canada, France, Germany, Hong Kong, Hungary, Israel, Italy, Mexico, Singapore, Slovakia, Switzerland and Thailand). Enrolment took place in 2005 and 2006. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (26% control response vs 48% and 59% in the active arms). In the primary analysis a pre-defined site based stratification was included in the model.Study A2 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had moderate to severe, active RA who had an inadequate response to MTX. In this study 2 dose levels of active treatment in combination with MTX were compared to a placebo plus MTX arm. Patients who did not achieve a 20% improvement from baseline in both Swollen and Tender Joint Counts at week 16 could, if requested and deemed necessary by the investigator, receive escape therapy. A total of 1196 patients were enrolled at 137 centres in 15 countries (Australia, Brazil, China, Denmark, Finland, France, Greece, Italy, Mexico, Norway, Poland, Switzerland, South Africa, Spain and the United States). Enrolment took place in 2005 and 2006. This study was designed to assess the reduction in signs and symptoms of RA after 24 weeks, prevention of joint damage at 52 weeks (evaluated by radiographs, with confirmation at 104 weeks), and physical function at 52 weeks (with confirmation at 104 weeks). After week 52, all remaining patients were switched to open-label active treatment for the second year, patients could opt for a 3rd year of open label treatment. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (27% control response vs 51% and 56% in active arms). In the primary analysis a pre-defined site based stratification was included in the model.Study A3 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and who had not been treated with MTX within 6 months prior to randomization, and who had not discontinued previous MTX treatment as a result of clinically important toxic effects or lack of response as determined by the investigator. In this study, one arm of active treatment in combination with MTX was compared to a placebo arm in combination with MTX. Patients who did not achieve a 20% improvement from baseline in both Swollen and Tender Joint Counts at week 16 could, if requested and deemed necessary by the investigator, receive escape therapy. In addition, active treatment was compared to a third arm of patients receiving placebo MTX. A total of 673 patients were enrolled at 120 centres in 18 countries (Argentina, Australia, Canada, China, Denmark, France, Israel, Italy, Lithuania, Mexico, Norway, Peru, Portugal, Serbia/Montenegro, Slovenia, South Africa, Spain and the United States). Enrolment took place in 2005 and 2006. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (52% control response vs 71% in active arm). In the primary analysis disease duration and a pre-defined site based stratifications were included in the model.Study A4 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had previously had an inadequate clinical response or were intolerant to treatment with one or more anti-TNF therapies within one year prior to randomization and who were receiving a stable dose of MTX. Patients who did not achieve a 20% improvement from baseline in both Swollen and Tender Joint Counts at week 16 could, if requested and deemed necessary by the investigator, receive escape therapy. In this study 2 dose levels of active treatment in combination with MTX were compared to a placebo plus MTX arm. A total of 499 patients were enrolled at 128 centres in 13 countries (Australia, Belgium, Canada, France, Germany, Great Britain, Iceland, Italy, Mexico, The Netherlands, Sweden, Switzerland, and the United States). Enrolment took place in 2005 and 2006. Statistically significant ACR20 responses at week 24 were observed in the active treatment arms over the control arm (10% control response vs 30% and 50% in active arms). In the primary analysis a pre-defined site based stratification was included in the model.Study A5 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had moderate to severe active RA who had an inadequate clinical response to current DMARD therapy. In this study 1 arm of active treatment in combination with MTX was compared to a placebo plus MTX arm. Patients who did not achieve a 20% improvement from baseline in both Swollen and Tender Joint Counts at week 16 could, if requested and deemed necessary by the investigator, receive escape therapy. A total of 1220 patients were enrolled at 130 centres in 18 countries (Argentina, Australia, Brazil, Canada, China, Costa Rica, Czech Republic, Finland, France, Germany, Mexico, Panama, Russia, Spain, Sweden, Thailand, South Africa and the United States). Enrolment took place in 2005 and 2006. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (25% control response vs 61% in active arm). In the primary analysis a pre-defined site based stratification was included in the model.The studies in Project A had a very similar design, with identical key inclusion criteria. There were some differences between studies with respect to previous experience of RA treatments. Other differences between the studies included choice of sites and countries, so that whilst the analysis of ACR20 included site based stratification, the grouping of sites was different for each study depending partly on enrolment in individual sites. 4.3.2. Project MProject M was a glycosylated immunoglobulin chimeric mouse/human antibody that binds to the CD20 antigen present on the majority of B cells. Expression of CD20 is restricted to the B-lymphocyte lineage from the pre–B cell stage until terminal differentiation into plasma cells. Treatment with this compound induces a rapid and sustained depletion of peripheral CD20+ B cells. Depletion of B cells is maintained from a few months to over 5 years. The pivotal phase III studies were conducted over several years and in coordination with a development partner. The impact of this is that there is substantial variability in study design, analysis strategy and programming consistency, making pooling of data for just this project non trivial (see Section 4.4 for further details). Five pivotal phase III studies from Project M development program were included in this research. All studies were randomised, double blind parallel group designs.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 2 Summary of Project M studiesPopulation Inclusion CriteriaTreatment armsStudy M1Active RA with inadequate response to MTXSJC ≥ 8 andTJC ≥ 8 andCRP ≥ 0.6 mg/dL or ESR ≥ 28 mm/hr2 active arms plus MTX vs placebo plus MTX Study M2Moderate to severe active RA and na?ve to MTXSJC ≥ 8 andTJC ≥ 8 andCRP ≥ 1 mg/dL 2 active arms plus MTX vs placebo plus MTXStudy M3Active RA with prior inadequate response to anti-TNF therapySJC ≥ 8 andTJC ≥ 8 andCRP ≥ 1.5 mg/dL or ESR ≥ 28 mm/hr1 active arm plus MTX vs placebo plus MTXStudy M4Severe active RA with prior inadequate response to DMARD and MTX therapySJC ≥ 8 andTJC ≥ 8 andCRP ≥ 1.5 mg/dL or ESR ≥ 28 mm/hr1 active arms, 1 active arm plus cyclophosphamide, 1 active arm plus MTX vs placebo plus MTXStudy M5Moderate to severe active RA with prior inadequate response to DMARD and MTX therapySJC ≥ 8 andTJC ≥ 8 and 2 of the following:CRP ≥ 1.5 mg/dL, ESR ≥ 30 mm/hr, Morning stiffness > 45 minutes1 active arms, 1 active arm plus cyclophosphamide, 1 active arm plus MTX vs placebo plus MTXStudy M1 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and had an inadequate clinical response to methotrexate (MTX) therapy. In this study 2 arms of active treatment in combination with MTX were compared to a placebo plus MTX arm. Between weeks 16 and 23, patients who had less than a 20% improvement in both tender and swollen joint counts compared to baseline were allowed to initiate “rescue” treatment with one non-biologic DMARD, the choice of which is at the discretion of their treating physician. These patients could receive further courses of active treatment if eligible, but were considered to be non-responders for all categorical efficacy endpoints at week 24. Patients received medication at day 1 and day 15 and then returned for assessments every 4 weeks until week 24, then every 8 weeks until week 48, and then every 12 weeks until B cells returned to baseline levels or to normal range. A total of 512 patients were enrolled at 102 centres in 11 countries (Canada, France, Germany, Great Britain, Guatemala, Mexico, Poland, Romania, Slovenia, Sweden, and the United States). Enrolment took place in 2005 and 2006. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (23% control response vs 55% and 51% in active arms). Analysis was stratified by RF status at baseline and region (US, Rest of World - ROW).Study M2 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and were na?ve to MTX therapy. In this study 2 arms of active treatment in combination with MTX were compared to a placebo plus MTX arm. Rescue treatment with one non biologic DMARD was available from week 52 for eligible patients. A total of 755 patients were enrolled at 169 centres in 27 countries (Australia, Belgium, Brazil, Canada, China, Czech Republic, Denmark, Finland, France, Germany, Great Britain, Guatemala, India, Italy, Mexico, Norway, Panama, Peru, Philippines, Poland, Romania, Russia, South Korea, Spain, Sweden, The Netherlands and The United States). Enrolment took place in 2006 and 2007. The primary endpoint for this study was not ACR20 at week 24 but rather the change in Sharp score at 52 weeks. The week 24 analysis was a pre-planned interim analysis. Patients were to receive study medication for a total of 3 years and followed until B Cells repleted. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (59% control response vs 73% and 75% in active arms).Study M3 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and had failed to respond or were intolerant to anti-TNF therapies. In this study 1 arm of active treatment in combination with MTX was compared to a placebo plus MTX arm. Patients who failed to respond to treatment during the study (i.e., achieved < 20% improvement in both swollen and tender joint counts) could receive rescue therapy from weeks 16 to 24. Patients received medication at day 1 and day 15 and then returned for assessments every 4 weeks until week 24. After week 24, patients entered the post-treatment period and were followed up every two months for 18 months, giving an overall study duration of 24 months. A total of 520 patients were enrolled at 114 centres in 11 countries (Belgium, Canada, France, Germany, Great Britain, Ireland, Israel, Italy, Norway, The Netherlands and the United States). Enrolment took place in 2003 and 2004. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (18% control response vs 51% in active arm). Analysis was stratified by RF status at baseline and region.Study M4 was a phase IIb, randomized, double-blind, double-dummy, controlled, multifactorial study of nine different treatment regimens in a 3 x 3 configuration, that comprised 2 different active doses plus placebo and three different corticosteroid regimens (including placebo), along with a weekly regimen of MTX. Patients were recruited who had RA, had failed 1-5 DMARDs, other than MTX, (for lack of efficacy) and who currently had an inadequate response to MTX. A rescue arm was available for eligible patients from weeks 16 to 24. Patients received medication at day 1 and day 15 and then returned for assessments every 4 weeks until week 24. After week 24, patients entered the post-treatment period and were followed up every two months for 18 months, giving an overall study duration of 24 months. A total of 465 patients were enrolled at 95 centres in 14 countries (Australia, Brazil, Canada, Czech Republic, Finland, Germany, Great Britain, Italy, Mexico, New Zealand, Poland, Spain, Sweden, and the United States). Enrolment took place in 2003 and 2004. Statistically significant ACR20 responses at week 24 were observed in the active treatment over control in this study (28% control response vs 55% and 54% in active arms). Analysis was stratified by corticosteroid use baseline and region.Study M5 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had RA, had failed 1-5 DMARDs and who had a partial clinical response to therapy with MTX monotherapy. Patients were assigned to 1 of 4 treatment arms: MTX, active treatment, active treatment plus cyclophosphamide, or active treatment plus MTX. Patients received medication at day 1 and day 15 and then returned for assessments every 4 weeks until week 24. Patients had a 4 week follow up assessment after week 24. A total of 161 patients were enrolled at 26 centres in 11 countries (Australia, Belgium, Canada, Czech Republic, Germany, Great Britain, Israel, Italy, Poland, Spain and The Netherlands). Enrolment took place in 2001 and 2002. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (38% control response vs 65%, 76% and 73% in active arms). Analysis was stratified by rheumatoid factor and region.The studies in Project M also had a very similar design and again differed in populations of patients studied, ranging from patients who were na?ve to MTX treatment through to patients who had poor response to not just MTX but to other biologic treatments. There was some variation in key inclusion criteria, particularly with respect to CRP and ESR, however joint count requirements were consistent across all studies. In addition in Project M the analysis of ACR20 included a more mixed choice of analysis stratification from study to study.?4.3.3. Project OProject O was a humanized anti-CD20 monoclonal antibody, and has the same method of action and had been shown to be at least as effective as the antibody in Project M in depleting B-cells. Following the results of the pivotal phase III studies, the indication for RA was not considered competitive and the project was not submitted for registration in RA. The pivotal phase III trials selected in this research were designed at the same time, although the enrolment dates differed slightly. This parallel development meant that the design of the protocols, analysis plans, programming were very consistent between studies, enabling very easy pooling. This project was the most recent development project of the four, and is the only project to incorporate the new Clinical Data Interchange Standards Consortium (CDISC) standards in data structure (see Section 4.4 for more detail). Project O comprised of 4 studies. All studies were randomised, double blind parallel group designs.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 3 Summary of Project O studiesPopulation Inclusion CriteriaTreatment armsStudy O1Active RA with inadequate response to MTXSJC ≥ 4 andTJC ≥ 4 andCRP ≥ 0.6 mg/dL andRF or CCP positive2 active arms plus MTX vs placebo plus MTX Study O2Moderate to severe active RA with inadequate response to anti-TNF therapySJC ≥ 4 andTJC ≥ 4 andCRP ≥ 0.6 mg/dL andRF or CCP positive2 active arms plus MTX vs placebo plus MTXStudy O3Active RA with inadequate response to MTXSJC ≥ 4 andTJC ≥ 4 andCRP ≥ 0.6 mg/dL or ESR ≥ 28 mm/hr andRF or CCP positive2 active arms plus MTX vs placebo plus MTXStudy O4Moderate to severe active RA and na?ve to MTXSJC ≥ 8 andTJC ≥ 8 andCRP ≥ 1 mg/dL andRF or CCP positive2 active arms plus MTX vs placebo plus MTXStudy O1 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and who had previously had an inadequate clinical response to MTX therapy. In this study 2 arms of active treatment in combination with MTX were compared to a placebo plus MTX arm. Patients received double blind medication up to week 48 and then were eligible for open label treatment for a further 2 years or until B-cells returned to baseline or normal levels. A total of 1015 patients were enrolled at 209 centres in 24 countries (Argentina, Australia, Austria, Belgium, Brazil, Canada, China, France, Germany, Great Britain, Greece, Guatemala, Israel, Mexico, New Zealand, Panama, Peru, Russia, South Korea, Spain, Taiwan, Thailand, Ukraine and the United States). Enrolment took place between 2006 and 2008. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (36% control response vs 57% and 55% in active arms). Analysis was stratified by RF status and region.Study O2 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and who had previously had an inadequate clinical response to an anti-TNF-α therapy. In this study 2 arms of active treatment in combination with MTX were compared to a placebo plus MTX arm. Patients received double blind medication up to week 48 and then were eligible for open label treatment for a further 2 years or until B-cells returned to baseline or normal levels. A total of 840 patients were enrolled at 227 centres in 25 countries (Argentina, Australia, Belgium, Brazil, Canada, Czech Republic, France, Germany, Hungary, Israel, Italy, Japan, Mexico, New Zealand, Panama, Peru, Poland, Slovakia, Slovenia, Spain, Sweden, Switzerland, Taiwan, The Netherlands, and the United States). Enrolment took place in 2007 and 2008. Statistically significant ACR20 responses at week 24 were observed in active treatment over control in this study (22% control response vs 42% and 48% in active arms). Analysis was stratified by baseline DMARD therapy and region.Study O3 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had active RA and who had an inadequate clinical response to at least 12 weeks of prior MTX therapy (with or without previous exposure to biologic DMARDs). In this study 2 arms of active treatment in combination with MTX were compared to a placebo plus MTX arm. Patients received double blind medication up to week 48 and then were eligible for open label treatment for a further 2 years or until B-cells returned to baseline or normal levels. A total of 314 patients were enrolled at 96 centres in 14 countries (Australia, Canada, France, Germany, Great Britain, Italy, Mexico, Poland, Romania, South Korea, Spain, Switzerland, Thailand and the United States). Enrolment took place in 2008. ACR20 responses at week 24 were not statistically significant (28% control response vs 38% and 53% in active arms). Analysis was stratified by baseline DMARD therapy and region. Study O4 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had moderate to severe RA and who were na?ve to MTX. In this study 2 arms of active treatment in combination with MTX were compared to a placebo plus MTX arm. Patients received double blind medication up to week 104 and then were eligible for open label treatment for a further 1 year or until B-cells returned to baseline or normal levels. A total of 613 patients were enrolled at 154 centres in 21 countries (Argentina, Australia, Brazil, Israel, Italy, Lithuania, Mexico, New Zealand, Panama, Peru, Philippines, Poland, Russia, South Korea, Spain, South Africa, Sweden, Switzerland, Taiwan, Thailand, and the United States). Enrolment took place in 2007 and 2008. Statistically significant ACR20 responses at week 52 were observed in active treatment over control in this study (57% control response vs 73% and 71% in active arms). Analysis was stratified by CRP status and region.The studies in Project O were perhaps more homogenous they had a very similar design, differed a little in populations of patients studied and included the same treatment arm regimens. There were more differences here in inclusion criteria compared with Project A and M, as well as variation within this project for the joint count criterion. Nonetheless the study reports still considered this population to be moderately to severe in severity. In the analysis of ACR20 stratification always included region, but differed slightly in second stratification factor.4.3.4. Project TProject T was a potent and selective inhibitor of collagenase, thus preventing cleavage of triple helical collagen fibres. It is orally active in models of cartilage degradation and preserves joint integrity in animal models of arthritis. Since the results of the studies were not positive in the radiographic endpoints, the project was not submitted for registration. Project T is the oldest of the 4 projects under consideration and presented challenges in terms of versions of SAS, and converting legacy datasets to newer versions (see Section 4.4 for more details). In addition, some endpoints such as ACR20 were not standard at the time of development planning and were therefore calculated post hoc from the available data for the purposes of this research, the primary endpoint in these studies was based on radiographic assessment.Project T comprised of 2 studies. Both studies were randomised, double blind parallel group designs.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 4 Summary of Project T studiesPopulation Inclusion CriteriaTreatment armsStudy T1Established RA SJC ≥ 4 and≥ 1 joint with erosion andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr4 active arms plus MTX vs placebo plus MTX Study T2Established RASJC ≥ 4 and≥ 1 joint with erosion andCRP ≥ 1 mg/dL or ESR ≥ 28 mm/hr4 active arms plus MTX vs placebo plus MTXStudy T1 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had established RA. Patients with RA were randomized to receive one of four doses of active treatment or placebo once daily for 104 weeks. All patients received background MTX therapy. Patients were stratified by low-dose oral corticosteroid use. A total of 1145 patients were enrolled at 98 centres in 20 countries (Australia, Austria, Belgium, Brazil, Canada, Canary Islands, Chile, Finland, France, Germany, Great Britain, Ireland, Italy, Mexico, New Zealand, Portugal, South Africa, Spain, Switzerland and The Netherlands). Enrolment took place between 1998 and 2000. ACR20 responses at week 24 were observed in active treatment over control in this study (26% control response vs 29%, 31%, 22% and 24% in active arms). The study failed to meet its primary radiographic endpoint (Sharp Score). Study T2 was a phase III, randomized, double-blind, placebo-controlled, parallel group study. Patients were recruited who had established RA. Patients with RA were randomized to receive one of four doses of active treatment or placebo once daily for 104 weeks. All patients received background MTX therapy. Patients were stratified by low-dose oral corticosteroid use. A total of 1153 patients were enrolled at 104 centres in 6 countries (Brazil, Canada, Costa Rica, Mexico, South Africa, and the United States). Enrolment took place between 1998 and 2000. ACR20 responses were not analysed in this study but were calculated post hoc in this thesis. The study failed to meet its primary radiographic endpoint (Sharp Score).Project T includes 2 identically designed studies in different regions. Despite the fact that the trials were not registrational due to the results, they nonetheless include useful data from over 2000 patients with RA. Some of these study characteristics are displayed and summarised in REF _Ref494121969 \h Table 4.5.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 5 Summary of RA database available for analysisSourceStart of Enrol-mentVersion of SASMTX Control patientsPlacebo PatientsActive Treatment Patients*TotalStudy A120058.2204419623Study A220058.23948021196Study A320058.2284101289674Study A420058.2161338499Study A520058.24158051220Total Project A145810126534212Study M120058.2172340512Study M220068.2251504755Study M320038.2209311520Study M420038.2148317465Study M520018.240121161Total Project M82015932413Study O120069.13246911015Study O220079.1277563840Study O320089.164250314Study O420079.1210403613Total Project O87519072782Study T119986.122209251145Study T219986.122299241153Total Project T44918492298?Overall3602101800211705* Patients on investigational drugEach study design follows patients for at least 24 weeks, and includes patient disposition, demographic information, medical history, concomitant medication, and efficacy measurements as described in Section 3.3. There are also safety laboratory variables from haematology, biochemistry, and RA specific markers (including Rheumatoid Factor). REF _Ref512261492 \h Table 4.6 displays the ACR20 response rates by study. In the MTX control group the ACR20 response ranges from 10.1% to 60.4%. The rates in the lower responding studies, A4, M3 and O2; are associated with study populations that had inadequate clinical response to an anti-TNF-α therapy. The rates in the higher responding studies, were A3, M2 and O4; are associated with populations that were na?ve to MTX or had at least a 6-month treatment break from MTX at the time of enrolment. Whilst this heterogeneity is noted, the dependent variables of ‘popir’, and ‘popra’ in the clinical trial dataset incorporate this information and are available for the analyses in Chapter 8.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 6 Summary of ACR20 response rates by study SourceMTX Control Placebo Active Treatment(%)(%)1 (%)2 (%)3 (%)4 (%)Study A126.547.958.5Study A226.950.656.2Study A352.543.669.9Study A410.130.450.0Study A524.560.8Study M123.354.550.6Study M258.072.575.0Study M317.951.3Study M427.955.354.1Study M532.541.442.5Study O135.756.954.5Study O222.042.247.9Study O328.137.652.7Study O460.468.377.6Study T127.629.931.821.425.0Study T222.622.516.222.221.34.4.Operational challenges in database creationAs discussed earlier in this chapter, the strength of this data is that the data comprise of adult patients diagnosed with RA from 16 Randomised Clinical Trials enrolled between 1998 and 2008 representing registration studies for the adult RA indications of 4 development products. This presented however a number of practical challenges that took over one year to program and collate, the details of these challenges are described in this section. The first general challenge for each of the studies was to identify the most relevant datasets to include. For example, study M2 in Project M was a study that as a primary endpoint evaluated radiographic endpoints at 52 weeks. In this research we are interested in the ACR20 response at 24 weeks; this was a planned interim analysis in study M2. For this study a database snapshot was created for the week 24 analysis and for the week 52 analysis to enable regulatory submission to the USA and Europe. Additional snapshots of the data were created for safety updates as required by regulatory process as well snapshots for later publication purposes at conferences and in journals. The decision to be made therefore is which snapshot to use for the week 24 ACR20 analyses. Although one might expect this to be the same, ongoing data management cleaning activities have the potential to impact already reported endpoints. So for each of the studies a careful review of study and project documentation was performed and the input of the original programmer and statistician were sought if they still worked for the company. In general, I decided to use the snapshot used for the original registration submission in order to be consistent with the published study reports as much as possible.Fortunately, during the 11-year span in study initiation the computer environment was consistent (UNIX environment on a HP Integrity rx7640 server) however many other aspects of the studies and projects varied. The version of SAS ranged from version 6.12 to version 9.1, and with these multiple versions of SAS, the formats of SAS datasets also varied ( REF _Ref494121969 \h \* MERGEFORMAT Table 4.5). In order to bring some consistency to the datasets, each study dataset was updated to the current working version on the UNIX server at the time of this research (version 9.2). Conventions in data collection and regulatory requirements changed over the 11 years spanned by the studies. For example, the earlier studies recorded race in greater granularity (Oriental, Chinese, Bangladesh, Philippino, Filipino, Cuban, Maori) these needed to be mapped in a manner consistent with later trials (Asian, Hispanic, and Pacific Islander).Similarly, some studies collected country information whilst others summarised by region, and regions were not always consistent. For example, some studies categorised region as “US, Europe and Other”, or “North America, Non North America” or by groupings of countries based on patient recruitment (sometimes Canada was “North America” and sometimes “Rest of World” depending on study). In studies, such as M4, a difference of ACR20 response in both placebo and active treatment appeared to be lower in USA and Canada. The most practicable way to divide region according to the available data was as “North America” and “Non North America”.Laboratory results were captured inconsistently between studies, and required substantial data manipulation. Laboratory results were recorded sometimes as numeric values and other times as character values. When collected as character, the fields contained such contents as “No Valid Result”, “NK”, “ND”, “Neg”, or “Trace”. All laboratory values were converted to numeric to allow computation, with the fields above converted to missing or 0 as appropriate. Laboratory values were not consistently collected in the same units. Conversions to consistent units were required. For example, Glucose was reported as both mg/dL and as mmol/L. For this database, Glucose was reported as mmol/L, so mg/dL reported values were converted by dividing the reported value by the conventional and accepted conversion factor of 18.016. In some studies, RF was captured as a numeric variable, in other studies RF was captured as a categorical variable either as Negative/Positive or 0/1. Medical History was recorded differently between studies. This was derived from different fields of the case report form. This was variously captured as Negative/Positive, 0/1 or (blank)/Yes. For consistency all medical history results were converted to a 0/1 format.For earlier studies DAS28, ACRn was not databased, so needed to be calculated from component values. Similarly, ACR70, ACR90 were calculated post hoc for some earlier studies. The nomenclature for ACR20 changed between studies, and care needed to be taken to be satisfied that ACR20 was calculated consistently. In some studies, ACR20 was based on ESR, others were based on CRP. There were differing ACR20 calculation conventions depending on how the individual components were imputed when values were missing. Since ACR20 was most commonly calculated using CRP, this was used whenever possible.Individual component percentage improvements that were reported in section 8.6.2, were not retained from ACR20 calculations in most studies. These values were calculated specifically for this thesis. Over the time period that these trials were conducted, many teams and individuals were involved in the clinical development programs. During that period versions of SAS changed, innovations in programming emerged and opinions changed on best data structure. In addition, depending on the needs of the project needs different data types and the decisions to accommodate them were made. Their decisions were not always consistent with previous RA drug development programs, so that for example, the same variable from trial to trial may be labelled differently, if not from trial to trial then almost certainly from project to project. In the course of this research, identifying and mapping equivalent variables from study to study proved a very time consuming process. In 1997 the Clinical Data Interchange Standards Consortium (CDISC) was created to create standards for data content and structure ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2016</Year><RecNum>188</RecNum><DisplayText>(FDA, 2016)</DisplayText><record><rec-number>188</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1506369120">188</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Study Data Standards: What you need to know</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2016</year></dates><urls><related-urls><url>;(FDA, 2016). The most recent studies in this research embraced these new standards, which were quite different in structure to earlier studies. In September 2017 the FDA required that Sponsors, whose studies started after 17th December 2016, must submit data in the data formats supported by FDA and listed in the FDA Data Standards Catalogue. In undertaking a pooled analysis of this type and scale, the amount of data manipulation should not be underestimated. Whilst it was expected that some changes in SAS and some differences in what was collected would be observed, the subtle changes in ACR20 calculation and different database conventions led to substantial data manipulation. In addition, the nature of clinical development of a chronic disease with multiple reporting events created an unexpected issue in choosing and finding the most appropriate data source.With the adoption of CDISC by FDA and other regulatory agencies, combining data from future clinical trials should become easier. Nonetheless care should be taken to ensure subtle changes in definitions of variables are recognised.Dependent and Explanatory Variables REF _Ref494122044 \h \* MERGEFORMAT Table 4.7 describes the types of the variables available for analysis. As can be seen there is a wealth of data in terms of patients, dependent variables and independent variables. The data comprise 151 variables, of which 7 are response variables measured at week 24 and 144 are independent variables measured at baseline. With such a large amount of data there is sufficient scope to apply sophisticated methods described in later chapters.All data, except for week 24 endpoints are baseline assessments recorded on or shortly after enrolment into the clinical trials. The patient disposition / demography domain captures RA independent patient characteristics such as age, gender, race, smoking history and histories of any other disease. The patient disposition / demography domain also captures RA specific information at enrolment into the trials, such as duration and severity of RA, and previous use of other agents to treat RA.The baseline disease status domain includes variables relating to assessment of RA. This domain includes the baseline individual components of the ACR20, Rheumatoid Factor (RF) status, DAS28, SF36 component and summary measures as well as baseline radiographic assessments.In addition, baseline safety and laboratory data are included. These domains represent vital signs variables as well as results from baseline haematology and biochemistry blood samples and urinalysis.Within these domains there is a mixture of data type. There are continuous data such as age, weight, duration of disease, blood pressure. Some of the continuous variables are distributed approximately Normally, e.g. HAQ, pain score, blood pressure, Haemoglobin. Other continuous variables have skewed distributions e.g. joint counts, RA duration, SF36 mental component score. Histograms, boxplots and Normal Q-Q plots for continuous variables are shown in Appendix B.2. The clinical trial data also includes dichotomous data such as gender, steroid use, and occurrence of previous diseases such as cardiac, gastrointestinal and other disorders. Further there are categorical measures relating to RA history, RA severity, SF36 questions, and urinalysis. Appendix B.1 details a complete list of endpoints available in the data. Appendix B.2 includes various plots of the distributions of these variables.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 7 Variables available for analysis in RA DatabaseDomainContinuousDichotomousCategoricalPatient Disposition/ DemographyAge*, Height*, Weight, Years of Smoking.Duration of disease. Gender, Region, Steroid Use History.Other Disease History: Blood & Lymphatic, Cardiac, CNS, Ear Disorders, Eye Disorders, Gastrointestinal Disorders, Genetic and Familial Disorders, Hepatobiliary Disorders, Immune Disorders, Infections, Injuries, Metabolic Disorders, Musculoskeletal Disorders, Neoplasms, Pregnancy and Lactation Disorders, Psychiatric Disorders, Renal Disorders, Reproductive Disorders, Respiratory Disorders, Skin Disorders, Social Disorders, Surgery, Vascular Disorders.Project, Protocol, RA Population, History of Inadequate Response, Treatment Assigned, Race, Smoking History, Previous use of DMARDs, Year of Enrolment.* Normally distributedTable 4.7 Variables available for analysis in RA Database (continued)DomainContinuousDichotomousCategoricalBaseline Disease statusACR Components: Tender joint assessments*, Swollen joint assessments, ESR, CRP, HAQ*, Pain Score*, Patient Global Assessment*, Physician Global Assessment*.DAS28* Short Form36 Summary Scores: Mental Component, Physical Component, Bodily Pain*, General Health*, Mental Health*, Physical Functioning*, Role Limitation Emotional, Role Limitation Physical, Social Functioning, Vitality*.Radiographic measures: Erosion Score, Joint space narrowing, Sharp Score.RF Status.Short Form36 36 Individual questions. Baseline Safety EndpointsVital Signs: Diastolic* and Systolic* Blood Pressure, Heart Rate*.Baseline Laboratory dataHaematology, Blood Chemistry.Urine.Efficacy Endpoints (Week 24)ACRn, DAS28.ACR20, ACR50, ACR70, ACR90.* Normally distributed4.6.Response & Demography and Disease History In order to rationalise the number of treatment groups across all projects and studies, drug type categorises treatment into 3 groups: Placebo, Methotrexate control arm (MTX) and experimental treatment (Active). ACR20 averaged a 50% response in the Active group compared with 32% in MTX and 44% in Placebo. The Placebo data was generated by Study A3 in patients with mild RA. With the exception of smoking years where the placebo arm had no data, demographic variables were balanced across treatment ( REF _Ref454297283 \h \* MERGEFORMAT Table 4.8). ACR20 response across categorical demographic factors such as gender also showed consistent results. Despite a difference in numbers of females vs males, response in ACR20 was similar, 48.8% vs 47.2% respectively in the active treatment arm and 31.4% vs 31.7% for the MTX group ( REF _Ref490048758 \h \* MERGEFORMAT Figure 4.1).Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 8 Summary of ACR20 and Demographic DataVariableTreat-mentn/N*MeanMini-mumMaxi-mumMedianSDACR20Placebo101/1010.440100.50MTX3515/36020.320100.47Active7774/80020.500100.50Age (years)Placebo101/10151.321845112.9MTX3602/360252.118855312.5Active8002/800252.518905312.5Duration of RA (years)Placebo101/1017.620.1641.683.279.43MTX3601/36027.500.0153.305.037.81Active7999/80027.370.0061.975.087.61Smoking history (years)Placebo101/1010.00000.000.00MTX2684/36021.200600.005.16Active5256/80021.330600.005.38Height (cm)Placebo101/101165.01501881649.12MTX3581/3602163.31262001639.41Active7950/8002163.31302001639.37Weight (kg) Placebo101/10178.8749.9131.874.820.40MTX3582/360274.4334.5186.071.519.11Active7952/800274.1328.0189.571.018.78Number of previous DMARDs (n)Placebo101/1011.30711.5MTX3153/36021.401011.6Active6153/80021.501111.6 * n/N – data-points evaluable / total number of patients SD = standard deviationFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 1 ACR20 Response by Treatment Type and Gender4.7.Available and Missing Data REF _Ref490048788 \h \* MERGEFORMAT Figure 4.2 gives a holistic view of those variables with missing data according to domain type (i.e. X-ray, SF36, Demography, Laboratory, Endpoint, ACR, Vital Signs). REF _Ref490048799 \h Figure 4.3 and REF _Ref490048806 \h Figure 4.4 express the detail of the missing data. Depending on the variable, the amount of missing data ranges from 0% to 93% (Unmodified Sharp Score). Some of the variables had missing data by design, for example those variables with more than 50% missing data related mostly to X-ray data which were not collected in every study ( REF _Ref490048799 \h \* MERGEFORMAT Figure 4.3). Scott ADDIN EN.CITE <EndNote><Cite><Author>Scott</Author><Year>2000</Year><RecNum>123</RecNum><DisplayText>(Scott, 2000)</DisplayText><record><rec-number>123</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">123</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Scott, D.L.</author></authors></contributors><titles><title>Prognostic factors in early rheumatoid arthritis</title><secondary-title>Rheumatology (Oxford England)</secondary-title></titles><pages>6</pages><volume>39(suppl. 1)</volume><section>24</section><dates><year>2000</year></dates><urls></urls></record></Cite></EndNote>(Scott, 2000) identified that radiographic damage at baseline as one of three important predictive factors for RA progression. RF score was not always collected as a numeric value in all studies, sometimes merely recorded as positive or negative. Of the 26 variables with 10% to 50% missing data, 11 concerned SF36, and 8 for laboratory measures. The remaining were baseline (n=5), X-ray (n=1) and endpoint (n=1) ( REF _Ref490048806 \h \* MERGEFORMAT Figure 4.4). SF36 was a tertiary endpoint in each trial, as well as being a patient recorded outcome, therefore the ability to query missing data during the study conduct would have been more limited. Conversely ACR outcomes were primary endpoints and for ACR20 patients who withdrew early or moved on to escape therapy were defined as non-responders, therefore missing values were much fewer. Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 2 Missing Data across all variablesBlue = X-ray, Green = SF36, Red = Demography, Black = Laboratory, Yellow = Endpoint, Orange = ACR, Purple = Vital Signs Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 3 Variables with more than 50% missing dataBlue = X-ray, Red = DemographyFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 4 Variables with 10% to 50% missing dataBlue = X-ray, Green = SF36, Red = Demography, Black = Laboratory, Yellow = Endpoint,4.8.Response DataThe amount of missing data varies by study and data domain. The calculation of ACR20, ACR50, ACR70 and ACR90 is missing for 2.7% of the patients, reducing the effective size of the analysis from 11,705 to 11,390 patients ( REF _Ref490048858 \h \* MERGEFORMAT Figure 4.5). A higher amount of missing data is observed for ACRn (15.7%), the higher amount is due to imputation rules for withdrawals not applicable to ACRn (for definition of ACRn, see section 3.3.1). Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 5 Percentage of patients with an ACR Response assessment4.9.Demography and Disease History DataFor most of the common demographic characteristics such as gender, age, region, race, RF status, medical history, weight, and height, the amount of missing data is very low (< 1%). The exceptions are smoking history and steroid use at baseline, numeric value for RF at baseline, and number of previous DMARDs which were not collected in all studies (missing rate 29% to 64%)( REF _Ref490048896 \h \* MERGEFORMAT Figure 4.6). Projects A and T did not collect RF at baseline for any study. All patients had disease history data. Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 6 Percentage of patients with Demographic dataRF = rheumatoid factor4.10. Baseline ACR components DataBaseline ACR component data is available on all variables for more than 99% of the patients ( REF _Ref490048937 \h \* MERGEFORMAT Figure 4.7). Project T did not collect Baseline DAS28 data.Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 7 Percentage of patients with baseline ACR component data4.11.Vital Signs and Radiography DataThe proportion of missing vital sign data is very low at <1%. Radiographic data were only collected in 50% of the studies and up to 56% of the patients, depending on the variable. Sharp score, Joint Space Narrowing and Erosion score were the most consistently collected variables across studies that assessed X-Rays ( REF _Ref490048973 \h \* MERGEFORMAT Figure 4.8 and REF _Ref459193243 \h \* MERGEFORMAT Figure 4.9). Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 8 Percentage of patients with Vital Sign and X-ray dataFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 9 Percentage of patients with Vital Sign and X-ray data by project4.12.Laboratory DataThe laboratory data comprises 30 variables, the rate of missing data ranges from 1% to 31%. Whilst many of the variables were collected consistently, 6 (Calcium, Cholesterol, LDH, Phosphate, RF, Uric Acid) were not collected at all in Project T - see REF _Ref454292553 \h \* MERGEFORMAT Figure 4.10 and REF _Ref454292561 \h \* MERGEFORMAT Figure 4.14. RF and Glycosuria were not collected in Project O - see REF _Ref454292560 \h \* MERGEFORMAT Figure 4.13.Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 10 Percentage of patients with laboratory dataFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 11 Percentage of patients with laboratory data in Project AFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 12 Percentage of patients with laboratory data in Project MFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 13 Percentage of patients with laboratory data in Project OFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 14 Percentage of patients with laboratory data in Project T4.13.SF36 DataThe proportion of missing SF36 individual component scores ranged from 3.0% to 11.7% ( REF _Ref454293125 \h \* MERGEFORMAT Figure 4.15). Similarly, 2.8% to 8.2% of SF36 summary score values were missing ( REF _Ref454293219 \h \* MERGEFORMAT Figure 4.16).Figure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 15 Percentage of patients with SF36 Component dataSee Appendix B.2 for SF36 component labelsFigure STYLEREF 1 \s 4. SEQ Figure \* ARABIC \s 1 16 Percentage of patients with SF36 Component scores4.14.Correlations and Associations with ACR20 REF _Ref456554817 \h \* MERGEFORMAT Table 4.9 shows the correlations between ordinal categorical variables and ACR20 by project and pooled data. Spearman’s rank statistics are presented due to ordinal nature of the variables and for consistency. The statistics give a broad indication of relationship with ACR20. The presence of steroid use at baseline and a positive RF have a small relationship with ACR20 response. Active and to some degree MTX control are more associated with response with the exception of Project T. REF _Ref458136066 \h \* MERGEFORMAT Table 4.10 shows the associations between ACR20 and categorical variables by project and pooled data. For most of the categorical variables the general associations were higher and null hypotheses could be rejected with very low p-values (<0.0001). This is mostly replicated for projects A, M and O. Notably Project T is the earliest project and least effective medicine which may account for this heterogeneity. Drug Type (Active treatment, MTX Control or Placebo) has the highest and consistent correlation amongst categorical variables.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 9 Correlations between ordinal categorical variables and ACR20 Variable Spearman’s Rank Correlation (p-value)ProjectAMOTTotal**Steroid Use at Baseline (order: No, Yes)0.0363 (0.0186)0.0316 (0.2350)0 na0.0405 (0.0643)0.0265 (0.0200)Drug Type (order: Active, MTX control, Placebo)-0.2405 (<.0001)-0.2345 (<.0001)-0.1691 (<.0001)0.0115 (0.5987)-0.1605 (<.0001)RF Category (order: Negative, Positive)0.0652 (<.0001)0.0584 (0.0047)0.0255 (0.1807)na*0.0125 (0.1825)* single category, **na?ve pooled, p-value = Cochrane-Mantel-HaenszelTable STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 10 Associations between non ordinal categorical variables and ACR20 response rates Variable ACR20 Response rate (p-value)ProjectAMOTTotal**Compound0.46290.53370.49110.24080.4436(<.0001)RA Population(<.0001) na*( <.0001) na*(<.0001) Active RA0.58610.53370.43640.24080.4255 Moderate / Severe RA0.4393-0.6860-0.4754IR Population(<.0001)(<.0001)( <.0001) na*(<.0001) DMARD IR0.48440.5589na-0.5081 MTX IR0.44590.42630.4757-0.4540 TNF IR0.58610.69250.6860-0.3581 No IR0.30520.38270.37440.24080.4450Race(<.0001)(<.0001)(<.0001)(0.8319)(<.0001) American Indian0.58800.27780.6237-0.5823 Asian0.44580.52330.47540.23990.3968 Black0.45000.44230.34250.21980.3785 Hispanic0.46880.5349--0.5169 Multiple0.50430.7135--0.5995 Pacific Islander1.00000.66670.6250-0.6923 Other1.0000.42860.59550.26290.4532 White0.44990.50000.48870.16670.4649Region(<.0001)(<.0001)(<.0001)(<.0001)(<.0001) Europe0.4847-0.56360.29170.4393 North America0.37500.41670.40530.20730.3606 South America0. 5762-0.6723-0.6049 Rest of World0. 47420.64090.52300.20050.5118 Non North America-0.5191-0.5191* single category,**na?ve pooled, IR=Inadequate response, p-value = Cochrane-Mantel-HaenszelTable 4.10 (continued) Associations between non ordinal categorical variables and ACR20 response ratesVariable ACR20 Response rate (p-value)ProjectAMOTTotal**Gender(0.9853)(0.4565) (0.2775)(0.5780)(0.5343) Female0.46280.53000.48590.23750.4451 Male0.46320.54950.51160.24910.4380Smoking History(0.0060)(<.0001) (0.0011) na(<.0001) Current Smoker0.41790.48020.4394-0.4349 Never Smoked0.47280.63210.5212-0.5072 Past Smoker-0.53490.4638-0.4846* single category,**na?ve pooled, p-value = Cochrane-Mantel-HaenszelCorrelations between numeric variables and ACR20 were in general quite low. The highest magnitude in the pooled analysis observed was 0.18566 ( REF _Ref456555601 \h \* MERGEFORMAT Table 4.11). Again Project T seems to behave differently to the other 3 projects. There appears to be some consistency between Projects A, M and O for Psychiatric disorder medical history and possibly number of previous DMARDs, RA Duration, and Vitality. Also of note is that when they are measured, the radiographic assessments at baseline have the highest correlations with ACR20 with the highest magnitude at 0.23379. Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 11 Correlations between continuous variables and ACR20VariablePearson Correlation CoefficientProjectAMOTTotal*Number of DMARDs-0.1210-0.0242-0.1377na-0.1197Year of Enrolmentna0.1167-0.0272na0.1857Psychiatric History-0.1100-0.114-0.1010-0.0014-0.0702GI History-0.0628-0.1090na-0.0624-0.0105Erosionna-0.2296-0.1103na-0.0894Joint Space Narrowingna-0.2273-0.1137na-0.1157Sharp Genantna-0.2338-0.1152na-0.1084RA Duration-0.0533-0.1245-0.1258-0.0665-0.0546Calcium-0.0087-0.1589-0.0019na-0.0589SF36 Mental Component0.3203-0.1225-0.0027-0.0641-0.1119SF36 Physical Component0.0858-0.10770.0869-0.0481-0.0845SF36 Vitality0.09940.12200.0819-0.02770.0160SF36 General Health0.06160.11630.0329-0.0649-0.0097SF36 Mental Health0.02580.1034-0.0063-0.0570-0.0428SF36 Feel full of pep-0.1032-0.0927-0.0592-0.0372-0.0545SF36 Lot of Energy-0.0762-0.1221-0.0655-0.0386-0.0308Baseline TJC-0.0358-0.0020-0.01840.13120.0868Baseline Pain VAS-0.01310.02380.00710.10580.0732Baseline Patient Global Score-0.01160.05240.02480.11830.0942Baseline Physician Global Score-0.01810.08430.00580.09100.1098*na?ve pooled, greater magnitude correlations shaded4.15.LimitationsWhilst all studies recruited adult patients with RA there are some potential sources of heterogeneity between the studies. As mentioned above patients were recruited over a period from 1998 to 2008, during this period the treatment of RA has evolved, as observed by changes in methotrexate dosing. Patients have varying response to other treatments for RA such as inadequate response to methotrexate, anti-TNF therapies and other DMARDs. The data also represents patients from 50 countries which introduces possible heterogeneity through factors such as diet and standard of medical care, see study descriptions in Section 4.3 for further details.As seen in REF _Ref490048799 \h \* MERGEFORMAT Figure 4.3 some variables have a high amount of missing values overall. Joint space narrowing of hands and of feet for example were only collected in Project T therefore it would be prudent to consider omitting these variables from the overall analysis and include them only when analysing the data at a project level.Some of the radiographic variables when considered by project have higher correlations than other variables. By design, not all studies performed X-rays to capture these data. When performing the analysis later in this research it would be prudent to consider an analysis subset on those studies that were designed to capture radiographic data.4.16.Criteria for method selectionFurther in this thesis, a method will be applied to the data described in this chapter to predict patient outcome. The features of the chosen methodology will in a large part need to be compatible with the characteristics of the data.4.16.1. Mixed DataThe data described in this chapter features explanatory variables of multiple types REF _Ref497235697 \h Table 4.7). Whilst the outcome variable of ACR20 is dichotomous, there are also dichotomous variables such as gender, steroid use, disease history (yes, no), and RF status (positive, negative). The data features categorical variables such as race, treatment assigned, smoking history (never, ex-smoker, current), patient population, and some SF36 scale data. Finally, the dataset comprises data that is continuous or can be treated as continuous such as vital sign and laboratory measurements, joint counts, age, height and weight. A method will be chosen that is able to manage well multiple variable types in order to make best use of the data.4.16.2. Missing ValuesAs discussed in Section 4.7 missing values are an important feature of this data. The amount of missing values ranges from 0% to 93% (Unmodified Sharp Score). Some of the missing data is by design, i.e. where data on a particular variable was not collected in all studies. Sometimes there is a higher level of missing data on a variable due to it not being of primary nor secondary importance. When considering all explanatory variables available in this data, every patient has at least one missing data point, therefore it is important that any method applied does not delete patients with missing data from the analysis. 4.16.3. Outliers Many of the continuous explanatory variables are subject to extreme values. For example, years of smoking, RA duration and Baseline CRP, display very skewed distributions (see Appendix B.2 for histograms and boxplots). 4.16.4. ScalabilityThis data contains data from over 11,000 patients and 151 variables. A viable method needs to be able to have computational scalability to manage a dataset of this size.4.16.5. High DimensionsSection 4.5 informed that the data comprise 151 variables, of which 7 are response variables measured at week 24 and 144 are independent variables measured at baseline. A data point of the 144 independent variables can be considered as residing in 144dimensional space. Any method therefore needs to be able to handle high dimensional data well.4.16.6. Interpretability Results from this research are intended to inform future clinical trial design and drug development. Decisions are made by multidisciplinary study teams and clinical development teams of which the statistician is usually one member of many. Although perhaps not an essential requirement, interpretability of findings for a non-statistician is a preferred criterion. 4.17.SummaryThis chapter has described a large database of randomised clinical trials in Adult RA patients. With 11,705 patients, to the best of my knowledge this is larger than any of the research described in the systematic review (Chapter 6). This data comprises a variety of drug projects and a substantial number of individual studies. Within these randomised clinical trials is a selection of RA patient types characterised by their inadequate response to various therapies, whose data have been collected in clinical trials from 6 months and up to one year. These studies have collected data points to enable derivation of the endpoints described in Section 3.3 as well as the many of the risk factors mentioned in Section REF _Ref461275508 \r \h 2.9. In addition, there is available an abundance of additional explanatory variables from sources such as laboratory data, safety data, vital sign data and immunology sampling. From my reflections on the data described in this chapter, REF _Ref490049353 \h \* MERGEFORMAT Table 4.12 lists out criteria that I consider important in selecting a methodology for predicting ACR20 response in this dataset.Table STYLEREF 1 \s 4. SEQ Table \* ARABIC \s 1 12 Methodology Selection CriteriaCharacteristicCan handle ‘mixed’ dataDataset contains continuous, dichotomous and categorical dataHandling of missing valuesIn some explanatory variables over 50% missing valuesRobust to outliersLarge clinical dataset, large number of variables. Potential for outliers is highComputational Scalability Large clinical dataset, large number of variables.Can manage high dimensionalityLarge clinical dataset, large number of variables.Interpretability Results intended to inform clinical development. Decision makers typically not statistical expertsIn my opinion, the methodology selected to identify predictors for ACR20 in this dataset must be able to handle continuous, dichotomous and categorical explanatory variables. A list of these variables is described in REF _Ref510186600 \h Table 4.7. These data were generated from 4 development programs across 16 clinical trials over 11 years. In that time there were some changes in choice of variables collected. In my dataset this leads to a wide range of missing data rates from 0% to 93% (Section 4.7); therefore, in my opinion the methodology selected must be able to manage missing data without deleting entire patients. With over 11,000 patients from multiple development programs, studies, years and countries any chosen methodology must be resistant to impact of outliers. This is a large dataset in terms of number of patients (n=11,705) and in the number of independent variables (144 variables), therefore computational scalability and ability to manage high dimensionality will be important requirements. Finally, in order to share the results of this research to a clinical audience, I maintain that the interpretability of findings is desirable.This data will inform the selection criteria for methods explored in Chapters 5 and 6. The features of the data with respect to amount of missing data, the mixture of categorical, dichotomous and continuous variables and the correlation structures will guide the parameters for the simulations described in the Chapter 7. The results of the missing value assessment of the clinical dataset together with the simulations investigation of missing values will inform how the final analysis will be performed. The level of actual correlations will be considered in the assessment of the simulations which will inform methods for statistical analysis in Chapter 8. : Data Mining Methodologies5.1. IntroductionAs indicated in the abstract and in Chapter 1, this research aims to explore and evaluate methods to predict response outcomes in RA patients. Chapters 2 and 3 presented important characteristics and risk factors for diagnosis and pathophysiology as well as a summary of measures of disease severity. Chapter 3 identified ACR20 as a primary endpoint of interest for the purposes of classification in this research. Chapter 4 summarised the structure and features of the clinical trial data available to this research and made recommendations for criteria required for a classification method. This chapter will describe some methods in more detail and describe how the methods may be used to help predict outcomes for individual patients based on clinical biomarkers. Chapter 6 will go on to review the literature to see how these and other methods have been used in practice to predict outcomes in RA data. By the end of Chapter 6 a recommendation will be made to choose which methods will be applied to simulated data, with an aim to select a single method to apply to the available clinical trial data ( REF _Ref490049461 \h \* MERGEFORMAT Figure 5.1). This research will attempt to predict ACR20, a categorical outcome of response. This type of outcome may be predicted by classification methods such as CART, nearest neighbour averaging and support vector machines or by logistic regression. The advantages, disadvantages and suitability for application to RA data will be explored in this chapter. Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 1 Planned analysis method refinement5.2. AimsThe aims of this chapter are:To describe a selection of data mining methodologies that are being considered for application to the available data;Through application of criteria proposed in Chapter 4, reduce the number of methods that will be explored in later simulations. The following methods will be discussed: CART (including Boosting, Bagging and Random Forest); K-nearest neighbour classification;Support vector machines;Logistic regression;Lasso;Principal components;Cross validation. 5.3.CARTClassification and regression tree (CART) models are a form of supervised learning, i.e. where the target of the analysis is to identify a relationship between a set of explanatory variables and one or more outcome variables ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011). CART was first introduced by Breiman ADDIN EN.CITE <EndNote><Cite><Author>Breiman</Author><Year>1984</Year><RecNum>125</RecNum><DisplayText>(Breiman et al., 1984)</DisplayText><record><rec-number>125</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">125</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Breiman, L.</author><author>Friedman, J.</author><author>Olshen, R.A.</author><author>Stone, C.J.</author></authors></contributors><titles><title>Classification and Regression Trees</title></titles><pages>358</pages><dates><year>1984</year></dates><publisher>Chapman and Hall</publisher><urls></urls></record></Cite></EndNote>(Breiman et al., 1984) who brought together the concepts classification tree analysis and regression tree analysis under a single method; where classification tree analysis is when the predicted outcome is a class (e.g. responder or non-responder), and regression tree analysis is when the outcome is numerical (e.g. change in cholesterol). The basic purpose of classification is to produce an accurate classifier to uncover predictive structure of a problem. i.e. which variables or interaction of variables drive a response. CART models are suited for the analysis of complex data as well as being flexible and robust analytically. They can deal with nonlinear relationships, high-order interactions, and missing values. Classification and regression trees explain variation of a single response variable by repeatedly splitting the data into more homogeneous groups, using combinations of explanatory variables that may be categorical and/or continuous. CART in a classification setting works in a stepwise fashion by first considering all the possible values of the explanatory variables and identifying a cut point whereby the responses in the outcome variable are best separated (primary splitter). Here ‘best’ is determined by the split that maximises a reduction in impurity. Impurity is a measure that takes the value zero when all of the cases in the node are the same (homogeneous) and increases as the contents of the node become heterogeneous. For classification, impurity is derived from the proportions, c, of responses in each category. Common methods of determining impurity are: the information index (entropy) and the Gini index. Entropy takes the form -∑c ln(c), where ∑ is summation over categories. The Gini index takes the form 1-∑c2. Entropy and Gini index have similar performance for growing trees ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011). In RA, a split may be say, the baseline CRP value of 0.37 that best partitions responders from non-responders. In the next step CART will consider each partition and again identify a cut point whereby the responses in the outcome variable are best separated (e.g. Age = 50). This continues until a stopping rule is applied. Stopping rules include splitting until a minimum node size is reached or until no splits further decrease the impurity beyond a specified threshold. The predicted value for a terminal node is decided by the majority vote within that node. CART methodology can effectively manage missing values by using surrogate splits. These are alternative variables and values that split the data in the same way as the primary splitter. When CART partitions the data, the remaining independent variables are ranked according to the proportion of cases in which they would misclassify the primary splitter. Predictors that do no better than the marginal distribution of the primary splitter are dropped. The variable with the lowest misclassification rate for the primary splitter is then used in place of the primary splitter to assign classes when primary splitter is missing. If this first surrogate is missing, then the next best surrogate can be used. If second surrogate is missing then third surrogate can be used, and so on. When it comes to making predictions on future data, the surrogate can be used when primary splitter is missing.The tree is usually represented graphically thus aiding exploration and understanding ADDIN EN.CITE <EndNote><Cite><Author>Breiman</Author><Year>1984</Year><RecNum>125</RecNum><DisplayText>(Breiman et al., 1984)</DisplayText><record><rec-number>125</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">125</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Breiman, L.</author><author>Friedman, J.</author><author>Olshen, R.A.</author><author>Stone, C.J.</author></authors></contributors><titles><title>Classification and Regression Trees</title></titles><pages>358</pages><dates><year>1984</year></dates><publisher>Chapman and Hall</publisher><urls></urls></record></Cite></EndNote>(Breiman et al., 1984). See REF _Ref460429174 \h \* MERGEFORMAT Figure 5.2 for a hypothetical example of CART illustrating factors leading to progression of a disease. From the first node (top circle), 25% of the patients are over 50 years old, and they had a progression of disease with probability 0.6. Fifteen percent of the patients who were 50 years old and younger, had no previous history of disease and were not female had a progression of disease with probability 0.2.Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 2 Example of CART Advantages of trees include: the flexibility to handle a broad range of response types, such as numeric, categorical, ratings, and survival data; they are intuitive to interpret; can handle large datasets; and the ability to handle missing values in both response and explanatory variables ADDIN EN.CITE <EndNote><Cite><Author>Breiman</Author><Year>1984</Year><RecNum>125</RecNum><DisplayText>(Breiman et al., 1984)</DisplayText><record><rec-number>125</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">125</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Breiman, L.</author><author>Friedman, J.</author><author>Olshen, R.A.</author><author>Stone, C.J.</author></authors></contributors><titles><title>Classification and Regression Trees</title></titles><pages>358</pages><dates><year>1984</year></dates><publisher>Chapman and Hall</publisher><urls></urls></record></Cite></EndNote>(Breiman et al., 1984). Disadvantages of CART include: large trees can become hard to interpret; prediction performance can be poor in higher dimensions where the number of explanatory variables is high; and over-emphasising interactions, i.e. the ‘best’ model may be main effects only ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011).Although it is a potential problem with classification, highly imbalanced classification where one class greatly outnumbers the other (in a binary setting) can lead to a distorted degree of accuracy with the majority class enjoying a high level of accuracy (close to 100%) and the minority class having low accuracy (0 to 10%) ADDIN EN.CITE <EndNote><Cite><Author>He</Author><Year>2009</Year><RecNum>404</RecNum><DisplayText>(He & Garcia, 2009)</DisplayText><record><rec-number>404</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1525021919">404</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>He, H.</author><author>Garcia, E. A.</author></authors></contributors><titles><title>Learning from Imbalanced Data</title><secondary-title>IEEE Transactions on Knowledge and Data Engineering</secondary-title></titles><periodical><full-title>IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING</full-title></periodical><pages>1263-1284</pages><volume>21</volume><number>9</number><dates><year>2009</year></dates><urls></urls></record></Cite></EndNote>(He & Garcia, 2009). To some degree this bias can be ameliorated by applying a cost function so that a higher or lower cost can be associated with misclassifying a responder compared to misclassifying a non-responder. Other methods to overcome the problems associated with imbalanced response include: under-sampling, and over-sampling. In under-sampling, a subset of observations is chosen (randomly or informatively) from the majority class, thus eliminating data until the responses are balanced. Over-sampling works with the minority class by replicating observations (randomly or informatively) to synthetically balance the classes. In the clinical data available for this research, there are differences in response rates, but those differences are not extreme enough to warrant adjustments.The performance of classification can be measured by sensitivity and specificity ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011). Sensitivity is the probability of predicting, for example, a disease when the true state is disease. Conversely specificity is the probability of predicting, non- disease when the true state is non-disease ADDIN EN.CITE <EndNote><Cite><Author>Altman</Author><Year>1994</Year><RecNum>126</RecNum><DisplayText>(Altman & Bland, 1994)</DisplayText><record><rec-number>126</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">126</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Altman, D.G.</author><author>Bland, J.M.</author></authors></contributors><titles><title>Diagnostic Tests 1: sensitivity and specificity</title><secondary-title>BMJ</secondary-title></titles><periodical><full-title>BMJ</full-title></periodical><pages>1552</pages><volume>308</volume><section>1552</section><dates><year>1994</year></dates><urls></urls></record></Cite></EndNote>(Altman & Bland, 1994). By varying cut points on trees (cost function) and plotting sensitivity vs. specificity, Receiver Operating Characteristic (ROC) Curves ADDIN EN.CITE <EndNote><Cite ExcludeYear="1"><Author>Casey</Author><Year>1996</Year><RecNum>155</RecNum><DisplayText>(Casey et al.)</DisplayText><record><rec-number>155</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">155</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Casey, A.T.</author><author>Bland, J.M.</author><author>Crockard, H.A. </author></authors></contributors><titles><title>Development of a functional scoring system for rheumatoid arthritis patients with cervical myelopathy</title><secondary-title>Annals of the rheumatic diseases </secondary-title></titles><periodical><full-title>Annals of the Rheumatic Diseases</full-title></periodical><pages>901-6</pages><volume>55</volume><number>12</number><dates><year>1996</year></dates><urls></urls></record></Cite></EndNote>(Casey et al.) curves can be constructed ( REF _Ref460429200 \h \* MERGEFORMAT Figure 5.3), to measure performance.Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 3 Example ROC CurveWhilst the performance of CART can be poor in high dimensions (i.e. a large number of explanatory variables), there are methods available which can greatly enhance performance such as Bagging, Random Forests and Boosting ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011).5.3.1. BaggingBagging, or bootstrap aggregation, begins by splitting the available dataset into training and test datasets. Bagging then generates many trees by constructing bootstrap samples with replacement from the training data, where each bootstrapped sample is the same size or smaller than the training data. For each bootstrapped sample a tree is constructed. In the prediction, using test data, each data point is tested against each tree and the most frequent classification is selected as the ‘bagged’ predictor, i.e. simple voting determines classification ADDIN EN.CITE <EndNote><Cite><Author>Breiman</Author><Year>1996</Year><RecNum>127</RecNum><DisplayText>(Breiman, 1996)</DisplayText><record><rec-number>127</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">127</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Breiman, L.</author></authors></contributors><titles><title>Bagging Predictors</title><secondary-title>Machine Learning</secondary-title></titles><pages>123-140</pages><volume>24</volume><number>2</number><section>123</section><dates><year>1996</year></dates><urls></urls></record></Cite></EndNote>(Breiman, 1996) (Figure 5.4). By taking an average of multiple trees bootstrapped from the training set, this method has the attribute of reducing the variance of the estimated prediction.21844010795Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 4 Bagging0Figure 5.4 BaggingTraining Data SetTest Data SetClass 0, 1Class 0, 1Class 0, 1Classification by majority voteBootstrapped resamplingTraining Data SetTest Data SetClass 0, 1Class 0, 1Class 0, 1Classification by majority voteBootstrapped resampling5.3.2. BoostingBoosting, like bagging, is an approach that can be used to learn from multiple trees ADDIN EN.CITE <EndNote><Cite><Author>Freund</Author><Year>1999</Year><RecNum>128</RecNum><DisplayText>(Freund & Schapire, 1999)</DisplayText><record><rec-number>128</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">128</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Freund, Y.</author><author>Schapire, R.E.</author></authors></contributors><titles><title>A Short Introduction to Boosting</title><secondary-title>Journal of Japanese Society for Artificial Intelligence</secondary-title></titles><pages>771-780</pages><volume>14</volume><number>5</number><dates><year>1999</year></dates><urls></urls></record></Cite></EndNote>(Freund & Schapire, 1999). Boosting fits many large or small trees to reweighted versions of the training data. Boosting begins in the same way as bagging by creating a resampled dataset from the training data. This first bagged dataset is used to create a tree. The performance of this first tree is tested on the original training data, and the data points in the original data that are poorly predicted are identified. This procedure is repeated in the second round resampling, with the difference that the previously identified points that were poorly predicted are given a greater weight of being selected in the next bagged dataset. Once again new poorly predicted data points in the training set are identified and the procedure continues. As with bagging the test data is applied to each generated tree and classification is determined by a majority vote ( REF _Ref490049786 \h \* MERGEFORMAT Figure 5.5). As with bagging this method can greatly reduce the variance of the estimated prediction. Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 5 BoostingTraining Data SetTest Data SetClass 0,1Class 0,1Class 0,1Classification by majority voteBootstrapped resamplingBootstrapped resampling weighted on previous poorly predicted training data points1Class 0,1Class 0,1Class 0,1232Training Data SetTest Data SetClass 0,1Class 0,1Class 0,1Classification by majority voteBootstrapped resamplingBootstrapped resampling weighted on previous poorly predicted training data points1Class 0,1Class 0,1Class 0,1232 5.3.3. Random ForestA more popular variant of Bagging is Random Forests ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011){Hastie, 2011 #124}. A known feature of bagging with large datasets is that even with bootstrapped resampling, the method can produce very similar trees with each iteration, thus increasing the likelihood of over-fitting and limiting the opportunity to reduce prediction variance. Random Forest methodology attempts to mitigate this by not only bootstrapping the cases in the training set but also randomly selecting a subset of features on which to create each bootstrapped tree. This brings some additional heterogeneity into the generated trees, thus reducing risk of over-fitting. This can result in an increased reduction of variance of the estimated prediction compared with bagging alone (Figure 5.6).-30480448310Training Data SetTest Data SetClass 0, 1Class 0, 1Class 0, 1Classification by majority voteBootstrapped resampling, on subset of variables from training 00Training Data SetTest Data SetClass 0, 1Class 0, 1Class 0, 1Classification by majority voteBootstrapped resampling, on subset of variables from training 24342366040Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 6 Random Forest0Figure 5.6 Random ForestThe advantages of CART and its refinements lend the methods well to RA data and an ACR20 endpoint ( REF _Ref490049823 \h \* MERGEFORMAT Table 5.1). From my review of the methodology, CART can manage missing values without loss of patient numbers and can handle a mixture of categorical and continuous explanatory variables. The RA data for analysis described in chapter 4 has over 11,000 patients and over 140 explanatory variables, in my opinion CART is an appropriate methodology to apply here.Table STYLEREF 1 \s 5. SEQ Table \* ARABIC \s 1 1 Methodology Selection Criteria for CARTCharacteristicAssessmentRatingCan handle ‘mixed’ dataCART can manage missing values without loss of patient numbers and can handle a mixture of categorical and continuous explanatory variables.GoodHandling of missing valuesCART can make use of surrogate variables to overcome missing values.FairRobust to outliersCART applies ‘best’ cut for each split. Outliers have little impact. GoodComputational Scalability Can handle large NGoodCan manage high dimensionalityPerforms internal feature selectionGoodInterpretability Trees can be too large to interpret if not pruned Fair5.4. k-Nearest Neighbour Classification Given a specific combination of dependent variables k-nearest neighbour classifiers use the responses of the k-nearest points to classify a response for a specific data point. For example, suppose we are given 3 factors; a patient’s age, time since diagnosis and CRP level. Then we can look for the nearest k points in 3 dimensional Euclidean space ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011). By noting whether these ‘neighbours’ are responders or non-responders to treatment we can apply a rule such as majority vote to classify the new patient, i.e. we classify the new patient as a responder if the majority of its k neighbours are responders. The advantages of nearest neighbour methods are: they are a good method when prediction dimension is low; when k=1 works well for 1/3 of problems and outperforms other methods for classification The disadvantages are that these methods often fail when we have a large number of predictor variables leading to high dimensionality and they are not straightforward to interpret ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011).Improvements can be made to nearest neighbour methods by applying kernel smoothing and / or assuming structure such as linear models, additive models or to restrict predictors to bring number of dimensions down. Kernel smoothing applies weights to the k nearest points (e.g. by inverse of distance).In my opinion, the application of k-nearest neighbour classification are limited in the context of RA data for this research ( REF _Ref490049850 \h \* MERGEFORMAT Table 5.2). Firstly, the data has high dimensionality (denoted by many explanatory variables) so that this method is known to perform badly. Secondly k-nearest neighbour classification deletes cases with missing values in explanatory variables when trying to find neighbours.Table STYLEREF 1 \s 5. SEQ Table \* ARABIC \s 1 2 Methodology Selection Criteria for k-Nearest NeighbourCharacteristicAssessmentRatingCan handle ‘mixed’ dataConcept of a ‘neighbour’ is not intuitive for non-ordered categorical variablesPoorHandling of missing valuesWill find next neighbour when encounters missing valuesFairRobust to outliersOutliers often not a neighbourGoodComputational Scalability Computationally intensive with large NFairCan manage high dimensionalityComputation increases exponentially with increase in dimensions (number of variables)PoorInterpretability Concept of neighbour easy to understandGood5.5.Support Vector MachinesA support vector machine ADDIN EN.CITE <EndNote><Cite><Author>Cortes</Author><Year>1995</Year><RecNum>129</RecNum><DisplayText>(Cortes & Vapnik, 1995)</DisplayText><record><rec-number>129</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">129</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cortes, C.</author><author>Vapnik, V.N.</author></authors></contributors><titles><title>Support-Vector Networks</title><secondary-title>Machine Learning</secondary-title></titles><pages>273-297</pages><volume>20</volume><number>3</number><dates><year>1995</year></dates><urls></urls></record></Cite></EndNote>(Cortes & Vapnik, 1995) is a procedure that creates hyperplanes that act as decision boundaries to enable classification of an outcome from a set of input data. The method identifies the largest boundary around the hyperplane based on input data. Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 7 Example Support Vector Machine REF _Ref460429232 \h \* MERGEFORMAT Figure 5.7 illustrates a support vector machine. The crosses and stars represent two types of outcome (e.g. ACR20 responder or non-responder) in multi-dimensional space. The solid line represents a hyperplane, outcomes below the hyperplane would be classified as a cross, items above as a star. The algorithm seeks to maximise the shaded margin area in determining the hyperplane ADDIN EN.CITE <EndNote><Cite><Author>Cortes</Author><Year>1995</Year><RecNum>129</RecNum><DisplayText>(Cortes & Vapnik, 1995)</DisplayText><record><rec-number>129</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">129</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cortes, C.</author><author>Vapnik, V.N.</author></authors></contributors><titles><title>Support-Vector Networks</title><secondary-title>Machine Learning</secondary-title></titles><pages>273-297</pages><volume>20</volume><number>3</number><dates><year>1995</year></dates><urls></urls></record></Cite></EndNote>(Cortes & Vapnik, 1995). REF _Ref460429232 \h \* MERGEFORMAT Figure 5.7 shows a linear decision boundary but the method can be extended to non-linear solutions, and where outcomes are not separable. Interpretation can become challenging where outcomes are not separable. Similar to nearest neighbour methodology, this method does not perform well in higher dimensions. Support vector machines prediction accuracy is generally high and robust to overfitting; however, interpretation can be difficult. Since RA data for this research has missing values, and highly dimensional, my assessment is that this method would not be suitable ( REF _Ref490049880 \h \* MERGEFORMAT Table 5.3).Table STYLEREF 1 \s 5. SEQ Table \* ARABIC \s 1 3 Methodology Selection Criteria for Support Vector MachinesCharacteristicAssessmentRatingCan handle ‘mixed’ dataConcept of a distance not defined for non-ordinal categorical variablesPoorHandling of missing valuesMissing values are pairwise deletedPoorRobust to outliersMethod uses all non-missing values. Outliers contribute to generation of hyperplaneFairComputational Scalability Computationally intensive with large NFairCan manage high dimensionalityPoor performance in high dimensionsPoorInterpretability Difficult to interpret in high dimensions and when outcomes not separablePoor5.6.Linear Regression and Logistic RegressionRegression, either as logistic regression or linear regression is a useful, powerful and a frequently used tool for prediction. Logistic regression is a linear method for classification and is widely used where there are dichotomous responses such as survival / non survival or responder / non-responder. Explanatory variables may be quantitative or qualitative, continuous or categorical. When using regression to predict outcomes we can build models from the available data, such methods as forward and backward stepwise are available to identify an optimal model. Adding variables to the model enables the model to fit to the given data better, i.e. reduces the bias of the model, where bias is the difference between our estimate and the true mean. However, if the model is too complex then small changes to the data can lead to substantial changes to the model (a high variance model). Since the data available for analysis in this research has many potential independent variables, overfitting is a risk. Overfitting occurs when the model is too complex and doesn’t generalise well outside the study sample. One way of controlling for overfitting is by shrinking the coefficients towards, or equal to, zero. This also has the advantage of encouraging simpler models, leading to easier interpretation. Two prominent shrinkage methods are Ridge Regression and Lasso ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011).5.6.1. Ridge RegressionWhen constructing a model, ridge regression imposes a penalty against models that have large coefficients ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011). Specifically, ridge regression limits the sum of squares of the coefficients to less than a given constant. The selection of this constant value can be determined through cross validation (see Section 5.8 below). The left hand side of Figure 5.8(a) represents this graphically. If β hat represents ordinary least squares regression, the ellipses represent the contours of the residual sums of squares. Ordinary least squares regression aims to select the smallest ellipse. The circular shaded area represents the constraint where the sums of squares of the β’s are less than a constant. In ridge regression the β’s selected are those where the contours of the ellipse and circle meet. 5.6.2. LassoSimilarly Lasso penalises models with large coefficients ADDIN EN.CITE <EndNote><Cite><Author>Tibshirani</Author><Year>1996</Year><RecNum>130</RecNum><DisplayText>(Tibshirani, 1996)</DisplayText><record><rec-number>130</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">130</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Tibshirani, R.</author></authors></contributors><titles><title>Regression shrinkage and selection via the lasso</title><secondary-title>Journal of the Royal Statistical Society, Series B</secondary-title></titles><pages>267-288</pages><volume>58</volume><number>1</number><dates><year>1996</year></dates><urls></urls></record></Cite></EndNote>(Tibshirani, 1996). It does this by requiring the sum of the absolute values of the coefficients to less than a constant value. In the right hand side of REF _Ref490049940 \h \* MERGEFORMAT Figure 5.8(b) this constraint is represented by a diamond. Although this is a subtle adaptation of ridge regression, it does have the useful property that very often the intersection between the ordinary least squares regression contour and lasso constraint will occur on an axis. In REF _Ref490049940 \h \* MERGEFORMAT Figure 5.8(b) we can see the intersection occurs when β2 is zero. In general, this allows us to control the number of features in our regression model.Once again this constant value can be determined through cross validation (see Section 5.8 below). The lasso has some advantages over ridge regression in that coefficients can be reduced to zero, effectively removing the parameter from the model. Figure STYLEREF 1 \s 5. SEQ Figure \* ARABIC \s 1 8 Comparison of Ridge Regression and Lassoβ1β2β1β2(a) Ridge Regression(b) Lassoβ1β2β1β2(a) Ridge Regression(b) LassoRegression models are powerful and frequently used. In the presence of collinearity regularisation methods such as ridge regression or lasso can reduce inflated variances of ordinary least squares estimates at the cost of a small amount of bias in the estimators. Regression models can be influenced by extreme values and treat missing values by pairwise deletion ( REF _Ref490049995 \h \* MERGEFORMAT Table 5.4). In the setting of a stepwise or lasso scenario with RA data this can potentially dramatically reduce the effective sample size. In my opinion regression can be a good to fair methodology to apply to the clinical trial database available to this research.Table STYLEREF 1 \s 5. SEQ Table \* ARABIC \s 1 4 Methodology Selection Criteria for Logistic RegressionCharacteristicAssessmentRatingCan handle ‘mixed’ dataEasily managed GoodHandling of missing valuesObservations are pairwise deleted in presence of missing valuesFairRobust to outliersMethod uses all non-missing valuesFairComputational Scalability Can manage large NGoodCan manage high dimensionalityForward selection methods manages high dimensions wellGoodInterpretability Well established method Good5.7.Principal ComponentsThe technique of principal components was introduced at the turn of the 20th century by Karl Pearson ADDIN EN.CITE <EndNote><Cite><Author>Pearson</Author><Year>1901</Year><RecNum>132</RecNum><DisplayText>(Pearson, 1901)</DisplayText><record><rec-number>132</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">132</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Pearson, K.</author></authors></contributors><titles><title>On lines and planes closest fit to a system of points in space</title><secondary-title>Philos. Mag.</secondary-title></titles><pages>559-572</pages><volume>2</volume><dates><year>1901</year></dates><urls></urls></record></Cite></EndNote>(Pearson, 1901). Principal components analysis (PCA), is a dimension reduction technique that can take a large number of variables and create a fewer number of linear combinations of these variables (principal components) that are uncorrelated to each other and yet retain information of the original set of data ADDIN EN.CITE <EndNote><Cite><Author>Fieller</Author><Year>2011</Year><RecNum>131</RecNum><DisplayText>(Fieller, 2011)</DisplayText><record><rec-number>131</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">131</key></foreign-keys><ref-type name="Unpublished Work">34</ref-type><contributors><authors><author>Fieller, N.</author></authors></contributors><titles><title>Multivariate Data Analysis course notes</title></titles><dates><year>2011</year></dates><publisher>University of Sheffield</publisher><urls></urls></record></Cite></EndNote>(Fieller, 2011). In the context of regression these reduced number of principal components can form the input variables to the regression models. For RA data principal components discards records with incomplete data ( REF _Ref490050025 \h \* MERGEFORMAT Table 5.5).Table STYLEREF 1 \s 5. SEQ Table \* ARABIC \s 1 5 Methodology Selection Criteria for Principal ComponentsCharacteristicAssessmentRatingCan handle ‘mixed’ dataProblematic if data contains non-ordinal categorical variablesFairHandling of missing valuesDiscards records with incomplete dataPoorRobust to outliersMethod uses all non-missing valuesFairComputational Scalability Manages large datasets wellGoodCan manage high dimensionalityMethod is designed to manage high dimensionsGoodInterpretability Individual principal components can be difficult to characteriseFair5.8. Cross ValidationThe intention of this research is to derive models through data mining methods that predict outcomes in RA trials. The performance of a model can be measured by its prediction rate on independent test data ADDIN EN.CITE <EndNote><Cite><Author>Kohavi</Author><Year>1995</Year><RecNum>133</RecNum><DisplayText>(Kohavi, 1995)</DisplayText><record><rec-number>133</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">133</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kohavi, R.</author></authors></contributors><titles><title>A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection</title><secondary-title>Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence </secondary-title></titles><pages>1137-1143</pages><volume>2</volume><number>12</number><dates><year>1995</year></dates><urls></urls></record></Cite></EndNote>(Kohavi, 1995). In many of the methods described in this chapter choices on model parameters are often required – these can be referred to as tuning parameters. For example in CART analysis cross validation can be used to identify the optimal tree size that minimises prediction error Tibshirani ADDIN EN.CITE <EndNote><Cite><Author>Hastie</Author><Year>2011</Year><RecNum>124</RecNum><DisplayText>(Hastie et al., 2011)</DisplayText><record><rec-number>124</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">124</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Hastie, T.</author><author>Tibshirani, R.</author><author>Friedman, J.</author></authors></contributors><titles><title>The Elements of Statistical Learning</title></titles><pages>745</pages><edition>Second</edition><dates><year>2011</year></dates><publisher>Springer </publisher><urls></urls><electronic-resource-num>10.1007/b94608</electronic-resource-num></record></Cite></EndNote>(Hastie et al., 2011).Cross validation is a commonly used method that is useful in this context ADDIN EN.CITE <EndNote><Cite><Author>Kohavi</Author><Year>1995</Year><RecNum>133</RecNum><DisplayText>(Kohavi, 1995)</DisplayText><record><rec-number>133</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">133</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kohavi, R.</author></authors></contributors><titles><title>A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection</title><secondary-title>Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence </secondary-title></titles><pages>1137-1143</pages><volume>2</volume><number>12</number><dates><year>1995</year></dates><urls></urls></record></Cite></EndNote>(Kohavi, 1995). In a data rich situation, such as that available for this research, cross validation can be applied by dividing the dataset into three parts: a training set, a validation set, and a test set. The training set would be used to fit the models, the validation set to assess the prediction error rate for the models, and the test set to assess the general error of the final model. Typically, the training set comprises 50%, and 25% for validation and test sets.A common method of cross validation is k-fold cross validation. In this method the training and validation datasets are combined and then split into k roughly equal parts. Then for each k=1,2,3…k, your model is fitted using your kth partition as validation set and the remaining k-1 partitions as training sets. Kohavi ADDIN EN.CITE <EndNote><Cite><Author>Kohavi</Author><Year>1995</Year><RecNum>133</RecNum><DisplayText>(Kohavi, 1995)</DisplayText><record><rec-number>133</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">133</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kohavi, R.</author></authors></contributors><titles><title>A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection</title><secondary-title>Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence </secondary-title></titles><pages>1137-1143</pages><volume>2</volume><number>12</number><dates><year>1995</year></dates><urls></urls></record></Cite></EndNote>(Kohavi, 1995) compared several approaches to estimate accuracy: cross-validation (including regular cross validation, leave-one-out cross-validation, stratified cross-validation) and bootstrap (sample with replacement), and recommended stratified 10-fold cross validation as the best model selection method, as it tends to provide less biased estimation of the accuracy. 5.9.Missing DataThe methods described in this chapter are affected to different degrees by missing data, and certainly this is an important feature of the clinical trial dataset available to this research. One approach to managing the effect of missing data is by pre-processing the data and imputing values. Missing data generally falls into 3 types: missing completely at random, missing at random, and not missing at random. A data point can be considered missing completely at random if the presence of missing is unrelated to any other observed or unobserved variable. A data point can be considered missing at random if the presence of missing is related to other observed variables but not to its own unobserved value. A data point can be considered not missing at random if neither missing completely at random nor missing at random. Most approaches to missing data assume missing completely at random or missing at random.Multiple imputation is a popular choice that approaches missing values based on repeated simulations. In multiple imputation, a set of complete datasets is generated from an existing incomplete dataset. Methods such as regression are used to fill in the missing data. The statistical methods of analysis are then applied to each dataset and outcomes combined to provide estimated results that take into account the uncertainty caused by missing values. One alternative to multiple imputation is simple imputation, in which the missing values are replaced by a single value (e.g. a mean or median), but is known to produce biased results for data that isn’t missing completely at random especially for large amounts of missing data ADDIN EN.CITE <EndNote><Cite><Author>Kabacoff</Author><Year>2015</Year><RecNum>405</RecNum><DisplayText>(Kabacoff, 2015)</DisplayText><record><rec-number>405</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1525082721">405</key></foreign-keys><ref-type name="Book">6</ref-type><contributors><authors><author>Kabacoff, R. I.</author></authors></contributors><titles><title>R In Action</title></titles><dates><year>2015</year></dates><publisher>Manning</publisher><urls></urls></record></Cite></EndNote>(Kabacoff, 2015).5.10.SummaryIn this chapter a number of data mining methodologies have been described that can be applied to a rich RA dataset to predict patient outcomes. At the end of Chapter 4 some desirable characteristics based upon the clinical trial data structure were identified. REF _Ref490050048 \h \* MERGEFORMAT Table 5.6 rates the methods described in this chapter against these characteristics. Table STYLEREF 1 \s 5. SEQ Table \* ARABIC \s 1 6 Some characteristics of classification and data mining methodsCharacteristicCARTk-NNSVMLogistic RegressionPrincipal ComponentsCan handle ‘mixed’ dataGoodPoorPoorGoodFairHandling of missing valuesGoodFairPoorPoorPoorRobust to outliersGoodGoodFairFairFairComputational Scalability (Large N)GoodFairFairGoodGoodCan manage high dimensionalityGoodPoorPoorGoodGoodInterpretabilityGoodGoodPoorGoodFairAs described in Section 5.3 CART is expected to perform well against these criteria, so long as trees are kept to a reasonable size. Performance of CART can also be enhanced by boosting, bagging and random forests, albeit at the cost of interpretability. On balance, logistic regression is a good choice as a classifier so long as missing data doesn’t lead to the deletion of too much data, and that the data is not subject to influence of extreme outliers. Regularisation methods such as lasso and ridge regression can improve logistic performance, but is still subject to data deletion in the presence of missing values. The concept of k-nearest neighbours is intuitive and simple to understand, but is limited of limited utility when applied to high dimensional and mixed data that are features of the clinical trial data described in Chapter 4. Similarly support vector machines fall foul of the same disadvantages without the advantages of robustness to outliers and interpretability. Although not a classification method, principal components analysis could have proved useful to reduce the dimensionality of the data. However, management of missing values, impact of outliers and interpretability of components may make this an undesirable method with respect to the intended clinical trial data.Not only have methods been described that can classify a patient’s response based on explanatory variables (e.g. CART, ridge regression, support vector machines) but also a methodology to control for overfitting (cross validation). At this juncture it appears to me that CART and Logistic regression would be the best methods to consider in simulations. Depending on the impact of missing data, multiple imputation methods can be applied and compared on both these methods.The next chapter will apply systematic reviews to identify where these methods have been applied to RA databases, and also to identify any other methods used in the context of RA and the prediction of patient response. The discussion from this chapter and findings in the next chapter will be applied to select which reduced set of methods will be applied in simulations. : Multivariate Methods Systematic Review6.1.IntroductionChapter 5 described a selection of data mining and statistical methods that may be useful in classifying ACR20 response in RA data. The purpose of this research is to identify any further methods which may have been used in practice to predict patient response with RA data. This chapter will apply systematic review methodology to explore which multivariate statistical methods have been used to identify clinical biomarkers for RA response. This chapter will also apply systematic review methodology to ascertain how data mining methods are applied in RA research. This will inform decisions on which methods to explore using the data available. Finally, the available literature for comparisons of logistic regression and CART in the context of variable selection will be reviewed to inform the simulations performed in Chapter 7.There are various risk factors known to predict an individual’s propensity to develop Rheumatoid Arthritis (RA). Alamanos ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005) discusses the factors smoking history, age, gender, genetics, socioeconomics, geography and ethnicity. Many papers discuss these factors in a univariate fashion, however very few, if any, discuss these factors in a multivariate way. A more sophisticated understanding of the influences of RA may lead to a better approach to personalised healthcare for patients with RA.The systematic reviews considered sources relating to an adult RA population, patients with juvenile variations of RA (e.g. systemic juvenile idiopathic arthritis (sJIA)) and related inflammatory diseases (e.g. Lupus, Castleman’s disease, psoriatic arthritis) were excluded.6.2.AimsThrough the application of systematic reviews, this chapter’s aims are: To identify through available literature which biological factors are considered to influence progression of the disease;To summarise methods and results that are assessed in a multivariate analysis;To identify which data mining methods have been applied in RA;To ascertain whether prediction of ACR20 response has been investigated before.Along with methods described in Chapter 5, these findings will enable recommendations for simulations in Chapter 7.6.3.Systematic review of multivariate methods applied in RA research6.3.1.Methods Systematic reviews were conducted in Medline and Embase. Three reviews were conducted with search terms refined and updated in second and third reviews. Details of the search criteria, results and review annotations can be found in Appendix A.1.The search criteria included data on adult patients with RA. In identifying methods, the following general and specific terms were included in the searches: CART;Multivariate analysis;Classification;Discriminant analysis;Cluster analysis;Factor analysis.These terms were expected to identify both methods described in Chapter 5, as well as any other data mining or statistical method used to predict classification of RA response. 6.3.2.ResultsFollowing systematic reviews conducted in Embase and Medline, 12 articles were identified that applied multivariate or data mining methodology to an Adult RA dataset. These 12 articles are summarised in REF _Ref490050776 \h \* MERGEFORMAT Table 6.1. None of these articles included prediction of ACR20 as an endpoint. Numbers of patients available to these research articles was in the main modest, although 2 papers had access to 1448 and 1905 patients. Methodology applied varied substantially, Generalised Estimating Equations (GEE) were applied in 3 papers to analyse longitudinal categorical endpoints, and principal components were applied on 2 occasions to reduce the dimensionality. CART and logistic regression were applied multiple times. In as much as there was variation in types of methodology applied, there was varied choice of explanatory variables (47 different variables). The most common explanatory variables in these 12 articles were: Joint count, age and ESR which were used in 6, 5, and 5 times respectively. Rheumatoid factor was mentioned in 4 articles and gender 3 times. Table STYLEREF 1 \s 6. SEQ Table \* ARABIC \s 1 1 Summary of Medline and Embase Systematic ReviewsReferenceOutcome Sample SizeStatistical MethodsExplanatory VariablesCurtis et alDisease Adjustment (longitudinal dichotomous assessment)59Pearson CorrelationHierarchical regression Principal ComponentsStressSocial SupportCopingAgeIllness DurationESR Joint Count Dessein et alAtherosclerosis (dichotomous assessment)74Mann Whitney UChi SquareLogistic regressionClassification & Regression Tree (CART)Factor analysis with Varimax rotationAgeGender RaceSmoking HistoryDisease duration, Disease severityHAQESR, CRPJoint Count, Joint space narrowingDrug therapyWaist HypertensionSystolic blood pressureTriglyceridesUric acidHypothyroidismPolymorphonuclear countGlomerular filtration rateRisk of Cardiovascular eventTable 6.1 (continued) Summary of Medline and Embase Systematic ReviewsReferenceOutcome Sample SizeStatistical MethodsExplanatory VariablesWolfe, Pincus and O’DellChange in therapy (longitudinal dichotomous assessment)1905Generalised Estimating Equations (GEE)Logistic regression CARTJoint countPainGlobal severityESR Grip strengthHAQDepressionAnxiety Morning stiffnessFatigueSleep DisturbanceWiles et alDisability (longitudinal dichotomous assessment)684 (325 RA)Principal ComponentsGEEAgeGenderDelay to presentationMorning StiffnessJoint countRheumatoid factor statusPresence of nodulesWatkins et alCoping (multiple outcomes)121Multivariate Analysis of Variance (MANOVA)Hierarchical regressionAge Pain Isaacs and FerraccioliDiscussion of biomarkers-None-Table 6.1 (continued) Summary of Medline and Embase Systematic ReviewsReferenceOutcome Sample SizeStatistical MethodsExplanatory VariablesRioja et alActive vs Quiescent RA using DAS28(continuous assessment)163Partial least squares discriminant analysis Age GenderCRP, ESR, RFJoint countsDASConcomitant medicationWild et alDiagnosis of RA (dichotomous assessment)1448(906 RA)Receiver Operating Curves ADDIN EN.CITE <EndNote><Cite ExcludeYear="1"><Author>Casey</Author><Year>1996</Year><RecNum>155</RecNum><DisplayText>(Casey et al.)</DisplayText><record><rec-number>155</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">155</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Casey, A.T.</author><author>Bland, J.M.</author><author>Crockard, H.A. </author></authors></contributors><titles><title>Development of a functional scoring system for rheumatoid arthritis patients with cervical myelopathy</title><secondary-title>Annals of the rheumatic diseases </secondary-title></titles><periodical><full-title>Annals of the Rheumatic Diseases</full-title></periodical><pages>901-6</pages><volume>55</volume><number>12</number><dates><year>1996</year></dates><urls></urls></record></Cite></EndNote>(Casey et al.)Regularised discriminant analysis Anti-CCPGourraud et alGenetic association160Chi-Square, Fishers exact testRF, Genetic Allelesden-Broeder et alRadiological outcomes (dichotomous assessment)47t-test, Wilcoxon, Pearson correlation, logistic regression.Markers of cartilage and synovium turnoverMarkers of endothelial activationTreatment statusImmunoglobulinsRheumatoid factor Baseline DAS & Sharp scoreVoll and BurkhardtErosive disease400NoneUnknownLin et alRA Disease Activity (longitudinal continuous assessment)212GEEserum secreted phospholipaseLansbury index, EffusionsJoint Counts ESRPlatelet count, Haemoglobin 6.4.Systematic Review of Data Mining Methodologies 6.4.1.MethodsAdditional systematic reviews were conducted on specific data mining terms to identify their use with RA data. The data mining terms searched on were: CART, Cross Validation, Nearest Neighbour, Lasso, Bagging, Kernel, Principal Components, Ridge Regression and Support Vector Machines. Details of the systematic review can be found in Appendix A.1.4.6.4.2.ResultsA summary of results is shown in REF _Ref499306157 \h Table 6.2. Even though over 40 articles were identified that applied data mining methodology to an RA setting, just one applied these methods to predict RA outcomes from biomarkers.Table STYLEREF 1 \s 6. SEQ Table \* ARABIC \s 1 2 Data Mining Systematic ReviewData Mining/ Statistical TermArticlesPrediction of RA outcomes in Adults using clinical biomarkersCART40Bagging00Boosting00Random Forest20Nearest Neighbour30Kernel20Support Vector Machines10Ridge Regression00Lasso00Principal Components140Cross Validation616.4.3.CARTThis systematic review yielded 4 results for the use of CART in RA data, none of which addressed predicting patient response in RA progression. Dessein ADDIN EN.CITE <EndNote><Cite><Author>Dessein</Author><Year>2005</Year><RecNum>139</RecNum><DisplayText>(Dessein et al., 2005)</DisplayText><record><rec-number>139</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">139</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Dessein, P.H.</author><author>Joffe, B.I.</author><author>Veller, M.G.</author><author>Stevens, B.A.</author><author>Tobias, M.</author><author>Reddi, K</author><author>Stanwix, A.E.</author></authors></contributors><titles><title>Traditional and Nontraditional Cardiovascular Risk Factors Are Associated with Atherosclerosis in Rheumatoid Arthritis</title><secondary-title>Journal of Rheumatology</secondary-title></titles><pages>435-42</pages><volume>32</volume><number>3</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(Dessein et al., 2005) investigated cardiovascular risk factors are associated with atherosclerosis in RA, CART was applied here to compare to models developed from application of logistic regression. Wolfe ADDIN EN.CITE <EndNote><Cite><Author>Wolfe</Author><Year>2001</Year><RecNum>140</RecNum><DisplayText>(Wolfe et al., 2001)</DisplayText><record><rec-number>140</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">140</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Wolfe, F.</author><author>Pincus, T.</author><author>O'Dell, J.</author></authors></contributors><titles><title>Evaluation and documentation of rheumatoid arthritis disease status in the clinic: which variables best predict change in therapy</title><secondary-title>The Journal of rheumatology </secondary-title></titles><pages>1712-7</pages><volume>28</volume><number>7</number><dates><year>2001</year></dates><urls></urls></record></Cite></EndNote>(Wolfe et al., 2001) aimed to predict change in therapy and also employed CART and logistic regression. The authors turned to CART for nonlinear prediction and to better manage missing values. Sokoloff and Varma ADDIN EN.CITE <EndNote><Cite><Author>Sokoloff</Author><Year>1988</Year><RecNum>145</RecNum><DisplayText>(Sokoloff & Varma, 1988)</DisplayText><record><rec-number>145</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">145</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Sokoloff, L</author><author>Varma, A.A</author></authors></contributors><titles><title>Chondrocalcinosis in surgically resected joints</title><secondary-title>Arthritis and Rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>6</pages><volume>1988</volume><number>31</number><section>750-6</section><dates><year>1988</year></dates><urls></urls></record></Cite></EndNote>(Sokoloff & Varma, 1988) studied Chondrocalcinosis in surgically resected joints, however ‘CART’ was picked up by the systematic review due to the included phrase “which….was the horse and which was the cart’. Philbin, Ries and French ADDIN EN.CITE <EndNote><Cite><Author>Philbin</Author><Year>1995</Year><RecNum>146</RecNum><DisplayText>(Philbin et al., 1995)</DisplayText><record><rec-number>146</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473705605">146</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Philbin, E. F.</author><author>Ries, M.D.</author><author>French, T.S.</author></authors></contributors><titles><title>Feasibility of maximal cardiopulmonary exercise testing in patients with end-stage arthritis of the hip and knee prior to total joint arthroplasty</title><secondary-title>Chest</secondary-title></titles><periodical><full-title>Chest</full-title></periodical><pages>174-81</pages><volume>108</volume><number>1</number><dates><year>1995</year></dates><urls></urls></record></Cite></EndNote>(Philbin et al., 1995) looked at the effect of exercise on hip and knee, and again CART in this paper was not a statistical method but an exercise measuring device (‘metabolic cart’).No results were found for Boosting and Bagging extensions of CART.6.4.4.Random ForestTwo articles were identified that applied Random Forest methodology to RA data. O’Hanlon ADDIN EN.CITE <EndNote><Cite><Author>O'Hanlon</Author><Year>2011</Year><RecNum>147</RecNum><DisplayText>(O'Hanlon et al., 2011)</DisplayText><record><rec-number>147</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473705907">147</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>O'Hanlon, T.P.</author><author>Li, Z.</author><author>Gan, L.</author><author>Gourley, M.F.</author><author>Rider, L.G</author><author>Miller, F.W.</author></authors></contributors><titles><title>Plasma proteomic profiles from disease-discordant monozygotic twins suggest that molecular pathways are shared in multiple systemic autoimmune diseases</title><secondary-title>Arthritis research & therapy </secondary-title></titles><periodical><full-title>Arthritis research & therapy</full-title></periodical><pages>R181</pages><volume>13</volume><number>6</number><dates><year>2011</year></dates><urls></urls></record></Cite></EndNote>(O'Hanlon et al., 2011) applied multivariate random forest modelling to differentiate different types of autoimmune diseases in a small selection of monozygotic twins. Wolfe and Michaud ADDIN EN.CITE <EndNote><Cite><Author>Wolfe</Author><Year>2009</Year><RecNum>148</RecNum><DisplayText>(Wolfe & Michaud, 2009)</DisplayText><record><rec-number>148</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">148</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Wolfe, F.,</author><author> Michaud, K., </author></authors></contributors><titles><title>Predicting depression in rheumatoid arthritis: the signal importance of pain extent and fatigue, and comorbidity</title><secondary-title>Arthritis and rheumatism</secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>667-73</pages><volume>61</volume><number>5</number><dates><year>2009</year></dates><urls></urls></record></Cite></EndNote>(Wolfe & Michaud, 2009) applied CART and Random Forest to predict depression in RA patients.6.4.5.Nearest NeighbourSystematic review revealed 3 results associating Nearest Neighbour methodology with Adult RA data. Park ADDIN EN.CITE <EndNote><Cite><Author>Park</Author><Year>2007</Year><RecNum>149</RecNum><DisplayText>(Park et al., 2007)</DisplayText><record><rec-number>149</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">149</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Park, G.S.</author><author>Wong, W.K.</author><author>Oh, M. </author><author>Khanna, D.</author><author>Gold, R.H</author><author>Sharp, J.T.</author><author>Paulus, H.E.</author></authors></contributors><titles><title>Classifying radiographic progression status in early rheumatoid arthritis patients using propensity scores to adjust for baseline differences</title><secondary-title>Statistical methods in medical research </secondary-title></titles><pages>13-29</pages><volume>16</volume><number>1</number><dates><year>2007</year></dates><urls></urls></record></Cite></EndNote>(Park et al., 2007) applied nearest neighbour as well as bootstrapping and kernel methods to establish propensity scores for matching patients in an observational radiographic RA dataset. Edwards ADDIN EN.CITE <EndNote><Cite><Author>Edwards</Author><Year>2007</Year><RecNum>150</RecNum><DisplayText>(Edwards et al., 2007)</DisplayText><record><rec-number>150</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">150</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Edwards, C.J. </author><author>Feldman, J.L.</author><author>Beech, J.</author><author>Shields, K.M.</author><author>Stover, J.A. </author><author>Trepicchio,W.L.</author><author>Larsen G.</author><author>Foxwell,B.M.J.</author><author>Brennan, F.M.</author><author>Feldmann, M.</author><author>Pittman, D.D.</author></authors></contributors><titles><title>Molecular profile of peripheral blood mononuclear cells from patients with rheumatoid arthritis</title><secondary-title>Molecular medicine </secondary-title></titles><pages>40-58</pages><volume>13</volume><number>1-2</number><dates><year>2007</year></dates><urls></urls></record></Cite></EndNote>(Edwards et al., 2007) applied a nearest neighbour analysis to identify genetic transcripts preferentially expressed in RA that they proposed could aid early diagnosis and treatment. Carano ADDIN EN.CITE <EndNote><Cite><Author>Carano</Author><Year>2004</Year><RecNum>151</RecNum><DisplayText>(Carano et al., 2004)</DisplayText><record><rec-number>151</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">151</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Carano, R.A.D.</author><author>Lynch, J.A.</author><author>Redei, J.</author><author>Ostrowitzki, S.</author><author>Miaux, Y</author><author>Zaim, S.</author><author>White, D.L.</author><author>Peterfy, C.G.</author><author>Genant H.K.</author></authors></contributors><titles><title>Multispectral analysis of bone lesions in the hands of patients with rheumatoid arthritis</title><secondary-title>Magnetic resonance imaging </secondary-title></titles><pages>505-14</pages><volume>22</volume><number>4</number><dates><year>2004</year></dates><urls></urls></record></Cite></EndNote>(Carano et al., 2004) used a nearest neighbour analysis to evaluate a new tool to classify RA patients by extent of bone lesions appearing in hands. 6.4.6.KernelThe keyword kernel flagged two articles, Park ADDIN EN.CITE <EndNote><Cite><Author>Park</Author><Year>2007</Year><RecNum>149</RecNum><DisplayText>(Park et al., 2007)</DisplayText><record><rec-number>149</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">149</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Park, G.S.</author><author>Wong, W.K.</author><author>Oh, M. </author><author>Khanna, D.</author><author>Gold, R.H</author><author>Sharp, J.T.</author><author>Paulus, H.E.</author></authors></contributors><titles><title>Classifying radiographic progression status in early rheumatoid arthritis patients using propensity scores to adjust for baseline differences</title><secondary-title>Statistical methods in medical research </secondary-title></titles><pages>13-29</pages><volume>16</volume><number>1</number><dates><year>2007</year></dates><urls></urls></record></Cite></EndNote>(Park et al., 2007) is described in Section 6.4.5. Yokoyama and Muto ADDIN EN.CITE <EndNote><Cite><Author>Yokoyama</Author><Year>2006</Year><RecNum>152</RecNum><DisplayText>(Yokoyama & Muto, 2006)</DisplayText><record><rec-number>152</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">152</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Yokoyama, E.</author><author>Muto, M. </author></authors></contributors><titles><title>Adult variant of self-healing papular mucinosis in a patient with rheumatoid arthritis: predominant proliferation of dermal dendritic cells expressing CD34 or factor XIIIa in association with dermal deposition of mucin</title><secondary-title>The Journal of dermatology </secondary-title></titles><pages>30-5</pages><volume>33</volume><number>1</number><dates><year>2006</year></dates><urls></urls></record></Cite></EndNote>(Yokoyama & Muto, 2006) presented a case study of a RA patient with rice kernel sized papules, kernel in this article refers to size of blisters rather than a statistical method.6.4.7.Support Vector Machines One article from Liu ADDIN EN.CITE <EndNote><Cite><Author>Liu</Author><Year>2008</Year><RecNum>153</RecNum><DisplayText>(Liu et al., 2008)</DisplayText><record><rec-number>153</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">153</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Liu, W.</author><author>Li, X.</author><author>Ding, F.</author><author>Li, Y </author></authors></contributors><titles><title>Using SELDI-TOF MS to identify serum biomarkers of rheumatoid arthritis</title><secondary-title>Scandinavian journal of rheumatology </secondary-title></titles><pages>94-102</pages><volume>37</volume><number>2</number><dates><year>2008</year></dates><urls></urls></record></Cite></EndNote>(Liu et al., 2008) used support vector machine as a classifier to illustrate how mass spectrometry could identify potential biomarkers for the diagnosis of RA. 6.4.8.Ridge Regression and LassoNo articles were identified that applied the shrinkage methods ridge regression and Lasso to RA data.6.4.9.Principal ComponentsOf the 14 results yielded for principal components analysis, Jones et al ADDIN EN.CITE <EndNote><Cite><Author>Jones</Author><Year>1993</Year><RecNum>158</RecNum><DisplayText>(Jones et al., 1993)</DisplayText><record><rec-number>158</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">158</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Jones, P.W.</author><author>Ziade, M.F.,</author><author>Davis, M J</author><author>Dawes, P.T</author></authors></contributors><titles><title>An index of disease activity in rheumatoid arthritis</title><secondary-title>Statistics in medicine </secondary-title></titles><pages>1171-81</pages><volume>12</volume><number>12</number><dates><year>1993</year></dates><urls></urls></record></Cite></EndNote>(Jones et al., 1993), Burckhardt and Bjelle ADDIN EN.CITE <EndNote><Cite><Author>Burckhardt</Author><Year>1994</Year><RecNum>159</RecNum><DisplayText>(Burckhardt & Bjelle, 1994)</DisplayText><record><rec-number>159</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">159</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Burckhardt, C.S.</author><author>Bjelle, A.A.</author></authors></contributors><titles><title>Swedish version of the short-form McGill Pain Questionnaire</title><secondary-title>Scandinavian journal of rheumatology </secondary-title></titles><pages>77-81</pages><volume>23</volume><number>2</number><dates><year>1994</year></dates><urls><related-urls><url>;(Burckhardt & Bjelle, 1994), Guillemin, Brian?on and Pourel ADDIN EN.CITE <EndNote><Cite><Author>Guillemin</Author><Year>1992</Year><RecNum>160</RecNum><DisplayText>(Guillemin et al., 1992)</DisplayText><record><rec-number>160</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">160</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Guillemin, F.</author><author>Brian?on, S.</author><author>Pourel, J,</author></authors></contributors><titles><title>Validity and discriminant ability of the HAQ Functional Index in early rheumatoid arthritis.</title><secondary-title>Disability and rehabilitation </secondary-title></titles><pages>71-7</pages><volume>14</volume><number>2</number><dates><year>1992</year></dates><urls><related-urls><url>;(Guillemin et al., 1992), Suurmeijer ADDIN EN.CITE <EndNote><Cite><Author>Suurmeijer</Author><Year>1994</Year><RecNum>161</RecNum><DisplayText>(Suurmeijer et al., 1994)</DisplayText><record><rec-number>161</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">161</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Suurmeijer, T.P.</author><author>Doeglas, D.M.</author><author>Moum, T.</author><author>Brian?on, S.</author><author>Krol, B.</author><author>Sanderman, R,</author><author>Guillemin, F.</author><author>Bjelle, A.</author><author>van den Heuvel, W.J.</author></authors></contributors><titles><title>The Groningen Activity Restriction Scale for measuring disability: its utility in international comparisons</title><secondary-title>American journal of public health</secondary-title></titles><pages>1270-3</pages><volume>84</volume><number>8</number><dates><year>1994</year></dates><urls></urls></record></Cite></EndNote>(Suurmeijer et al., 1994), Barlow and Wright ADDIN EN.CITE <EndNote><Cite><Author>Barlow</Author><Year>1998</Year><RecNum>154</RecNum><DisplayText>(Barlow & Wright, 1998)</DisplayText><record><rec-number>154</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">154</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Barlow, J.H. </author><author>Wright, C.C.</author></authors></contributors><titles><title>Dimensions of the Center of Epidemiological Studies-Depression Scale for people with arthritis from the UK</title><secondary-title>Psychological reports </secondary-title></titles><pages>915-9</pages><volume>83</volume><number>3</number><dates><year>1998</year></dates><urls></urls></record></Cite></EndNote>(Barlow & Wright, 1998), and Szilasiová et al ADDIN EN.CITE <EndNote><Cite><Author>Szilasiová</Author><Year>2002</Year><RecNum>162</RecNum><DisplayText>(Szilasiová et al., 2002)</DisplayText><record><rec-number>162</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">162</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Szilasiová, A. </author><author>Macejová, Z.</author><author>Nagyová, I,.</author><author>Kovarová, M.</author><author>Béresová, A</author><author>Szilasiová J.</author></authors></contributors><titles><title>Reliability and validation of the Slovak modified version of the Stanford Health Assessment Questionnaire using the functional disability index in patients with rheumatoid arthritis </title><secondary-title>Vnitr?ní lékar?ství </secondary-title></titles><pages>8-16</pages><volume>48</volume><number>1</number><dates><year>2002</year></dates><urls><related-urls><url>;(Szilasiová et al., 2002) implemented the technique to validate various scales related to RA outcomes. Korman et al ADDIN EN.CITE <EndNote><Cite><Author>Korman</Author><Year>2009</Year><RecNum>157</RecNum><DisplayText>(Korman et al., 2009)</DisplayText><record><rec-number>157</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">157</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Korman, B.D.</author><author>Seldin, M.F.</author><author>Taylor, K.E.</author><author>Le, J.M.</author><author>Lee, A.T.</author><author>Plenge, R.M.</author><author>Amos,C.I.</author><author>Criswell,L.A.</author><author>Gregersen,P.K.</author><author>Kastner, D.L.</author><author>Remmers, E.F.</author></authors></contributors><titles><title>The chromosome 7q region association with rheumatoid arthritis in females in a British population is not replicated in a North American case-control series</title><secondary-title>Arthritis and rheumatism </secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>47-52</pages><volume>60</volume><number>1</number><dates><year>2009</year></dates><urls></urls></record></Cite></EndNote>(Korman et al., 2009) and Juang et al ADDIN EN.CITE <EndNote><Cite><Author>Juang</Author><Year>2011</Year><RecNum>163</RecNum><DisplayText>(Juang et al., 2011)</DisplayText><record><rec-number>163</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">163</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Juang, Y.T. </author><author>Peoples, C.</author><author>Kafri, R.</author><author>Kyttaris, V.C.</author><author>Sunahori, K.</author><author>Kis-Toth, K.</author><author>Fitzgerald, L.</author><author>Ergin, S.</author><author>Finnell, M.</author><author>Tsokos, G.C.</author></authors></contributors><titles><title>A systemic lupus erythematosus gene expression array in disease diagnosis and classification: a preliminary report. </title><secondary-title>Lupus</secondary-title></titles><pages>243-9</pages><volume>20</volume><number>3</number><dates><year>2011</year></dates><urls><related-urls><url>;(Juang et al., 2011) used principal components analysis in gene expression studies. Iverson et al ADDIN EN.CITE <EndNote><Cite><Author>Iversen</Author><Year>2004</Year><RecNum>156</RecNum><DisplayText>(Iversen et al., 2004)</DisplayText><record><rec-number>156</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">156</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Iversen, M.D.</author><author>Eaton, H.M.</author><author>Daltroy, L.H., </author></authors></contributors><titles><title>How rheumatologists and patients with rheumatoid arthritis discuss exercise and the influence of discussions on exercise prescriptions</title><secondary-title>Arthritis and rheumatism </secondary-title></titles><periodical><full-title>Arthritis and Rheumatism</full-title></periodical><pages>63-72</pages><volume>51</volume><number>1</number><dates><year>2004</year></dates><urls></urls></record></Cite></EndNote>(Iversen et al., 2004) developed exercise profiles, Chiaravalloti et al ADDIN EN.CITE <EndNote><Cite><Author>Chiaravalloti</Author><Year>2003</Year><RecNum>164</RecNum><DisplayText>(Chiaravalloti et al., 2003)</DisplayText><record><rec-number>164</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473713536">164</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Chiaravalloti, N.D.</author><author>Christodoulou, C.</author><author>Demaree, H.A.,</author><author>DeLuca, J.</author></authors></contributors><titles><title>Differentiating simple versus complex processing speed: influence on new learning and memory performance</title><secondary-title> Journal of clinical and experimental neuropsychology</secondary-title></titles><pages>489-501</pages><volume>25</volume><number>4</number><dates><year>2003</year></dates><urls></urls></record></Cite></EndNote>(Chiaravalloti et al., 2003) identified factors that influence learning and memory, and Girard et al ADDIN EN.CITE <EndNote><Cite><Author>Girard</Author><Year>2002</Year><RecNum>165</RecNum><DisplayText>(Girard et al., 2002)</DisplayText><record><rec-number>165</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">165</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Girard, F.</author><author>Guillemin, F. </author><author>Novella, J.L.</author><author>Valckenaere, I.</author><author>Krzanowska, K. </author><author>Vitry, F. </author><author>Vittecoq, O. </author><author>Eschard, J.P.</author><author>Blanchard, F. </author><author>Le Lo?t, X.</author></authors></contributors><titles><title>Health-care use by rheumatoid arthritis patients compared with non-arthritic subjects</title><secondary-title>Rheumatology </secondary-title></titles><pages>167-75</pages><volume>41</volume><number>2</number><dates><year>2002</year></dates><urls></urls></record></Cite></EndNote>(Girard et al., 2002) found patterns in health care use. Casey, Bland and Crockard ADDIN EN.CITE <EndNote><Cite><Author>Casey</Author><Year>1996</Year><RecNum>155</RecNum><DisplayText>(Casey et al., 1996)</DisplayText><record><rec-number>155</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">155</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Casey, A.T.</author><author>Bland, J.M.</author><author>Crockard, H.A. </author></authors></contributors><titles><title>Development of a functional scoring system for rheumatoid arthritis patients with cervical myelopathy</title><secondary-title>Annals of the rheumatic diseases </secondary-title></titles><periodical><full-title>Annals of the Rheumatic Diseases</full-title></periodical><pages>901-6</pages><volume>55</volume><number>12</number><dates><year>1996</year></dates><urls></urls></record></Cite></EndNote>(Casey et al., 1996), used principal components analysis to simplify the 20 question HAQ down to 10 questions for RA patients with cervical myelopathy. Mielenz et al ADDIN EN.CITE <EndNote><Cite><Author>Mielenz</Author><Year>2006</Year><RecNum>166</RecNum><DisplayText>(Mielenz et al., 2006)</DisplayText><record><rec-number>166</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473714884">166</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Mielenz, T.</author><author>Jackson, E.</author><author>Currey, S.</author><author>DeVellis, R.</author><author>Callahan, L.F., </author></authors></contributors><titles><title>Psychometric properties of the Centers for Disease Control and Prevention Health-Related Quality of Life (CDC HRQOL) items in adults with arthritis</title><secondary-title>Health and quality of life outcomes </secondary-title></titles><periodical><full-title>Health and quality of life outcomes</full-title></periodical><pages>1-8</pages><volume>4</volume><number>66</number><dates><year>2006</year></dates><urls></urls></record></Cite></EndNote>(Mielenz et al., 2006) and Goeppinger et al ADDIN EN.CITE <EndNote><Cite><Author>Goeppinger</Author><Year>1998</Year><RecNum>167</RecNum><DisplayText>(Goeppinger et al., 1998)</DisplayText><record><rec-number>167</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">167</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Goeppinger, J.</author><author>Doyle, M.A. </author><author>Charlton, S.L.</author><author>Lorig, K.</author></authors></contributors><titles><title>A nursing perspective on the assessment of function in persons with arthritis</title><secondary-title>Research in nursing & health </secondary-title></titles><pages>321-31</pages><volume>11</volume><number>5</number><dates><year>1998</year></dates><urls></urls></record></Cite></EndNote>(Goeppinger et al., 1998) both used principal components analysis to identify factors that were most influential in RA scales (Health Related Quality of Life and HAQ respectively). 6.4.10. Cross ValidationSix results were identified by the systematic review linking cross validation with RA. Dessein et al ADDIN EN.CITE <EndNote><Cite><Author>Dessein</Author><Year>2005</Year><RecNum>139</RecNum><DisplayText>(Dessein et al., 2005)</DisplayText><record><rec-number>139</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">139</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Dessein, P.H.</author><author>Joffe, B.I.</author><author>Veller, M.G.</author><author>Stevens, B.A.</author><author>Tobias, M.</author><author>Reddi, K</author><author>Stanwix, A.E.</author></authors></contributors><titles><title>Traditional and Nontraditional Cardiovascular Risk Factors Are Associated with Atherosclerosis in Rheumatoid Arthritis</title><secondary-title>Journal of Rheumatology</secondary-title></titles><pages>435-42</pages><volume>32</volume><number>3</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(Dessein et al., 2005) investigated cardiovascular risk factors associated with atherosclerosis in RA and used cross validation to validate prediction of plaques in CART modelling. Ware et al ADDIN EN.CITE <EndNote><Cite><Author>Ware</Author><Year>1999</Year><RecNum>168</RecNum><DisplayText>(Ware et al., 1999)</DisplayText><record><rec-number>168</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">168</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Ware, J.E. </author><author>Keller, S.D.</author><author>Hatoum, H.T.</author><author>Kong, S.X</author></authors></contributors><titles><title>The SF-36 Arthritis-Specific Health Index (ASHI): I. Development and cross-validation of scoring algorithms</title><secondary-title>Medical care </secondary-title></titles><pages>MS40-50. </pages><volume>37</volume><number>5</number><dates><year>1999</year></dates><urls><related-urls><url>;(Ware et al., 1999) developed an Arthritis scale based on SF-36 using multiple studies, and Metsios et al ADDIN EN.CITE <EndNote><Cite><Author>Metsios</Author><Year>2008</Year><RecNum>169</RecNum><DisplayText>(Metsios et al., 2008)</DisplayText><record><rec-number>169</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473715779">169</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Metsios, G.S.</author><author>Stavropoulos-Kalinoglou, A. </author><author>Panoulas, V.F.</author><author>Koutedakis, Y.</author><author>Nevill, A.M.</author><author>Douglas K.M.</author><author>Kita, M.</author><author>Kitas, G.D.</author></authors></contributors><titles><title>New resting energy expenditure prediction equations for patients with rheumatoid arthritis </title><secondary-title>Rheumatology </secondary-title></titles><periodical><full-title>Rheumatology</full-title></periodical><pages>500-506</pages><volume>47</volume><number>4</number><dates><year>2008</year></dates><urls></urls></record></Cite></EndNote>(Metsios et al., 2008) predict resting energy expenditure. Julia et al ADDIN EN.CITE <EndNote><Cite><Author>Julià</Author><Year>2009</Year><RecNum>170</RecNum><DisplayText>(Julià et al., 2009)</DisplayText><record><rec-number>170</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473716007">170</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Julià, A.</author><author>Erra, A.</author><author>Palacio, C.</author><author>Tomas, C.</author><author>Sans, X.</author><author>Barceló, P. </author><author>Marsal , S.</author></authors></contributors><titles><title>An eight-gene blood expression profile predicts the response to infliximab in rheumatoid arthritis</title><secondary-title>PloS one </secondary-title></titles><periodical><full-title>PloS one</full-title></periodical><pages>e7556</pages><volume>4</volume><number>10</number><dates><year>2009</year></dates><urls></urls></record></Cite></EndNote>(Julià et al., 2009) used cross validation to build prediction models to predict response according to EULAR, this had some similarities with the aims of this research albeit in a genetic marker framework. Van der Helm ADDIN EN.CITE <EndNote><Cite><Author>van der Helm</Author><Year>2007</Year><RecNum>46</RecNum><DisplayText>(van der Helm, 2007)</DisplayText><record><rec-number>46</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">46</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>van der Helm, A</author></authors></contributors><titles><title><style face="normal" font="Dutch801BT-Roman" size="100%">Emerging Patterns of Risk Factor Make-Up Enable Subclassification of Rheumatoid Arthritis</style></title><secondary-title>ARTHRITIS & RHEUMATISM</secondary-title></titles><pages>1728-1735</pages><volume>56</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>RA</keyword></keywords><dates><year>2007</year><pub-dates><date>2007</date></pub-dates></dates><label>47</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(van der Helm, 2007) used logistic regression to predict progression from undifferentiated (early) arthritis to RA or other diagnosis. Cross validation was utilised in order to avoid over-fitting of logistic regression models. Cole et al ADDIN EN.CITE <EndNote><Cite><Author>Cole</Author><Year>2005</Year><RecNum>171</RecNum><DisplayText>(Cole et al., 2005)</DisplayText><record><rec-number>171</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473716241">171</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Cole, J.C. </author><author>Motivala, S.J. </author><author>Khanna, D.</author><author>Lee, J.Y.</author><author>Paulus, H.E.</author><author>Irwin, M.R.</author></authors></contributors><titles><title>Validation of single-factor structure and scoring protocol for the Health Assessment Questionnaire-Disability Index</title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>536-42</pages><volume>53</volume><number>4</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(Cole et al., 2005) applied cross validation to RA patient data in order to validate the HAQ DI scale.6.5.Systematic Review SummaryThe data mining systematic reviews identified 43 articles that apply data mining methodology to RA data. Of these just one directly attempted to predict patient outcomes from clinical biomarker data. Van der Helm et al ADDIN EN.CITE <EndNote><Cite><Author>van der Helm</Author><Year>2007</Year><RecNum>46</RecNum><DisplayText>(van der Helm, 2007)</DisplayText><record><rec-number>46</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">46</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>van der Helm, A</author></authors></contributors><titles><title><style face="normal" font="Dutch801BT-Roman" size="100%">Emerging Patterns of Risk Factor Make-Up Enable Subclassification of Rheumatoid Arthritis</style></title><secondary-title>ARTHRITIS & RHEUMATISM</secondary-title></titles><pages>1728-1735</pages><volume>56</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>RA</keyword></keywords><dates><year>2007</year><pub-dates><date>2007</date></pub-dates></dates><label>47</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(van der Helm, 2007) used logistic regression to predict progression to full RA from early undifferentiated arthritis using baseline characteristics. Through a combination of logistic regression and cross validation, 9 clinical variables were identified that together predict progression, to an accuracy of around 87%.Considered together the articles offer some insights into statistical methods utilised used to explore relationships between outcomes and demographic and clinical measures. Wiles et al ADDIN EN.CITE <EndNote><Cite><Author>Wiles</Author><Year>2000</Year><RecNum>141</RecNum><DisplayText>(Wiles et al., 2000)</DisplayText><record><rec-number>141</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">141</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Wiles, N..</author><author>Dunn, G.,</author><author>Barrett, E.,</author><author>Silman, A.,</author><author>Symmons, D. </author></authors></contributors><titles><title>Associations between demographic and disease-related variables and disability over the first five years of inflammatory polyarthritis: a longitudinal analysis using generalized estimating equations. </title><secondary-title>Journal of Clinical Epidemiology</secondary-title></titles><pages>988-96</pages><volume>53</volume><number>10</number><dates><year>2000</year></dates><urls></urls></record></Cite></EndNote>(Wiles et al., 2000) and Watkins et al ADDIN EN.CITE <EndNote><Cite><Author>Watkins</Author><Year>1999</Year><RecNum>142</RecNum><DisplayText>(Watkins et al., 1999)</DisplayText><record><rec-number>142</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">142</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Watkins, K.W.</author><author>Shifren, K.</author><author>Park, D.C.</author><author>Morrell, R.W. </author></authors></contributors><titles><title>Age, pain, and coping with rheumatoid arthritis.</title><secondary-title>Pain</secondary-title></titles><pages>217-28</pages><volume>82</volume><number>3</number><dates><year>1999</year></dates><urls></urls></record></Cite></EndNote>(Watkins et al., 1999) go a little further than the others and begin to consider interactions and longitudinal nature of data. Wild et al PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5XaWxkPC9BdXRob3I+PFllYXI+MjAwODwvWWVhcj48UmVj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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5XaWxkPC9BdXRob3I+PFllYXI+MjAwODwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Wild et al., 2008) included a large number of patients and a modest number of markers to aid diagnosis of RA. In chapter 5 we described some data mining methodologies, in this chapter we applied broad search criteria in systematic reviews to attempt to find other methodology applied to predict RA patient response. No methods were found in addition to those described in Chapter 5. The rating of methods therefore, remains unchanged, REF _Ref490050827 \h \* MERGEFORMAT Table 6.3 summarises some characteristics of these methods. Table STYLEREF 1 \s 6. SEQ Table \* ARABIC \s 1 3 Some characteristics of classification and data mining methodsCharacteristicCARTk-NNSVMLogistic RegressionPrincipal ComponentsCan handle ‘mixed’ dataGoodPoorPoorGoodGoodHandling of missing valuesFairFairPoorFairPoorRobust to outliersGoodGoodPoorFairFairComputational Scalability (Large N)GoodPoorPoorGoodGoodCan manage high dimensionalityGoodPoorPoorGoodGoodInterpretabilityFairGoodPoorGoodFairFollowing findings in chapter 5 and this chapter, it appears that CART and Logistic regression would be methods to consider in simulations. This chapter will next look a systematic review to identify comparisons between CART and logistic regression in variable selection6.6.Systematic Review comparing CART and Logistic RegressionAppendix A.1.5 describes in detail a systematic review that was conducted seeking comparisons of logistic regression and CART in the context of variable selection. Five articles were found that compared logistic regression and CART, none in RA.There has been some literature comparing logistic regression and CART in the context of variable selection. Largely the focus has been on prediction performance rather than variable selection. Only one reference was found that simulated data, and this was in a “semi-artificial” way ADDIN EN.CITE <EndNote><Cite><Author>Rousseau</Author><Year>2008</Year><RecNum>336</RecNum><DisplayText>(Rousseau et al., 2008)</DisplayText><record><rec-number>336</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1519904466">336</key><key app="ENWeb" db-id="">0</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rousseau, R.</author><author>Govaerts, B.</author><author>Verleysen, M.</author><author>Boulanger, B.</author></authors></contributors><titles><title>Comparison of some chemometric tools for metabonomics biomarker identification</title><secondary-title>Chemometrics and Intelligent Laboratory Systems</secondary-title></titles><periodical><full-title>Chemometrics and Intelligent Laboratory Systems</full-title></periodical><pages>54-66</pages><volume>91</volume><number>1</number><section>54</section><dates><year>2008</year></dates><isbn>01697439</isbn><urls></urls><electronic-resource-num>10.1016/j.chemolab.2007.06.008</electronic-resource-num></record></Cite></EndNote>(Rousseau et al., 2008). Literature neither compared the performance of CART and logistic regression to select variables from a large dataset with many variables, nor assessed performance in a variety of scenarios. In each of the articles, there appeared to be no concern with volume of missing data. The numbers of subjects in each of the manuscripts varied from 252 to over 180,000 but the number of explanatory variables was low (<47) compared to the data available in this thesis. Apart from some minor fluctuations, each author found the performance of CART and logistic regression to be comparable. 6.7.SummaryFive systematic reviews were conducted in exploring methods for use later in this research. The first 3 systematic reviews, searched for methods used in adult RA data. The search included cluster analysis, factor analysis, discriminant analysis and more general terms such as multivariate analysis, classification. This search yielded some interesting results but was limited to the context of RA, and thus other methods used in other disease areas or outside medicine may be overlooked. The fourth review considered methods identified in these reviews and their application to predict RA outcomes.In the multiple systematic reviews, no evidence was discovered that research has been published to predict ACR20 response in RA patients from baseline characteristic and biomarker data. Therefore, there is an opportunity with the clinical data available in this research to answer this outstanding question. Through systematic review, joint counts, age, ESR, and rheumatoid factor were identified as common biological factors considered to influence progression of the disease. In this chapter a broad search criteria were applied in systematic reviews to attempt to find other methodology applied to predict RA patient response. No methods were found applied in practice in addition to those described in Chapter 5. In chapter 4 we reviewed the clinical data available for analysis. In REF _Ref490050827 \h \* MERGEFORMAT Table 6.3 we compared the reviewed methods and recommended CART and logistic regression as the highest rated methods for this data. The fifth literature review revealed little consensus on performance of CART over logistic regression in the context of variable selection. Although the search was limited by the term variable selection, the resulting literature also included discussion of prediction performance of CART and logistic regression. Following reviews in this chapter, we will use the characteristics of this data to generate simulated datasets for comparing CART and logistic regression ( REF _Ref490050912 \h \* MERGEFORMAT Figure 6.1).Figure STYLEREF 1 \s 6. SEQ Figure \* ARABIC \s 1 1 Flowchart of planned analysis method refinement: Simulations 7.1.IntroductionChapters 5 and 6 described data mining methodologies and pros and cons of these methods for analysing RA data such as CART, nearest neighbour classification, support vectors machines and logistic regression. The systematic review in Chapter 6 revealed that no research has been published comparing logistic regression and CART in the context of variable selection for RA data. This chapter will describe simulations conducted to explore how CART and logistic regression perform under differing scenarios with known data relationships with the intention of reducing the number of methods to apply to the real patient database as well as to indicate which analysis parameters to apply as illustrated in REF _Ref461551712 \h \* MERGEFORMAT Figure 7.1. The scenarios will be developed with known predetermined features, including but not limited to the characteristics of the data described in Chapter 4. Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 1 Flowchart of planned analysis method refinementThis chapter will describe methods for generating simulated datasets and steps taken to fine tune parameters of CART and logistic regression for use with a real patient dataset. 7.2.AimsThe aims of this chapter are:to describe the simulations performed; to identify how CART and logistic regression perform in predicting patient ACR response in an RA setting; to select a shortlist of methods and parameters to later apply to the available real patient RA clinical trial database.7.3.Methods for Generating Simulations The generation of simulations incorporates recommendations for conducting simulation studies proposed by Burton ADDIN EN.CITE <EndNote><Cite><Author>Burton A.</Author><Year>2006</Year><RecNum>121</RecNum><DisplayText>(Burton A., 2006)</DisplayText><record><rec-number>121</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1445434297">121</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Burton A., Altman D.G., Royston P., Holder, R.L. </author></authors></contributors><titles><title>The design of simulation studies in medical statistics</title><secondary-title>Statistics in medicine</secondary-title></titles><periodical><full-title>Statistics in medicine</full-title></periodical><pages>4279-92</pages><volume>25</volume><section>4279</section><dates><year>2006</year></dates><urls></urls></record></Cite></EndNote>(Burton A., 2006) and Smith ADDIN EN.CITE <EndNote><Cite><Author>Smith</Author><Year>2010</Year><RecNum>122</RecNum><DisplayText>(Smith, 2010)</DisplayText><record><rec-number>122</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1445434749">122</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Smith, M.K., Marshall, A</author></authors></contributors><titles><title>Importance of protocols for simulation studies in clinical drug development</title><secondary-title>Statistical Methods in Medical Research</secondary-title></titles><periodical><full-title>Statistical Methods in Medical Research</full-title></periodical><pages>1-12</pages><volume>00</volume><section>1</section><dates><year>2010</year></dates><urls></urls></record></Cite><Cite ExcludeAuth="1"><Author>Smith</Author><Year>2010</Year><RecNum>122</RecNum><record><rec-number>122</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1445434749">122</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Smith, M.K., Marshall, A</author></authors></contributors><titles><title>Importance of protocols for simulation studies in clinical drug development</title><secondary-title>Statistical Methods in Medical Research</secondary-title></titles><periodical><full-title>Statistical Methods in Medical Research</full-title></periodical><pages>1-12</pages><volume>00</volume><section>1</section><dates><year>2010</year></dates><urls></urls></record></Cite></EndNote>(Smith, 2010). The simulations follow a simulation protocol (Appendix C.1) which includes: Definition of aims and objectivesSimulation proceduresLevel of dependence between simulated datasetsSoftwareRandom number generatorSeed specificationData generation methodsScenarios (e.g. missing data structure)Statistical methods to be evaluatedNumber of simulations to perform (with justification)Performance criteriaPresentation of resultsThe aims and objectives given in the simulation protocol will identify how CART and logistic regression perform in predicting patient ACR20 response in an RA setting under differing scenarios, and to select a shortlist of methods to later apply to available real life RA clinical trial database. In the simulation protocol, the simulated datasets were planned to be independent via the application of different seeds being applied. The software used was specified as SAS version 9.4 and R-Studio installed on a UNIX environment on a HP Integrity rx7640 server. In chapter 4 the patient database was described. This represents over 11,000 adult patients diagnosed with RA from 18 randomised clinical trials in 4 development drug projects enrolled between 1998 and 2008. There are over a 100 explanatory variables available for inclusion in the analysis. Variables available include such patient disposition and demographic data such as country, gender, age, height, weight, BMI, race, and duration of disease. Included also are baseline assessments of ACR (Tender joint assessments, Swollen joint assessments, ESR, CRP, HAQ, pain score, patient global assessment, physician global assessment), SF36 and DAS28. Laboratory data from haematology, blood chemistry were also to be pooled. Simulations were generated to mimic this real patient database. Two types of variable will be generated in the simulated datasets: continuous and dichotomous. In this simulation, continuous variables will be generated from an underlying Normal distribution, using a formula developed by Frison and Pocock ADDIN EN.CITE <EndNote><Cite><Author>Frison</Author><Year>1997</Year><RecNum>407</RecNum><DisplayText>(Frison & Pocock, 1997)</DisplayText><record><rec-number>407</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1527062231">407</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Frison, L.J.</author><author>Pocock, S.J.</author></authors></contributors><titles><title>Linearly divergent treatment effects in clinical trials with repeated measures: efficient analysis using summary statistics</title><secondary-title>Stat Med</secondary-title></titles><periodical><full-title>Stat Med</full-title></periodical><pages>2855-2872</pages><volume>16</volume><dates><year>1997</year></dates><urls></urls></record></Cite></EndNote>(Frison & Pocock, 1997) and adapted by Julious ADDIN EN.CITE <EndNote><Cite><Author>Julious</Author><Year>2000</Year><RecNum>406</RecNum><DisplayText>(Julious, 2000)</DisplayText><record><rec-number>406</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1527061838">406</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Julious, S.</author></authors></contributors><titles><title>Letter to the Editor Repeated measures in clinical trials: analysis using mean summary statistics and its implications for design</title><secondary-title>Stat Med</secondary-title></titles><periodical><full-title>Stat Med</full-title></periodical><pages>3133-3135</pages><volume>19</volume><dates><year>2000</year></dates><urls></urls></record></Cite></EndNote>(Julious, 2000). This formula will generate independent variables with a specified correlation to a given dependent outcome. SAS Code for generating simulated data can be found in Appendix C.2.2. The clinical data contains a mixture of dichotomous, categorical and continuous data and is fully described in chapter 4. Simulated datasets were generated in stages:In the first stage, 100 continuous variables will be generated with different means and variances to reflect the real patient dataset as described in Chapter 4. Ten of these independent variables will be generated to have higher correlation with the response and 90 will be generated to have lower correlation with the response. These simulations are intended to see how well CART and logistic regression select explanatory variables from a large set. Since over 100 variables are available in the real patient database, this is considered sufficient and manageable in the context of multiple simulations. Following the first stage, based on observed results, the scope of simulations will be reduced. In the second stage half of the explanatory variables will be converted to dichotomous and categorical variables in order to reflect the different types of variable in the real patient database.In the third stage missing data will be introduced. Missing data will be generated missing completely at random whereby the rate of missing data will be generated randomly across all subjects and all explanatory variables. In addition, missing data will be generated by informed missing where the rate of missing data will still be generated randomly but the rate is dependent on the response variable. See chapter 4 for description of missing data in this real patient database.Full details of simulation criteria, including SAS and R code can be found in the Simulation Protocol (Appendix C.1).7.4.Simulations with continuous explanatory variables and no missing dataThe performance of CART and logistic regression in selecting the higher correlated variables will be assessed by calculation of variable selection sensitivity and specificity and presenting these paired values in Variable Selection (VaSe) plots. For this dissertation VaSe plots have been developed based on ROC curves. REF _Ref490998851 \h \* MERGEFORMAT Figure 7.2 depicts an example of a VaSe plot. This plot specifically simulates 100 patients, where the higher correlated variables are correlated at 0.90 with response. In this plot, points are plotted on scale where the x-axis display the probability of correctly excluding low correlated variables and the y-axis display probability of correctly selecting high correlated variables. When the method selects only the 10 higher correlated variables and none of the lower correlated variables the points will be plotted in the top left corner of the VaSe plots. In later sections of this chapter these plots will be presented in a trellis form with multiple choices for patient numbers and multiple values of high correlations.Please note, this plot is not an ROC curve for individual patient response prediction, but rather it measures the precision of CART and logistic regression in selecting variables that are correlated to patient response.Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 2 Example VaSe plot7.5.Simulations with continuous explanatory variables and no missing data7.5.1.Initial Simulations IntroductionThe first stage of the simulations will investigate how logistic regression and CART performs in selecting variables with a known correlation structure, under varying sample sizes, varying response rates, with continuous explanatory variables and no missing data.7.5.2.MethodsIn the first stage each simulation will generate a dichotomous response with 10 random Normal variables correlated with this response, and 90 random Normal variables not correlated with this response. Simulations of size 100, 1,000 and 10,000 will be generated in SAS 9.2. For both continuous and dichotomous data, a seed will be specified to enable reproducibility. The variables generated will be correlated in separate simulations at 0.90, 0.80. 0.70, 0.60, 0.30, 0.20, 0.10 and 0.0. The dichotomous response will be generated at 90%, 50% and 25% to approximately resemble response rates for ACR20, ACR50 and ACR70 respectively (see Section 4.7). A simulated dataset will be generated in SAS for each combination of sample size (3 options), response rate (3 options), higher correlation (4 options), and lower correlation (4 options) leading to 144 separate scenarios. Each scenario will be simulated 100 times, creating a total of 14,400 datasets to be applied to the 2 methods. The CART analysis will be performed using rpart package of the R software application ADDIN EN.CITE <EndNote><Cite><Author>Therneau</Author><Year>2017</Year><RecNum>183</RecNum><DisplayText>(Therneau et al., 2017)</DisplayText><record><rec-number>183</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502290933">183</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Therneau, T. </author><author>Atkinson, B.</author><author>Ripley, B.</author></authors></contributors><titles><title>rpart: Recursive Partitioning and Regression Trees. R package version 4.1-11</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2017</year></dates><urls><related-urls><url>;(Therneau et al., 2017). The R routine rpart, authored by Terry Therneau and Elizabeth Atkinson from the Mayo Foundation, will build classification and regression trees. Parameters selected in this R routine include: equal weights, ANOVA modelling, the retention of observations if one or more predictors are missing, equal costs, and a minimum of 20 observations per node in order for a split to be attempted. Stepwise logistic regression will be performed in SAS with the significance level for entry (SLE) PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+V2lsZDwvQXV0aG9yPjxZZWFy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ADDIN EN.CITE PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+V2lsZDwvQXV0aG9yPjxZZWFy
PjIwMDg8L1llYXI+PFJlY051bT4zMjE8L1JlY051bT48RGlzcGxheVRleHQ+KFdpbGQgZXQgYWwu
KTwvRGlzcGxheVRleHQ+PHJlY29yZD48cmVjLW51bWJlcj4zMjE8L3JlYy1udW1iZXI+PGZvcmVp
Z24ta2V5cz48a2V5IGFwcD0iRU4iIGRiLWlkPSI1YTk5d2RhYXp3ZHh0M2V4NXBmcHBhZTNmc3Bz
d3Z0ZnJ4cmQiIHRpbWVzdGFtcD0iMTUxOTkwNDQxNCI+MzIxPC9rZXk+PGtleSBhcHA9IkVOV2Vi
IiBkYi1pZD0iIj4wPC9rZXk+PC9mb3JlaWduLWtleXM+PHJlZi10eXBlIG5hbWU9IkpvdXJuYWwg
QXJ0aWNsZSI+MTc8L3JlZi10eXBlPjxjb250cmlidXRvcnM+PGF1dGhvcnM+PGF1dGhvcj5XaWxk
LCBOLjwvYXV0aG9yPjxhdXRob3I+S2FybCwgSi48L2F1dGhvcj48YXV0aG9yPkdydW5lcnQsIFYu
IFAuPC9hdXRob3I+PGF1dGhvcj5TY2htaXR0LCBSLiBJLjwvYXV0aG9yPjxhdXRob3I+R2FyY3ph
cmVrLCBVLjwvYXV0aG9yPjxhdXRob3I+S3JhdXNlLCBGLjwvYXV0aG9yPjxhdXRob3I+SGFzbGVy
LCBGLjwvYXV0aG9yPjxhdXRob3I+dmFuIFJpZWwsIFAuIEwuPC9hdXRob3I+PGF1dGhvcj5CYXll
ciwgUC4gTS48L2F1dGhvcj48YXV0aG9yPlRodW4sIE0uPC9hdXRob3I+PGF1dGhvcj5NYXR0ZXks
IEQuIEwuPC9hdXRob3I+PGF1dGhvcj5TaGFyaWYsIE0uPC9hdXRob3I+PGF1dGhvcj5ab2xnLCBX
LjwvYXV0aG9yPjwvYXV0aG9ycz48L2NvbnRyaWJ1dG9ycz48YXV0aC1hZGRyZXNzPkRlcGFydG1l
bnQgb2YgTWFya2VyIERpc2NvdmVyeS9Qcm90ZW9taWNzLCBSb2NoZSBQcm9mZXNzaW9uYWwgRGlh
Z25vc3RpY3MsIFJvY2hlIERpYWdub3N0aWNzIEdtYkgsIFBlbnpiZXJnLCBHZXJtYW55LiBub3Ji
ZXJ0LndpbGRAcm9jaGUuY29tPC9hdXRoLWFkZHJlc3M+PHRpdGxlcz48dGl0bGU+RGlhZ25vc2lz
IG9mIHJoZXVtYXRvaWQgYXJ0aHJpdGlzOiBtdWx0aXZhcmlhdGUgYW5hbHlzaXMgb2YgYmlvbWFy
a2VyczwvdGl0bGU+PHNlY29uZGFyeS10aXRsZT5CaW9tYXJrZXJzPC9zZWNvbmRhcnktdGl0bGU+
PC90aXRsZXM+PHBlcmlvZGljYWw+PGZ1bGwtdGl0bGU+QmlvbWFya2VyczwvZnVsbC10aXRsZT48
L3BlcmlvZGljYWw+PHBhZ2VzPjg4LTEwNTwvcGFnZXM+PHZvbHVtZT4xMzwvdm9sdW1lPjxudW1i
ZXI+MTwvbnVtYmVyPjxlZGl0aW9uPjIwMDgvMDEvMTI8L2VkaXRpb24+PGtleXdvcmRzPjxrZXl3
b3JkPkFudGlib2RpZXMsIEFudGludWNsZWFyLyphbmFseXNpczwva2V5d29yZD48a2V5d29yZD5B
cnRocml0aXMsIFJoZXVtYXRvaWQvKmRpYWdub3Npczwva2V5d29yZD48a2V5d29yZD5CaW9tYXJr
ZXJzLyphbmFseXNpczwva2V5d29yZD48a2V5d29yZD5DaXRydWxsaW5lL2ltbXVub2xvZ3k8L2tl
eXdvcmQ+PGtleXdvcmQ+RmVtYWxlPC9rZXl3b3JkPjxrZXl3b3JkPkh1bWFuczwva2V5d29yZD48
a2V5d29yZD5Mb25naXR1ZGluYWwgU3R1ZGllczwva2V5d29yZD48a2V5d29yZD5NYWxlPC9rZXl3
b3JkPjxrZXl3b3JkPk1pZGRsZSBBZ2VkPC9rZXl3b3JkPjxrZXl3b3JkPk11bHRpdmFyaWF0ZSBB
bmFseXNpczwva2V5d29yZD48a2V5d29yZD5QZXB0aWRlcywgQ3ljbGljLyppbW11bm9sb2d5PC9r
ZXl3b3JkPjxrZXl3b3JkPlByb3NwZWN0aXZlIFN0dWRpZXM8L2tleXdvcmQ+PGtleXdvcmQ+U2Vu
c2l0aXZpdHkgYW5kIFNwZWNpZmljaXR5PC9rZXl3b3JkPjxrZXl3b3JkPlNlcnVtIEFteWxvaWQg
QSBQcm90ZWluL2FuYWx5c2lzPC9rZXl3b3JkPjwva2V5d29yZHM+PGRhdGVzPjx5ZWFyPjIwMDg8
L3llYXI+PHB1Yi1kYXRlcz48ZGF0ZT5GZWI8L2RhdGU+PC9wdWItZGF0ZXM+PC9kYXRlcz48aXNi
bj4xMzU0LTc1MFggKFByaW50KSYjeEQ7MTM1NC03NTBYIChMaW5raW5nKTwvaXNibj48YWNjZXNz
aW9uLW51bT4xODE4ODcyNjwvYWNjZXNzaW9uLW51bT48dXJscz48cmVsYXRlZC11cmxzPjx1cmw+
aHR0cHM6Ly93d3cubmNiaS5ubG0ubmloLmdvdi9wdWJtZWQvMTgxODg3MjY8L3VybD48L3JlbGF0
ZWQtdXJscz48L3VybHM+PGVsZWN0cm9uaWMtcmVzb3VyY2UtbnVtPjEwLjEwODAvMTM1NDc1MDA3
MDE2Njk0MTA8L2VsZWN0cm9uaWMtcmVzb3VyY2UtbnVtPjwvcmVjb3JkPjwvQ2l0ZT48L0VuZE5v
dGU+AG==
ADDIN EN.CITE.DATA (Wild et al.) and significance level for staying (SLS) in the model set to the same value. This value will be varied and set to 0.30, 0.20, 0.10, 0.05, 0.01 and 0.001. Each value will be applied to stepwise logistic regression for each simulated dataset, for a total of 86,400 analyses on logistic regression and 14,400 analyses on CART.In addition, 4 simulations of missing data (Section 7.6) and a simulation with dichotomised data (Section 7.5) will be created for each of CART and logistic regression leading to a total of 1028 simulated datasets (Appendix C.3). REF _Ref433193217 \h \* MERGEFORMAT Figure 7.3 illustrates how simulated datasets will be generated. Each simulated dataset will be created in SAS using a macro called simmacro. The data for the simulation will be exported to an R instance where the package rpart will be applied to create a classification and regression tree. The nodes of the generated tree will then be exported back into the SAS environment. The same simulated data will also be used by SAS in Proc Logistic where a stepwise logistic regression will be applied, and the variables selected by the final model will be extracted. The specificity and sensitivity of logistic regression and CART for selecting the higher correlated variables will then be calculated. This procedure will be repeated 100 times and the resultant 100 specificities and 100 sensitivities for CART and logistic regression will be output to permanent datasets.Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 3 Generation of Simulated DatasetsThe main SAS program will utilise macros written for the dissertation in SAS to generate each simulation. For each dataset created, the SAS program will use a Unix utility to run R code within SAS for the CART analysis to identify CART selected nodes to compare with variables logistic regression identified in stepwise regression. The explanatory variables selected by CART and logistic regression will be displayed in VaSe plots. REF _Ref433654270 \h \* MERGEFORMAT Figure 7.4 displays logistic regression and CART in VaSe plots for the scenario where the response rate of the dependent variable is 50% and the lower correlated independent variables are simulated to have 0.0 correlation with the outcome. REF _Ref433654270 \h Figure 7.4 displays 12 separate VaSe plots. The columns are arranged by simulation size of 100, 1000 and 10,000 subjects. The rows represent the correlation of the higher correlated variables with correlation of 0.60 at the top, then 0.70, 0.80 and then 0.90 correlation on the bottom row. Within each VaSe plot the x-axis displays the probability of correctly excluding low correlated variables and the y-axis displays probability of correctly selecting high correlated variables. With respect to the data points within each VaSe plot, the CART results on 100 simulations are represented by blue circles and the logistic regressions on 100 simulations by red crosses.7.5.3.Results of Comparisons of CART versus Logistic Regression The simulations are computationally intensive. For the larger sample sizes (N=10,000) each stepwise logistic regression analysis alone can take 2 to 3 minutes, for a simulation of 100 repetitions this represents 200 to 300 minutes of processing time for a single scenario on logistic regression. The practical implications mean that each run of simulations can take 2 to 3 days of processing time, with frequent, unpredictable early terminations of the submitted program, due to memory limitations even on a Unix network. At the time of this research, better hardware was not available. This practical consideration limits the options to increase the number explanatory variables to generate beyond 100 or to increase the number of repetitions of each scenario beyond 100. As mentioned in section 7.5.2, the simulations comprise some 1028 datasets, creating 3,929,000 simulated patients with a total of 39,688,960,000 simulated data points (Appendix C.3).With respect to results, for simulations based on 100 patients there is clear separation on the VaSe plots of CART and Logistic Regression (first column of REF _Ref433654270 \h \* MERGEFORMAT Figure 7.4 to REF _Ref433654279 \h \* MERGEFORMAT Figure 7.7). Whilst both methods exhibit similar low correlated variable selection, CART displays superior high correlated variable selection (true positive rate). This is consistent for responses at 25%, 50% and 90% and for all combinations of correlation between response and high (0.9, 0.8, 0.7, 0.6 the rows of the trellis plots) and low (0, 0.1, 0.2, 0.3 figures 2, 3, 4, 5 respectively) correlated explanatory variables.For simulations based on 1,000 patients (second column of REF _Ref433654270 \h \* MERGEFORMAT Figure 7.4 to REF _Ref433654279 \h \* MERGEFORMAT Figure 7.7), CART tends to perform better, but different trends are expressed. Compared to 100 patient simulations both CART and logistic regression are increasingly sensitive but have poorer low correlated variable selection (false positive rate). In other words, for the 1,000 patient simulations the 2 methods are better at selecting the higher correlated variables, but also erroneously select a greater amount of lower correlated variables. There is also apparent, a gradual shift left to right (worsening of low correlated variable selection) as the correlation of highly correlated variables move from 0.9 to 0.6, this would be expected as high correlated variables would be less differentiated to lower correlated variable. This reduction of distinction between higher and lower correlated variables leads to a higher number of lower correlated variables being selected. Simulations based on 10,000 patients (third column of REF _Ref433654270 \h Figure 7.4 to REF _Ref433654279 \h Figure 7.7), are more pertinent to the overall aims of this research as they more closely match the numbers of patients available on the real patient clinical database. Compared to 1,000 patient simulations, the increased sample sizes lead to much more tightly clustered estimates of high and low correlated variable selection. The CART data has consistently almost perfect classification with 1.0 high correlated variable selection and low correlated variable selection of < 0.1. Whilst logistic regression matches CART for high correlated variable selection, low correlated variable selection varies across almost the whole range. As observed in the 1,000 patient based simulations, for the same reasons improvements in low correlated variable selection are seen with higher correlations in highly correlated variables. Also more apparent are improvements for low correlations of lowly correlated variables, this is consistent with expecting better results when correlations are more differentiated. Take for example the top right hand corner of REF _Ref433654270 \h \* MERGEFORMAT Figure 7.4. In this 10,000 patient simulation, 90 of the 100 variables are 0.20 correlated with the response and 10 are 0.60 correlated with the response. Both CART and logistic regression analyses select all 10 higher correlated variables as shown by 100% high correlated variable selection. However logistic regression selects many a greater number of lower correlated variables in the final model. Section 7.4.4 will explore ways to reduce this over-selection by logistic regression.Performance in all scenarios is similar between all response rates (25%, 50%, and 90%). Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 4 VaSe plot of logistic regression and CART for 50% response rate and zero correlation Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 5 VaSe plot of logistic regression and CART for 50% response rate and low correlation at 0.1 Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 6 VaSe plot of logistic regression and CART for 50% response rate and low correlation at 0.2 Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 7 VaSe plot of logistic regression and CART for 50% response rate and low correlation at 0.3 7.5.4.Improving the performance of Logistic RegressionAs described in Section 7.5.3, with increasing numbers of subjects in the simulations logistic regression improved in high correlated variable selection but performance worsened in low correlated variable selection. That is logistic regression tended to over-select variables in the final model. The comparisons of logistic regression with CART were based upon a stepwise logistic regression with significance levels for entry PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+V2lsZDwvQXV0aG9yPjxZZWFy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ADDIN EN.CITE PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+V2lsZDwvQXV0aG9yPjxZZWFy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ADDIN EN.CITE.DATA (Wild et al.) and staying (SLS) set at alpha = 0.30. This boundary was further explored at values of 0.2, 0.1, 0.05, 0.01 and 0.001 ( REF _Ref499900932 \h Figure 7.8, REF _Ref433656866 \h \* MERGEFORMAT Figure 7.9, REF _Ref433656869 \h \* MERGEFORMAT Figure 7.10, and REF _Ref433656871 \h \* MERGEFORMAT Figure 7.11). For each combination of response, high correlation and low correlation, no observable differences were seen for high correlated variable selection, however lower values of significance improved percent of low correlated variables exclusion. Across all scenarios a reduction in significance for entry and staying is associated with improvement in low correlated variable selection. So, as the criteria for selection in the model become stricter, the tendency for logistic regression to over-select is reduced. The impact on selection of high correlated variables is a more mixed picture. The changes in entry and staying significance values made little visible difference for 100 subject simulations (the first column in figures). One can see slight improvements in low correlated variable selection as significance decreases, but no improvements in high variable selection. For simulations based on 1,000 subjects, lower significance values are associated with improved low correlated variable selection as described above. However, there is not a consistent pattern with respect to high correlated variable selection. For high correlations at 0.60 (top row, middle column of figures), an entry and staying significance of 0.001 performs the best, however at other values of high correlation, a less strict significance of 0.01 often performs better for high correlated variable selection. This suggests that a too strict significance sometimes throws the baby out with the bathwater when excluding higher correlated variables. For lower values of significance and for simulations based on 10,000 patients (third column of REF _Ref433656857 \h \* MERGEFORMAT Figure 7.8 to REF _Ref433656871 \h \* MERGEFORMAT Figure 7.11), logistic regression competes well with the performance of CART. Also seen in these figures, high correlated variable selection was consistently high and stricter significance leads to improved low correlated variable selection. Where there is a greater separation in low and high correlated variables and for a significance of 0.001 ( REF _Ref433656857 \h \* MERGEFORMAT Figure 7.8 and last row of REF _Ref433656866 \h \* MERGEFORMAT Figure 7.9) logistic regression can achieve perfect variable selection. Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 8 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with zero correlationFigure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 9 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with low correlation at 0.1 Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 10 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with low correlation at 0.2 Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 11 Effect of changing entry and exit criteria in Logistic Regression for 50% Response with low correlation at 0.3 7.5.5.Initial Simulations ConclusionsWithin the parameters of this dissertation, the greater interest was in the performance of the 10,000 patient simulations since these are comparable to the number of subjects available for analysis in the real patient dataset described in Section 4. Thus next stages were to be conducted on these only. In Section 7.5.3, observed results were consistent between lower correlated values set at 0.10, 0.20 and 0.30 correlation. Similarly within each of REF _Ref433654270 \h \* MERGEFORMAT Figure 7.4 to REF _Ref433654279 \h \* MERGEFORMAT Figure 7.7 the higher correlations of 0.70 and 0.80 showed little differentiation. In REF _Ref433656857 \h Figure 7.8 to REF _Ref433656871 \h \* MERGEFORMAT Figure 7.11 for the 10,000 patient based simulations inform the reader that smaller values of SLS and SLE improve the performance of the logistic regression to be more comparable to CART. Thus further explorations of the performance of CART and logistic regression were performed on a more focussed selection of scenarios. For the investigation of categorical response in Section 7.6 and impact of missing values in Section 7.7, further simulations were based on a sample size of 10,000 patients, with lower correlated variables reduced to 0.0 and 0.3, with higher correlated variables reduced to 0.60 and 0.90. In addition, for further simulations, the values of SLS and SLE will be explored at 0.001 only.7.6.Simulations with continuous and categorical explanatory variables and no missing data7.6.1.Categorical and Continuous Simulations IntroductionIn Section 7.5, we have shown by using continuous explanatory variables that CART analysis tends to outperform logistic regression, particularly for simulations with a higher numbers of subjects. For these larger simulated datasets, logistic regression is more competitive with CART only when entry and staying significance set at stricter (lower) levels. However, the available real patient clinical database, as described in Chapter 4, contains both continuous and categorical independent variables, so in this section simulated datasets were adapted to reflect this. SAS code is included in Appendix C.2.7.6.2.MethodsIn the following analyses, five of the 10 higher correlated variables and 45 of the 90 lower correlated variables were converted to categorical values. Each of the 50 variables was split into equal dichotomous groups according to median value. Whilst this may not reflect all the possible permutations in the real dataset, some gross trends may become apparent with this simplest approach. Simulations of sample size 10,000, with entry and staying significance for logistic regression set at 0.001, were generated on this mixture of 50 continuous and 50 categorical variables, for 2 scenarios: low correlations at zero vs high correlations at 0.90 and low correlations at 0.30 versus high correlations at 0.60 as proposed in Section 7.4.7.6.3.Results REF _Ref461552007 \h \* MERGEFORMAT Figure 7.12 displays the results of simulations introducing dichotomous variables for 50% response. Trends were similar for 25% and 90% response. The second column shows the effect on logistic regression and CART by introducing categorical variables, compared with only continuous variables displayed in the first column. With the introduction of categorical variables, the performance of CART in selecting higher correlated variables deteriorated. In the first row of REF _Ref461552007 \h \* MERGEFORMAT Figure 7.12 where higher correlated variables are correlated at 0.90 with response and lower correlated variable correlated at 0.0, low correlated variable selection was maintained at similar levels for CART whilst high correlated variable selection declined indicating that CART was less efficient at selecting all higher correlated variables. Conversely in this scenario, logistic regression maintained high correlated variable selection whilst overcoming the previous tendency to over-select, to the level where logistic regression now outperforms CART. For CART, a similar effect is seen in the second row of REF _Ref461552007 \h \* MERGEFORMAT Figure 7.12 where the differentiation between low correlated variables and high correlated variables is less pronounced (0.30 and 0.60, respectively). Here low correlated variable selection is similar albeit slightly impaired, and high correlated variable selection deteriorates in a similar manner to the top row. Once again logistic regression improves markedly, particularly for high correlated variable selection but not as much as the first row and is not as competitive with CART. In both scenarios of correlations, the logistic regression results improved and the CART results appeared to worsen. Where correlations were more differentiated (0.00 vs 0.90 first row), logistic regression outperforms CART. Where the correlations were less differentiated CART was still a better performer although some high correlated variable selection was still natively impacted.Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 12 Effect of dichotomising explanatory variables in Logistic Regression vs CART for 50% Response L0.0/H0.9 – Low correlations at 0.0, high correlations at 0.9L0.3/H0.6 – Low correlations at 0.3, high correlations at 0.6Categorical and Continuous Simulations ConclusionsIn simulations with higher sample sizes CART out performs logistic regression where explanatory variables are continuous. However as categorical variables are introduced, logistic regression improves in low correlated variable selection whilst maintaining high correlated variable selection. We have seen that in some scenarios where there are no missing values, that logistic regression can outperform CART in variable selection. In the next section we will explore how the presence of missing values could impact variable selection performance. 7.7.Simulations with continuous explanatory variables and missing data7.7.1.Missing Data Simulations IntroductionIn Section 7.6 we observed how a change in characteristics of the explanatory variables by introducing categorical variables impacted the performance of CART and logistic regression. In clinical development in general and for RA trials specifically drug developers must expect missing data despite efforts to reduce the levels. The real patient clinical database described in chapter 4 contains different rates of missing values depending on the variable and also on the source clinical trial. In this section we will explore the impact of missing values. 7.7.2.MethodsTwo scenarios were selected for assessment of impact of missing values. Maintaining consistency with Section 7.5, sample size was set at 10,000 to reflect size of clinical database for use in later chapters. Response rate was set at 50% to reflect expected ACR20 response. Further, low correlation at 0.00 and high correlation at 0.90 were contrasted with low correlation of 0.30 and high correlation of 0.60 as these parameter selections exhibited different behaviours in Section 7.5.Two types of missing values were generated. Firstly, data missing completely at random were created, whereby any explanatory variables value could be missing at a rate of 1% and 0.1%. Thus for any subject none, one or more than one explanatory variable value could be missing. Secondly, data missing dependent on response was produced, such that data was missing at a higher rate when response was 0 (missing rate 0.002 and 0.02) compared to 1 (missing rate 0.001 and 0.01 respectively) to reflect the possibility that a patient may be more likely to have missing data if treatment response is poor. See Appendix C.2 for SAS code.7.7.3.ResultsIn SAS PROC Logistic any observation with missing values for the response, strata, or explanatory variables is excluded (listwise deletion) from the analysis, although for categorical explanatory variables, missing can be defined as an extra category.For missingatrandom data, in a single simulation, a low rate of missing values for each variable of 0.1% excluded 959 and 985 observations out of 10,000 (approximately 10%) in the 2 sets of parameter selections. Similarly, at a higher rate of missing values of 1% a single simulation excluded 6319 and 6354 observations out of 10,000 (approximately 63%).For not missingatrandom data, a single simulation, a low rate of missing values of 0.1%, excluded 952 and 927 observations out of 10,000 (approximately 10%) in the 2 sets of parameter selections. Similarly, at a higher rate of missing values of 1% a single simulation excluded 6351 and 6330 observations out of 10,000 (approximately 63%). Conversely CART does not exclude subjects due to missing values. In a single simulated dataset, where the rate of missing values is 1% CART recognises up to 97 and 112 values missing when calculating node splits. Similarly, where the rate of missing values is 0.1% CART recognises up to 10 and 15 values missing when calculating node splits. CART recognises the missing values by variable rather than by observation.So whether missingatrandom or not missingatrandom, logistic regression excludes a large proportion of data when rates of missing are at 1% or more. The risk of excluding data is magnified when including more explanatory variables that are expected to have missing values. Results of effects of rate of missing values, where missing is completely at random amongst the explanatory variables, are seen in REF _Ref461552258 \h \* MERGEFORMAT Figure 7.13. The first column displays the high correlated variable selection and low correlated variable selection where no missing values are present. The second and third columns display the effect of 0.1% and 1% missing values respectively. For missing-completelyatrandom, it can be seen that as the rate of missing changes, CART performance changes very little however the performance of logistic regression improves particularly with respect to low correlated variable selection. Considering that for logistic regression missing values reduces the effective sample size, this impact is not surprising and is consistent with logistic regression performance observed in REF _Ref433654270 \h \* MERGEFORMAT Figure 7.4 to REF _Ref433654279 \h \* MERGEFORMAT Figure 7.7 in Section 7.5 where smaller sample sizes had improved low correlated variable selection. This may be explained by the expectation that lower correlated variables less likely to meet the criteria for entry in smaller datasets.Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 13 VaSe plot of logistic regression and CART by rate of missing values none, 0.1% and 1% (missing-completely-at-random) L0.0/H0.9 – Low correlations at 0.0, high correlations at 0.9L0.3/H0.6 – Low correlations at 0.3, high correlations at 0.6Results of effects of rate of missing values, where missing is not at random amongst the explanatory variables, are seen in REF _Ref490052659 \h \* MERGEFORMAT Figure 7.14. For lower missing rates, the behaviour is much the same as missingatrandom. However, for a larger rate of missing values the CART performance again begins to decline in both high and low correlated variable selection, whilst logistic regression improves in high correlated variable selection. Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 14 VaSe plot of logistic regression and CART by rate of missing values none, 0.001 and 0.01 (missing not at random) L0.0/H0.9 – Low correlations at 0.0, high correlations at 0.9L0.3/H0.6 – Low correlations at 0.3, high correlations at 0.67.7.4. Missing Data Simulations ConclusionsIn this section we have seen how CART and logistic regression manage missing values. The biggest impact is how logistic regression manages missing values by excluding a whole observation if one or more observations are missing. Since a missing value could occur in any of the 100 explanatory variables, with just 1% missing values throughout the simulated dataset, as much as 63% of the data could be excluded. CART on the other hand still utilises data from non-missing variables even if an observation contains variables with missing values. 7.8.SummaryIn this chapter we have seen the introduction of variable selection (VaSe) plots (Section 7.4). These have been developed in order to visually establish the performance of competing multivariate methods to be able to select highly correlated variables for a large complex dataset. Through the use of VaSe plots we have seen throughout this chapter that within the scope of the simulations it seems clear that CART outperforms logistic regression in most scenarios, as illustrated by the development and introduction of the VaSe plots. Section 7.5 simulated an ideal situation where there are no missing values and all explanatory variables are continuous. Throughout 144 variations of this scenario CART clearly outperforms logistic regression, particularly in the case of high subject numbers (n=10,000). In Section 7.5.4 we saw that we can improve the performance of logistic regression by selecting significance levels for entry and staying set at low values. In the simulations with high subject numbers, a significance of 0.001 seemed to deliver the most comparative performance with CART.Section 7.5 also allowed for the refining down of scenarios to employ for further simulations. Results were consistent between the different outcomes of the response variable (25%, 50% and 90% response). When varying lower correlations at 0.10, 0.20 and 0.30 homogeneity was observed in performance as was by varying higher correlations between 0.70, 0.80, and 0.90. Further investigations were performed on 50% response, sample size 10,000, significance 0.001 on 2 scenarios: low correlation 0.0, high correlation 0.9low correlation 0.3, high correlation 0.6In Section 7.5.5 these scenarios gave the greatest differentiation in performance of CART and logistic regression and were used in Sections 7.6 and 7.7.Section 7.6 introduced dichotomous variables to the simulations. Half of both the higher and lower correlated explanatory variables were dichotomised. Here the arguably more realistic scenario where explanatory variables can be continuous or dichotomous, CART maintained low correlated variable selection but worsened in high correlated variable selection. Logistic regression improved dramatically in low correlated variable selection and seemed to outperform CART where the difference in lower and higher correlations in explanatory was most extreme. In perhaps a more representative situation with less difference in correlations, CART was still superior. Section 7.7 introduced missing values to the simulations. Two types of missing data were explored: missing-completelyatrandom, and missingnotatrandom. For missing completely-atrandom, missing values were assigned at random to the 100 explanatory variables and to each subject, at a missing rate of 1% and 0.1%. So any explanatory variable or subject could have none, one, or more than one missing value. For missingnotatrandom the missing value rate was assigned according to level of response variable, missing value rate was doubled when response was 0, compared to a response of 1.Exploring higher rates of missing data proved prohibitive by the way logistic regression in SAS handles missing data. SAS excludes the entire subject from the analysis if one or more response, strata or explanatory variable contains a missing value. At a rate of just 1% missing data, logistic regression excluded more than 60% of the data. CART excluded just 1% or 0.1% according to given missing data rate. In terms of performance, in the presence of missingcompletely-atrandom data, CART changed very little. Much like phenomenon seen in adding categorical variables, logistic regression improved in low correlated variable selection, particularly in the more extreme 0.0 vs 0.9 correlations. For missingnotatrandom, logistic regression acted similarly to missingatrandom, but CART loses some performance in both high and low correlated variable selection. CART still outperforms logistic regression in more realistic 0.3 vs 0.6 scenario.The handling of missing values by logistic regression would seem to be the most significant disadvantage of the method for use in a RA dataset. Particularly since, as we saw from Chapter 4, large amounts missing values are expected in datasets of this source. The ability to manage missing values and the pairwise deletion would make this the most important consideration for logistic regression as a selection method.In the simulations in this chapter we have seen that CART usually outperforms logistic regression in variable selection. The few scenarios where logistic regression outperforms CART are those that are least likely such as where the lower correlated variable are at a minimum of 0.0 correlation with the response and the higher correlated variables are correlated at a maximum of 0.90 with the response.7.9.LimitationsThe simulations performed in this chapter have not been exhaustive. We have not explored the combination of categorical and missing values, or non-binary categorisation of explanatory variables and we have not explored correlations between explanatory variables. Neither have we explored compensating for the way SAS Proc Logistic manages missing data by nominating missing data as a class variable: in real practice we wouldn’t be able to assume data is missing for the same reason. In order to keep number of simulations manageable we have not considered nonNormal continuous data. Nonetheless, in the scenarios explored, CART has been robust in performance in both high and low correlated variable selection compared to logistic regression. Whilst this staged approach intends to mimic aspects of the real patient database, it is recognised that it may be limited in expressing the nuances in such a large database and complex disease area. For example, the correlations explored in the simulations go beyond that seen in the real patient database. Also of note was that higher correlated variables tended to be correlated highly with each other and lower correlated variables also highly correlated with each other. Missing data was generated either missing completely at random or missing dependent on outcome response. No account was taken with respect to other independent variables when generating missing values. In reality a missing record of say tender joint count may well be associated with swollen joint count, and also if one value is present of the 2 then one would expect a high correlation between these two data-points.These simulations can only inform an approach to analysing the real patient data rather than definitively compare these methods in a real life data setting. As such conclusions drawn may only be valid for the situations explored. However, despite these limitations, CART appeared more robust in its performance compared to logistic regression.Nonetheless this chapter has shown the value of undertaking simulations. For example, for large datasets we have seen that by applying a very strict entry and staying criteria for the logistic regression we can improve performance in terms of high and low correlated variable selection. 7.10RecommendationsIn the next chapter we will use CART to analyse the real patient RA clinical trial data. In simulations, logistic regression performed well in most of the criteria for method selection ( REF _Ref490052717 \h \* MERGEFORMAT Table 7.1), however it became clear that the handling of missing values became too large an issue to overcome ( REF _Ref490052763 \h \* MERGEFORMAT Figure 7.15). Therefore, my assessment for handling of missing values was changed from fair to poor. In contrast, CART handled missing values with ease and also pruning of trees produced interpretable trees with high performance. As such, I upgraded missing value and interpretability ratings to good on the basis of the simulations. In the analysis of the real patient data, CART will be applied and additional data mining methodologies such as cross validation as described in Chapter 5 will be considered to maximise validity of findings. Table STYLEREF 1 \s 7. SEQ Table \* ARABIC \s 1 1 Some characteristics of CART and Logistic RegressionCharacteristicCARTLogistic RegressionCan handle ‘mixed’ dataGoodGoodHandling of missing valuesGoodPoorRobust to outliersGoodFairComputational Scalability (Large N)GoodGoodInterpretabilityGoodGoodCan manage high dimensionalityGoodGoodWe will apply CART to the data as a whole and also explore heterogeneity by comparing subgroups of data such as drug compound, dates study was conducted, geographical region of patients. Since CART seemed to be superior without the enhancements of boosting, bagging and random forest, the method will be applied first without these adaptations. Figure STYLEREF 1 \s 7. SEQ Figure \* ARABIC \s 1 15 Flowchart of analysis method refinement: Application of CART using Clinical Trial Data 8.1.IntroductionIn chapter 2 we depicted RA as a disease. We described the epidemiology, postulated causes of RA and diagnosis. Chapter 2 went on lay out the manifestations and co-morbidities of RA, as well as pathophysiology and risk factors for incidence and severity. Some of the hypothesised factors believed to affect occurrence and severity included age, gender, smoking, ethnicity, genetics, diet and socioeconomic status. Since the data for this thesis is generated from clinical trials at Roche, chapter 3 illustrated how phase III clinical trials are typically designed at Roche. This chapter details endpoints, trial designs and regulatory strategies that are acceptable for gaining approval for new therapies in RA.Chapter 4 follows on from chapter 3, by summarising the clinical trial data available for analysis in this research. These data are sourced from 16 phase 3 clinical trials conducted to support 4 separate drug development projects. The trials enrolled 11,705 adult patients diagnosed with RA over a period from 1998 to 2008. This chapter goes into some detail to convey which variables are available to predict outcome and the numbers of patients with those data. We also explore the amount of missing data and some correlations and associations with study endpoints. This information is later used to inform simulations.In this research we intended to compare multivariate and data mining methods for predicting patient response in RA patients. Chapter 5 describes a selection of data mining methodologies that are being considered for application to the available data, and through reflection, reduce the number of methods that will be explored in the later simulations. Methods for logistic regression and CART included some enhancements to improve performance; Lasso and ridge regression for logistic regression, boosting, bagging and random forest for CART.Through systematic reviews, chapter 6 identified biological factors that influence progression of the disease and to summarise methods and results that are assessed in a multivariate analysis. Further systematic reviews identified which data mining methods have been applied in RA. By the end of chapter 6, by understanding the characteristics of the data available to us for analysis and evaluating benefits and limitations of each method, we were able to reduce the number of statistical methods under consideration to logistic regression and CART. The conclusions of chapter 6 to progress with 2 methods, along with the characteristics of the data described on chapter 4, were used to inform the simulations conducted in chapter 7. The simulations conducted explored how CART and logistic regression perform under differing scenarios with known data relationships with the intention of identifying preferred methods to apply to the real patient database. Simulations tested how well each method selected variables with known correlation with outcome. Simulations were repeated under differing sample sizes, correlation levels, missing value rates, and variable type. This chapter concluded that within the scope of the simulations CART outperformed logistic regression in most scenarios, and was selected as the primary analysis method to apply to the clinical trial dataset.8.2.AimsThe aims of this chapter are to apply and evaluate CART analyses for prediction of patient response in the RA clinical trial database. 8.3.MethodsThe CART procedure from the R routine ‘rpart’ ADDIN EN.CITE <EndNote><Cite><Author>Therneau</Author><Year>2017</Year><RecNum>183</RecNum><DisplayText>(Therneau et al., 2017)</DisplayText><record><rec-number>183</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502290933">183</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Therneau, T. </author><author>Atkinson, B.</author><author>Ripley, B.</author></authors></contributors><titles><title>rpart: Recursive Partitioning and Regression Trees. R package version 4.1-11</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2017</year></dates><urls><related-urls><url>;(Therneau et al., 2017) will be applied to these data. In order to control for over-fitting a cross validation procedure will be used whereby 75% of the data will be randomly selected and used to build the training set. The CART model will be created from the training set. The remaining 25% of the data (test set) will be used measure the performance of the CART model; sensitivity, specificity and accuracy will be calculated. A receiver operating characteristic curve will be generated by adjusting the loss function in the CART procedure. Further exploratory analyses will be applied to explore the robustness of results through applying boosting, bagging and random forest enhancements to CART, as well as attempting to apply logistic regression to confirm choice of CART as a superior methodology in the context of these data. Random Forest methodology will also be applied using the R routine ‘randomForestSRC’ ADDIN EN.CITE <EndNote><Cite><Author>Ishwaran</Author><Year>2017</Year><RecNum>185</RecNum><DisplayText>(Ishwaran & Kogalur, 2017)</DisplayText><record><rec-number>185</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502291610">185</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Ishwaran, H.</author><author>Kogalur, U.</author></authors></contributors><titles><title>randomForestSRC: Random Forests for Survival, Regression, and Classification (RF-SRC). R package version 2.5.0</title></titles><dates><year>2017</year></dates><urls><related-urls><url>;(Ishwaran & Kogalur, 2017). Boosting and Bagging will utilise the R routine ‘adabag’ ADDIN EN.CITE <EndNote><Cite><Author>Alfaro</Author><Year>2015</Year><RecNum>184</RecNum><DisplayText>(Alfaro et al., 2015)</DisplayText><record><rec-number>184</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502291217">184</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Alfaro, E.</author><author>Gamez, M.</author><author>Garcia, N</author></authors></contributors><titles><title>adabag: Applies Multiclass AdaBoost.M1, SAMME and Bagging: R package version 4.1</title></titles><dates><year>2015</year></dates><urls><related-urls><url>;(Alfaro et al., 2015).8.4.Primary Analysis Results REF _Ref490053181 \h \* MERGEFORMAT Figure 8.1 shows the results of the CART analysis on the RA clinical trial dataset. See Appendix D.1 for detailed ‘rpart’ output. In this pruned tree a patient is predicted to be a responder via two routes: a baseline tender joint count of ≥ 7.5 and region not North America and Joint Space Narrowing Score < 0.49.a baseline tender joint count of ≥ 7.5 and region not North America andJoint Space Narrowing Score ≥ 0.49 andNumber of Previous DMARDs < 3.5 and Race not equal to Asian, Other or Unknown.Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 1 CART primary analysis REF _Ref490054590 \h \* MERGEFORMAT Table 8.1 shows the results of applying the test dataset to the CART model, the test data comprised 25% of the available data (n=2848) as described in Section 8.3. The model performed poorly in identifying those patients who responded to treatment with sensitivity of 0.41 (522/1279), although it was better at recognising non-responders with specificity 0.76 (1200/1569). Overall accuracy was still low at 0.60 (?(1200+522)/2848 ). Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 1 CART prediction results PredictedTotalNon-ResponderResponderActualNon-Responder12003691569Responder7575221279Total19578912848Sensitivity0.41Specificity0.76Accuracy0.60Table 8.2 describes the surrogates selected in the CART analysis. Details of the surrogate variables and analysis can be found in Appendix D.1. One can see for example that Baseline Swollen Joint Count can be substituted for Baseline Tender Joint Count when the latter has a missing assessment. Baseline Swollen Joint Count will agree with Baseline Tender Joint Count 91.2% of the time. For other nodes of primary analysis CART, although surrogates have been identified, they do not have the as good levels of agreement, and perhaps the medical rational is not so clear. For example, Region (defined as North America vs Non North America), can be replaced with surrogates such as History of Psychiatric, Immune, General, CNS disorders, or weight.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 2 CART prediction results Primary SplitSurrogate SplitsAgreementBaseline Tender Joint CountBaseline Swollen Joint Count0.912RegionHistory of Psychiatric Disorders 0.684History of Immune Disorders0.663Weight0.662History of General Disorders0.653History of Nervous System and Neurological Disorders0.650JSN ScoreRA Duration0.797Age0.760Race0.759Number of DMARDsBaseline CRP0.866RaceBaseline CRP0.691SF36 Mental Component Score0.654SF36 Physical Component Score0.653Red Blood Cell Count0.651SF36 General Health0.641To further explore the characteristics of the CART analysis a loss function operating characteristic curve was generated by adjusting the loss function in the CART procedure ( REF _Ref490054620 \h \* MERGEFORMAT Table 8.3 and REF _Ref490054639 \h \* MERGEFORMAT Figure 8.2). A loss function was applied where the loss due to misclassifying a non-responder ranged from 0.1 to 2. That is, a loss function of 2 assumes that the cost of misclassifying a non-responder as a responder was twice as high as that of misclassifying a responder as a non-responder. In REF _Ref490054639 \h \* MERGEFORMAT Figure 8.2, lower loss functions appear in top right corner and increase in value towards bottom left corner. The accuracy for all CART analyses with loss functions ranged from 0.49 to 0.60, with the highest accuracy relating to analysis with equal loss, this is the point at the furthest distance to the solid red diagonal line and illustrated by the green dashed line on REF _Ref490054639 \h \* MERGEFORMAT Figure 8.2. Adjusting the loss function did not seem to improve overall accuracy of the prediction. Extreme values for the loss function improve sensitivity or specificity, but at the cost to accuracy.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 3 Impact on accuracy of adapting loss functionLoss FunctionSensitivitySpecificityAccuracy0.10.011.000.560.30.001.000.550.60.120.950.570.90.280.850.591.0 (equal losses)0.400.760.601.10.520.660.601.20.640.530.581.30.720.460.581.40.730.470.591.50.720.460.581.60.800.370.571.70.820.350.561.80.830.360.571.90.820.350.562.00.970.100.49Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 2 Loss function Operating Characteristic Curve51498515430500514350152400 8.5.Additional analyses on primary endpointIn Section 5.3 we described methods that can improve the performance of CART, namely boosting, bagging and random forests. Bagging or bootstrap aggregation generates many trees by constructing bootstrap samples with replacement from the available data, a prediction method is applied to each bootstrap sample, simple voting determines classification. Boosting, like bagging, is an approach that can be used to combine multiple trees these into a single tree. As with bagging this method greatly reduces the variance of the estimated prediction. Random Forests select a random set of features on which to create each bootstrapped tree. This can result in a reduction of variance of the estimated prediction compared with boosting alone.In the following section we will also apply k-fold cross validation to explore whether any improvement to prediction can be observed. Logistic regression will also be implemented to confirm whether the choice made following simulations in Chapter 7 was justified. 8.5.1.CART analysis with boosting & baggingFor the purposes of exploratory analyses, boosting and bagging methods were applied to CART. The R package ‘adabag’ ADDIN EN.CITE <EndNote><Cite><Author>Alfaro</Author><Year>2015</Year><RecNum>184</RecNum><DisplayText>(Alfaro et al., 2015)</DisplayText><record><rec-number>184</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502291217">184</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Alfaro, E.</author><author>Gamez, M.</author><author>Garcia, N</author></authors></contributors><titles><title>adabag: Applies Multiclass AdaBoost.M1, SAMME and Bagging: R package version 4.1</title></titles><dates><year>2015</year></dates><urls><related-urls><url>;(Alfaro et al., 2015), was used for boosting and bagging analysis.The prediction results of boosting and bagging are shown in REF _Ref490054748 \h \* MERGEFORMAT Table 8.4 and REF _Ref490054754 \h \* MERGEFORMAT Table 8.5. Overall accuracy remains similar to results seen in primary CART and there is a little fluctuation in sensitivity and specificity. Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 4 Boosting prediction resultsPredictedTotalNon-ResponderResponderActualNon-Responder11284411569Responder6456341279Total177310752848Sensitivity0.50Specificity0.71Accuracy0.62Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 5 Bagging prediction results PredictedTotalNon-ResponderResponderActualNon-Responder11883811569Responder7225571279Total19109382848Sensitivity0.44Specificity0.76Accuracy0.61Since these methods generate and then aggregate multiple trees, no tree chart is available, but comparative variable importance is shown in REF _Ref490054721 \h \* MERGEFORMAT Figure 8.3 and REF _Ref490054728 \h \* MERGEFORMAT Figure 8.4. Variable importance is computed from the number of times each variable is selected as a split for any tree. Baseline tender joint count is the most important variable in both methods.Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 3 Boosting Variable ImportanceFigure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 4 Bagging Variable Importance8.5.2.CART analysis with Random ForestThe primary CART analysis was repeated by applying the Random Forest R routine (‘randomForestSRC’), the results are shown in REF _Ref490054777 \h \* MERGEFORMAT Table 8.6. Compared with CART, Random Forest prediction improved sensitivity but specificity worsened and overall accuracy was similar. REF _Ref490054810 \h \* MERGEFORMAT Figure 8.5 shows the variables with the highest relative importance in the random forest. Both CART and Random Forest select the variables: Baseline Tender Joint Count, Region, Joint Space Narrowing Score and Race. Random Forest selects some additional variables related to X-Ray assessments (Sharp Score and Erosion Score), likely due to high correlation with Joint Space Narrowing Score ( REF _Ref490054789 \h \* MERGEFORMAT Table 8.7). Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 6 Random Forest prediction results on all dataPredictedTotalNon-ResponderResponderActualNon-Responder13042651569Responder8114681279Total21157332848Sensitivity0.64Specificity0.62Accuracy0.62Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 5 Random Forest Variable Importance Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 7 Correlations between X-Ray assessmentsPearson Correlation/Spearman Correlation/NJoint Space narrowingSharp ScoreErosion Score Joint Space narrowing 1.001.0066020.960.9657990.860.825799Sharp Score1.001.0057990.970.955799Erosion Score1.001.0057998.5.3.k-fold cross validationIn Section 5.8 we described a common method of cross validation: K-fold cross validation. In this method the training and validation datasets are combined and then split into k roughly equal parts. Then for each k=1,2,3…k, your model is fitted using the kth partition as validation set and the remaining k-1 partitions as training sets. The literature suggests that a value of k =10 provides a least biased estimation of the accuracy. When 10-fold cross-validation is applied to the primary CART analysis, accuracy remains almost unchanged at 0.61.8.5.4.Logistic regression analysis In Chapters 5, and 6, we considered logistic regression as a potential method before finally choosing CART over logistic regression following simulations in Chapter 7. In order to confirm our choice of method, a forward stepwise logistic regression was run on all the data. Since every patient had at least one missing value in at least one variable, all data were deleted from the analysis. In a further exploratory analysis the variables identified by the primary CART analysis were applied in a logistic regression analysis. The logistic therefore considered the following dependent variables: Baseline Tender Joint CountRegion (North America or Non North America)Joint Space Narrowing ScoreNumber of Previous DMARDsRaceWith just these 5 variables only 4375 of 11705 patients (37%) were available for logistic regression analysis due to pairwise deletion in the presence of missing data. The data set was split into training and test data sets (ratio 75%: 25% respectively), with the model estimates generated from the training set and prediction results reported from fitting the test data. In the logistic regression, all dependent variables were significant in the model (p<.0001) apart from Baseline Tender Joint Count (p=0.8714). Once again, the prediction was very similar to the CART primary analysis ( REF _Ref490054978 \h \* MERGEFORMAT Table 8.8) (0.60 for CART vs 0.59 for logistic regression), and the area under the curve for the ROC was low at 0.6355 ( REF _Ref490054999 \h \* MERGEFORMAT Figure 8.6). Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 8 Logistic Regression prediction results on CART final modelPredictedTotalNon-ResponderResponderActualNon-Responder324220544Responder243350593Total5675701137Sensitivity0.61Specificity0.57Accuracy0.59Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 6 Receiver operating characteristic curve for logistic regression 8.5.5.CART analysis on completers For the purposes of comparison, a CART analysis was performed on the set of patients analysable in the logistic regression of section 8.5.4. That is patients for whom ACR20, baseline tender joint count, region (North America or Non North America), joint space narrowing score, number of previous DMARDs and race were not missing. The slight variation in overall totals is due to different seeds for random split of training and test sets between SAS and R and the subsequent numbers of patients deleted in the test sets. Results of CART completers analysis are shown in REF _Ref515976646 \h Table 8.9 and REF _Ref515976752 \h Figure 8.7. The sensitivity, specificity and accuracy of the predictions mirror those of the logistic regression, although the predictors are much fewer with just region and joint space narrowing score selected by the model.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 9 CART completers prediction resultsPredictedTotalNon-ResponderResponderActualNon-Responder293223516Responder223355578Total5165781094Sensitivity0.61Specificity0.57Accuracy0.59Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 7 CART completers analysis8.6.Exploratory Analysis 8.6.1.CART analysis on subgroups REF _Ref490055035 \h \* MERGEFORMAT Table 8.10 displays results of subgroup analyses performed to explore where prediction could be improved. Subgroup analyses were generated on each project separately as well as patients that had radiographic assessments and on patients on methotrexate control (MTX). Since smoking and Rheumatoid Factor were identified in Section 2.9 as risk factors for RA, the subgroup analyses were repeated on the group of patients where these assessments were available. Finally, further subgroup analyses based on groups of patients identified in the primary analysis were also explored.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 10 CART and Random Forest Subgroup AnalysesSubgroupTrainNACR20ResponseCART SensitivityCART SpecificityCART AccuracyRandom Forest AccuracyAll patientsAll patients85560.440.410.760.600.62X-ray patients48490.430.570.650.620.66Project A31650.470.650.530.580.66Project M17370.530.520.580.550.61Project O20840.490.590.570.580.59Project T15700.24na*na*na*0.73MTX patients26300.320.090.960.670.69Smoking Status not missingAll patients85020.440.440.740.610.63X-ray patients 48490.430.440.770.630.66Project A 31650.470.580.590.580.62Project M 17370.530.500.640.570.62Project O 20840.490.540.640.590.60Project T 12890.24na*na*na*0.73MTX patients26300.320.090.960.670.70RF not missingAll patients31300.530.680.490.580.62X-ray 23110.570.660.540.610.62Project A** 0nananananaProject M 10810.580.730.380.60.65Project O20490.500.560.640.60.6Project T**0nananananaMTX patients9960.400.340.840.650.65Other SubgroupsJSN Score < 0.59840.590.790.340.64ndBTJC < 7.58190.210.120.920.77ndBTJC <7.5 or JSN Score < 0.517640.420.740.720.73ndRF= Rheumatoid Factor, JSN=Joint Space Narrowing, BTJC=Baseline Tender Joint Count* Tree not computable**Numeric RF not recordedThe primary CART analysis yielded a predictive accuracy of just 0.60 (marginally increased to 0.62 when applying random forest methodology). In addition to the primary analysis, CART was applied to each project separately. Whilst there were some variations in sensitivity and specificity, overall accuracy was poorer on projects A, M and O. A CART was not computable for project T as complexity parameter threshold could not be achieved, however the random forest led to an improved accuracy over primary analysis (0.73 vs 0.62). Project T is the oldest project, recruiting patients from 1998. It is also the project with the lowest ACR20 response (approximately 24%). In addition to looking patients project by project, CART was also applied to those patients on the control arm (Methotrexate background therapy) and on patients with x-ray evaluations. The overall accuracy for patients on methotrexate control arms was higher than the primary CART analysis but this was driven by high specificity at the expense of low sensitivity meaning that the analysis seems to over-select non-responders. Of the patients with x-ray evaluations of joints, there was little improvement in prediction accuracy. In the description of the demographic data in Section 4.9, we can see that approximately 29% of patients do not have smoking data, however Section 2.9 discusses literature postulating the association between RA and cigarette smoking. Therefore, the subset of patients with data on smoking status were analysed as a subset. For this subset of patients, results were very similar across all available patients, by project and on methotrexate and x-ray patients. Similarly, Section 2.8 describes how patients with active RA have significantly higher serum rheumatoid factor (RF) levels compared with those with inactive disease, and Section 4.8 shows 74% of patients without a numeric RF value. In all patients and subsets of patients no significant improvement was seen in prediction. Based on CART primary analysis results, some small subgroups were explored. i.e. where BTJC <7.5 or JSN Score < 0.5. Some bigger improvements to overall accuracy of prediction could be seen (0.73, 0.77) but in a very small number of patients. In most subgroups there was no significant improvement in overall accuracy, either from CART analysis or random forest. In the instances where accuracy reached 0.70 or more, this came at cost to sensitivity, i.e. the model selected fewer patients overall as responders.8.6.2.CART analysis exploring study specific variables The intent of this research was to identify biomarkers and baseline patient assessments that predict patient outcome for disease progression, and from these results to improve study design in RA clinical trials through patient selection. Therefore, study specific identifiers such as compound, protocol, drug were omitted in order to identify a set of biomarkers that might have a general application. REF _Ref514917727 \h Table 8.11 and REF _Ref515458861 \h Figure 8.8 show the results of including study specific variables. Compound, and specifically Compound T, was identified as a splitter in the first node of this tree, followed by drug type (MTX, Placebo vs Active treatment). This is perhaps not surprising where the development programs feature potent medicines, however this may have limited utility in predicting response in a new or general setting. With these parameters the sensitivity increased (0.41 to 0.56) but there was little change in specificity (0.76 to 0.73) and still accuracy remained modest (0.60 to 0.65).Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 11 CART prediction results including study parameters PredictedTotalNon-ResponderResponderActualNon-Responder11404291569Responder5587211279Total169811502848Sensitivity0.56Specificity0.73Accuracy0.65Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 8 CART including study parametersGiven the impact of Project T in the previous analysis, the primary analysis was rerun excluding this project. Results can be seen in REF _Ref514919578 \h Table 8.12 and REF _Ref514919637 \h Figure 8.9. Here accuracy remains unchanged and there is an improvement in sensitivity balanced with a worsening of specificity.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 12 CART prediction excluding Project T PredictedTotalNon-ResponderResponderActualNon-Responder7304571187Responder4786621140Total120811192327Sensitivity0.58Specificity0.61Accuracy0.60Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 9 CART excluding Project T8.6.3.CART analysis on ACR componentsThe primary endpoint of this research is ACR20. As described in Section 3.3.1 the ACR20 is defined as a 20% reduction in: the number of swollen joints on a 68-joint count, the number of tender joints on a 66-joint count and in 3 of the 5 following criteria: patient’s assessment of pain, patient’s global assessment of disease activity, physician’s global assessment of disease activity, patient’s assessment of physical function and a markers of inflammation (i.e. ESR or CRP). Since the prediction for ACR20 was less than desirable, the ability to predict each of these 8 components as 20% response variables was investigated. REF _Ref490055075 \h \* MERGEFORMAT Table 8.13 shows the sensitivity, specificity and accuracy of predicting each component when primary CART analysis is applied.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 13 CART summary of ACR componentsOverall Response20% response variableActive Control SensitivitySpecificityAccuracyACR200.500.320.410.760.60Tender joint count 0.630.600.980.540.81Swollen joint count0.640.600.960.560.82CRP0.570.410.990.560.79ESR0.610.440.920.520.75Health assessment questionnaire0.500.430.850.480.67Patient’s assessment of pain0.550.470.980.410.71Physician’s global assessment of disease activity0.640.570.980.510.80Patient’s global assessment of disease activity0.560.490.960.420.71One can see that the sensitivity is very high for each of the components, indicating that responders are selected by the model with high probability. Conversely, specificity is lower than primary CART analysis for each component, indicating that fewer non-responders are correctly classified. Nonetheless, aside from health assessment questionnaire, overall accuracy of individual component classification is impressive and ranges from 0.70 to 0.82. Results were very similar to within 0.02 improvements when boosting and bagging methods are applied. Other than the Health Assessment Questionnaire Disability Index (HAQ-DI), these 8 components are not usually used individually for regulatory claims but do impact the patient (e.g. pain, tender joints, and global health). The HAQ-DI is occasionally filed as a secondary endpoint.The tree nodes that were selected by applying CART to each component response can be seen in REF _Ref490055106 \h \* MERGEFORMAT Table 8.14 and REF _Ref490055136 \h \* MERGEFORMAT Figure 8.10 to REF _Ref490055146 \h \* MERGEFORMAT Figure 8.17. Interestingly 3 of 5 nodes selected by the primary analysis, region, joint space narrowing score, and number of previous DMARDs, were not selected by any of the ACR component trees. Conversely, serum creatinine at baseline was selected as a tree node in 7 of 8 and baseline CRP was selected by 6 of 8 of the components but neither were selected by the primary analysis. Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 10 CART 20% Improvement in Tender Joint Count The first 2 nodes for swollen joint count are similar to tender joint count, but diverge somewhat as we progress further down the tree. Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 11 CART 20% Improvement in Swollen Joint CountCRP has higher affinity with SF36 scores than perhaps other ACR20 components. Of note is that ESR, which is often interchanged with CRP in the calculation of ACR20 also includes SF36 Mental Component Score higher up in the hierarchy of nodes. Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 12 CART 20% Improvement in CRPFigure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 13 CART 20% Improvement in ESRThe HAQ has a small number of nodes selected in order to predict the response Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 14 CART 20% Improvement in HAQAlso of note, perhaps intuitively, is that patient’s assessment of pain is predicted by baseline bodily pain and bodily pain standardised score from SF36 (with 0.71 accuracy).Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 15 CART 20% Improvement in Patient's Assessment of PainPatient’s and physician’s global assessment of disease activity share some similar nodes (serum creatinine and baseline CRP) at the top of their respective trees. Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 16 CART 20% Improvement in Physician's Global Assessment of Disease ActivityFigure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 17 CART 20% Improvement in Patient's Global Assessment of Disease ActivityTable STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 14 Tree nodes from CART analysis of ACR components20% response variableTree nodes (in order of importance)ACR20Baseline tender joint countRegionJoint space narrowing scoreNumber of previous DMARDsRaceTender joint Count Serum creatinineBaseline CRP SF36 general healthBaseline tender joint countSF36 General health standard scoreSwollen joint CountSerum creatinineBaseline CRPSerum albumin SF36 Bodily painSF36 Bodily pain standard scoreCRPSF36 Mental component scoreSF36 Role limitation - physical standard scoreSerum creatinineBaseline tender joint countBaseline CRPESRSerum creatinineSF36 Mental component scoreSteroid use at baseline Baseline tender joint count Baseline ESRHealth assessment questionnaireSerum creatinineBaseline CRPPatient’s assessment of painSF36 Bodily painSF36 Bodily pain standard scorePhysician’s global assessment of disease activitySerum creatinineBaseline CRPRaceBaseline tender joint countPatient’s global assessment of disease activitySerum creatinineBaseline CRPSerum albumin8.6.4.Logistic regression with multiple imputation As discussed in simulations in Section 7.7, the management of missing values by logistic regression was a significant disadvantage compared to CART which ameliorates the effect of missing data through identification of surrogate splits.In this section we applied multiple imputation to explore whether the results of logistic regression could be improved. In section 4.7 we investigated the amount of missing data in the available clinical dataset. We noted that some of the missing data is by design, i.e. where data on a particular variable was not collected in all studies and other data was missing at random. In the first step of multiple imputation we identified structurally missing data such as x-rays, i.e. data that was not collected by design of the study protocol. Variables not identified as structurally missing, we considered as missing at random. In the second step, we split the available dataset randomly into a training dataset (75%) and a test dataset (25%). For the training dataset we imputed missing values through multiple imputation using the SAS procedure PROC MI. Imputed values are generated by regression (linear or logistic as appropriate). By default, this method generates 5 iterations of imputed data. Since we had so many variables and so many missing data points, 25 iterations were applied, yielding 25 imputed datasets. A forward stepwise logistic regression was performed on each of the 25 imputed datasets using strict entry and exit criteria of 0.001, as informed by the simulations in Chapter 7. The results were combined into a single model using SAS procedure PROC MIANALYZE. The coefficients of the final model were the arithmetic mean of the individual coefficients estimated for each of the 25 logistic regression models. The parameters identified in the final model then applied to the non-imputed test dataset.When we applied the above steps to the clinical data, the 25 iterations of logistic regression selected similar but not the same variables in the models. The final combined model from PROC MIANLYZE included: Region (North America, Non North America);Inadequate Response Population (MTX IR, DMARD IR, TNF IR, No IR);Year of Enrolment;Drug Type (MTX, Placebo, Active);Baseline Patient Global Score;SF36 Pain interfere with work.These variables were then included in a predictive logistic regression model ( REF _Ref515463255 \h Table 8.15). Even though the test data set was not imputed only 9% of the data was lost due to missing values (2730 out of 2926).Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 15 Logistic Regression prediction results with multiple imputationPredictedTotalNon-ResponderResponderActualNon-Responder11683761544Responder5506361186Total171810122730Sensitivity0.54Specificity0.76Accuracy0.66The overall accuracy of the predicted model of 0.66 is a little better than non-imputed CART (0.60) and non-imputed logistic regression models (0.59), however the magnitude is still modest.8.6.5.CART analysis with multiple imputation For completeness and for the purposes of comparison, the training dataset generated through multiple imputation in section 8.6.4 was applied to CART and the resulting CART model was used with the test dataset to predict outcome. Results in REF _Ref516649189 \h Table 8.16 and REF _Ref516649209 \h Figure 8.18. Note the total number of patients differ to results in REF _Ref515463255 \h Table 8.15 due to handling of missing values present in test dataset.Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 16 CART prediction results with multiple imputationPredictedTotalNon-ResponderResponderActualNon-Responder13093081617Responder8144271241Total21237352858Sensitivity0.34Specificity0.81Accuracy0.61Figure STYLEREF 1 \s 8. SEQ Figure \* ARABIC \s 1 18 CART analysis with multiple imputationTJC= Tender Joint Count, SJC= Swollen Joint CountIn this analysis with multiple imputation the first 4 variables mirror that of the primary CART analysis (Baseline Tender Joint Count, Region, Joint Space Narrowing Score, Number of previous DMARDs). This similarity is not a surprise as the data originates from the same source. The tree begins to differ with the Baseline CRP node, however race is selected as a surrogate in the detailed analysis. Since the multiple imputation generates 25 datasets the resulting tree is able to include additional nodes before stopping rules are activated. Compared to the primary CART analysis the results are very similar (sensitivity 0.34 vs 0.41, specificity 0.81 vs 0.76, accuracy 0.61 vs 0.60), so in this situation little is to be gained by CART in applying multiple imputation methodology.8.7.SummaryIn this chapter we have presented the results of the CART analyses to predict patient response in the RA clinical trial database. The primary CART analysis, considering all available baseline and biomarker variables on all patients, generated a model that had an overall predictive accuracy of just 60%, although the explanatory variables selected in the model seem plausible for predicting patient outcome: a baseline tender joint count, region, joint space narrowing score, number of previous DMARDs and race. Of note, was that baseline swollen joint count was a good surrogate for baseline tender joint count (agreement 0.912), another strong surrogate was baseline CRP for number of DMARDs (agreement 0.866).Analysing the same patient population using random forest, bagging and boosting selected many of the same variables as important predictors but did little to improve predictive performance. Subgroup analyses by project, on patients with radiographic data, on control group (methotrexate), by smoking status and by rheumatoid factor saw some fluctuations in sensitivity and specificity also did little to improve overall predictive performance. As expected from simulations in chapter 7, a logistic regression analysis deleted too many patients to be able to conduct a stepwise analysis, thus justifying the selection of CART. As a further exercise, a logistic regression analysis was performed using the 5 variables identified by the primary CART analysis and using training and test datasets (Section 8.5.4). Even so, more than 60% of these data were excluded from the analysis due to missing data. The predictive performance of logistic regression proved to be similar to the CART analysis. The intent of this research was to identify biomarkers that might have a general application, therefore study specific variables were excluded from the above analyses. When study specific variables were allowed back into the model, compound, and drug type were primary and secondary splits in the tree, however performance of the model was still not impressive even when the oldest and least potent project (Project T) was excluded. Handling of missing values has been identified as a significant disadvantage of logistic regression especially as compared to CART, and was an important factor in selecting an appropriate method. A further exploratory analysis was therefore conducted whereby multiple imputation was performed on a training subset of the data to build a predictive model, and then applied to a non-imputed test set. Whilst this allowed an unrestricted forward stepwise model to be conducted, good prediction of ACR20 still remained elusive. In order to make fairer comparisons between CART and logistic regression with respect to handling of missing values, CART analyses were additionally performed on completers and on multiple imputed datasets ( REF _Ref516659492 \h Table 8.17).Table STYLEREF 1 \s 8. SEQ Table \* ARABIC \s 1 17 Summary of CART and Logistic regression methods on ACR20 MethodSensitivitySpecificityAccuracyCART Primary0.410.760.60CART on completers0.610.570.59CART on imputed data0.340.810.61Logistic Regression on completers 0.610.570.59Logistic Regression on imputed data0.540.760.66Nonetheless, overall accuracy across CART and logistic regression on raw data, complete records (on a limited subset of variables), or on multiple imputed data remained within a low, narrow range (0.59 to 0.66). Logistic regression with imputation has a slight edge over the other methods, but this comes with a number of assumptions and decisions to make such as, how to model the distribution for regression, deciding which variables are missing at random, missing completely at random or not missing at random. Although it was not an aim of this PhD, prediction of classification of ACR20 components was much more accurate, mostly driven in all components by sensitivities ranging from 0.85 to 0.99. There was some consistency in variables selected as tree nodes for the component scores but little consistency with overall ACR20 classification. Although there is relatively high accuracy of classification of the component scores, there is still variation in node selection between components. The ACR20 is a reductive measure to summarise patient response in a systemic, multi-faceted, chronic disease. This variation may indicate a certain insensitivity of the ACR20 and the ability to predict outcome. In the next chapter we will reflect on these results and their relevance for future clinical trials in RA. : Discussion and Conclusions9.1.IntroductionRheumatoid Arthritis (RA) is the most common inflammatory joint disease. It is a systemic, chronic, destructive, autoimmune disorder of unknown cause and no known cure. There is a considerable variation of disease occurrence and expression among different populations and groups of patients. RA is a multi-factorial disease with many postulated factors influencing onset, severity and outcome.There is extensive research on the disease and a significant amount of studies on the triggers and factors influencing duration and severity of the disease. The risk factors discussed in literature include gender, age, smoking, diet, ethnicity, and genetic factors. Studies investigating factors are largely limited to utilising univariate methodology, very few employ multivariate methods and even less consider the relationships between explanatory markers. When developing a new medicine, a question a clinical development team may wish to explore is what lessons can be learnt from past clinical development experiences. In particular, when there is in-house experience in the disease area, whether this valuable information could be used for the development of a new asset in the most efficient way. This thesis pursued 3 objectives; To identify the statistical or data mining method that best identifies biomarkers;To identify biomarkers and baseline patient assessments that predict patient outcome for disease progression;From these results to improve study design in RA clinical trials through patient selection and collection of most relevant data points attempted to predict patient response from data available at enrolment in clinical trials. If successful, this would lead to more efficient study designs that would take into account a patient’s likelihood of response by, for example, inclusion criteria and or stratification of analyses. Chapters 2 and 3 set the scene for this thesis by describing the disease of RA as well as how clinical trials are designed to demonstrate the effectiveness of investigational drugs. The clinical trial data available for exploitation in this thesis was described in Chapter 4. This is a large dataset pooled across 16 trials in 4 drug development programs, in this selection, any findings could potentially be generalised and applied to future programs. 9.2.Main findingsFrom the disease background and in particular the description of RA clinical trials and regulatory pathways, it was clear that the ACR20 would be the endpoint of interest for prediction. Since the ACR20 is a dichotomous response endpoint, then this became an exercise in classification. Chapters 5 and 6 explored and explained various data mining and statistical methods for classification. Chapter 5 focussed on a selection of methods of interest, whilst Chapter 6 described the outcomes of systematic reviews to identify methods applied in RA. In Chapter 6 it became clear that based on the features of the available clinical trial dataset, that logistic regression and CART would likely be the best methods for classification. Therefore, Chapter 6 went on to describe a systematic review on comparisons of CART and logistic regression not restricted to a RA setting.In the next step, Chapter 7 aimed to reduce the number of methods to a single choice based upon performance in simulated data with known correlations and structure. In Chapter 8, the winner, CART was applied to the clinical trial dataset. Chapter 1 laid out 3 specific objectives for this research:Identify the statistical or data mining method that best identifies biomarkers;Identify biomarkers and baseline patient assessments that predict patient outcome for disease progression;Improve study design in RA clinical trials through patient selection and collection of most relevant data points.The next 3 sections summarise how this research addressed these objectives.9.3.Objective 1: Choice of statistical or data mining method that best identifies biomarkersThis objective was approached by considering the following steps in section 1.2: Review and describe data mining methodologies for classification;Systematically review available literature for multivariate methods identifying risk factors for RA;Through the application of simulations, identify the best classification method for selecting risk factors.Methods that were considered included: CART (including Boosting Bagging and Random Forest);k-nearest neighbour classification;support vector machines;logistic regression (including Lasso and Ridge Regression);principal components;cross validation. Systematic reviews revealed a few other methods that have been utilised in RA data, but these were not relevant in predicting a dichotomous outcome, such as ACR20.In identifying and choosing a suitable method for classifying ACR20 response, I developed a shortlist of characteristics based upon general preferences and features I recognised in the available RA clinical trial dataset, and that I thought were important.These were: Ability to handle ‘mixed’ data;Handling of missing values;Robustness to outliers;Computational scalability;Interpretability;Ability to manage high dimensionality.Table STYLEREF 1 \s 9. SEQ Table \* ARABIC \s 1 1 Selection assessment for method selectionCharacteristicCARTk-NNSVMLogistic RegressionPrincipal ComponentsCan handle ‘mixed’ dataGoodPoorPoorGoodGoodHandling of missing valuesFairFairPoorFairPoorRobust to outliersGoodGoodPoorFairFairComputational scalability (Large N)GoodPoorPoorGoodGoodCan manage high dimensionalityGoodPoorPoorGoodGoodInterpretabilityFairGoodPoorGoodFairBased on these criteria, I considered CART and logistic regression to be the most preferred methods for an RA dataset ( REF _Ref500252433 \h Table 9.1).9.3.1.Systematic ReviewsAs mentioned in the previous section, systematic reviews were conducted to reveal which data mining and statistical methods I thought would be useful in classifying ACR20 response. Whilst restricting searches to an RA setting, the search terms for methods aimed to be broad. The terms used to search for methods were: CART, Multivariate analysis, Classification, Discriminant analysis, Cluster analysis, and Factor analysis. One could have considered not restricting the searches to just RA, however I considered the number of methods identified to be sufficient even if the search would not be considered exhaustive. However, when comparing the shortlisted methods of CART and logistic regression, no therapeutic area restriction was applied, some articles were identified outside of the field of medicine. Although articles were identified that used most of these methods in literature, none were used in the context of predicting ACR20 or dichotomous response in a RA dataset. Therefore, to the best of my knowledge, this thesis addresses a gap in known research.9.3.2.SimulationsA substantial number of simulations were conducted to explore the performance of logistic regression and CART in selecting variables with known high correlations to a dichotomous response variable. In the simulations 100 explanatory variables were generated, 10 of which were highly correlated with response and 90 with low correlation with response. In the simulations, over a thousand scenarios, replicated 100 times, were created ranging in size from 100 to 10,000 simulated patients (Appendix C.3). The computational intensity of the simulations was challenging. For the larger simulated patient datasets (N= 10,000), each scenario of 100 repetitions could take 200 to 300 minutes of processing time. Each run of the simulations could take 2 to 3 days, with frequent, unpredictable early terminations of the submitted program, due to memory limitations even on a Unix network. Scenarios were created by: varying the size of data sets, varying the level of high and low correlations with response, creating continuous and then dichotomous explanatory variables, and generating missing data. The scenarios generated were not an exact match for the available clinical trial dataset, nor were they exhaustive. Correlations for example were extended beyond that seen in the data, however in other respects such as types of variables and missing data rates, I considered the simulations to be anchored in the real world data. Therefore, I believe the simulations to be a valuable tool in selecting an analysis method. In the course of generating simulations, I created, developed and used VaSe plots to compare the performance of the chosen methods in being able to select higher correlated variables. In the example VaSe plot ( REF _Ref500252954 \h Figure 9.1), higher performing methods will plot closer to the top right hand corner. The advantage of this type of display is that the reader can quickly compare the performance of multiple methods. I found this very useful when looking at the multiple plots in chapter 7, for example to visually see how changing the sample size from 100 to 1,000 to 10,000 affected the accuracy of selecting highly correlated variables and not selecting lower correlated variables. A potential disadvantage is that these plots are potentially confused with Receiver Operating Characteristic curves, so care has been taken in the text when describing results of variable selection. Figure STYLEREF 1 \s 9. SEQ Figure \* ARABIC \s 1 1 Example VaSe plotIn almost all cases, we saw that CART usually outperforms logistic regression in variable selection. The few scenarios where logistic regression outperforms CART are those that are least likely such as where the lower correlated variable are at a minimum of 0.0 correlation with the response and the higher correlated variables are correlated at a maximum of 0.90 with the response.CART seemed to be particularly robust to missing values in the dataset. Logistic regression on the other hand, lost much of the data through listwise deletion so that with even a small rate of missing of just 1%, over 63% of the data was ignored. Higher rates of simulated missing data became untenable for a logistic regression analysis. 9.4.Objective 2: Identification of biomarkers and baseline patient assessments that predict patient outcomeThis objective was approached by considering the following steps in section 1.2: Describe the epidemiology, aetiology and diagnosis of RA, as well as risk factors and assessments of disease activity;Collate and describe the available clinical data;Apply the best method from simulations to clinical database.The background research, data processing, systematic reviews, methods research, and simulations in this thesis led to the analysis in Chapter 8. This was a planned and systematic process with the aim to be able to present robust, repeatable results. As identified in the systematic reviews, there is a gap in the literature concerning the prediction of ACR20 for patients with RA. In Chapter 8 CART was implemented to attempt to classify ACR20. The primary model identified baseline tender joint count, geographic region, joint space narrowing score, number of previous disease modifying anti-rheumatic drugs and race as the most important variables for predicting ACR20. Whilst these seem to be quite plausible features for disease response the prediction accuracy was disappointing at just 60%. The model identified just 41% of those patients who responded to treatment (sensitivity), although it was better at recognising non-responders at 76% (specificity). Enhancements to CART procedure (Bagging, Boosting and Random Forest) did little to improve accuracy of prediction. Further exploratory analyses looking at subgroups such as drug project, smoking status, or rheumatoid factor, also had little effect on improving prediction. The primary model originally excluded study specific variables in the hope that a model generated could be applied generally. In an exploratory analysis, I decided to implement CART on a dataset that included compound, protocol number and drug type. As may be expected in a dataset featuring approved medicines for RA; compound and drug type featured as primary and secondary splitters of the tree. However, performance of prediction changed little. In a further analysis, Compound T (the oldest and least potent compound) was excluded from the data and again there was no improvement in prediction. I found this result consistent with the primary CART analysis in that baseline tender joint count and region from the primary analysis were good surrogate splitters for compound and drug type, indicating good interchangeability of these variables and I believe explains the similar prediction accuracy between analyses. As I expected from simulations, a na?ve forward selection logistic regression, listwise deleted all of the data from the analysis. Nevertheless, when I applied logistic regression using just those variables selected by the primary CART analysis, I found the prediction accuracy remained almost unchanged. A rationale for this may be that when the 2 methods are able to use the same level of information from the dataset (without over deletion) then their performance is similar. CART manages missing data by identifying surrogate variables for those chosen in the tree, so that when it comes to prediction, more of the data is utilised. In a dataset of this type where missing data are a prominent feature, this gives a distinct advantage for CART over logistic regression. Methods exist to deal with missing data, from complete case analysis to single imputation to multiple imputation. In order to tip the balance more in favour of logistic regression, I did some further investigation through implementing multiple imputation. I did this by applying multiple imputation on a training set, on which a logistic regression model was re-generated, and then used this predictive model to a non-imputed test set. In this scenario, logistic regression performed very similar to CART, albeit with a different but overlapping set of predictor variables. The use of multiple imputation for logistic regression enabled the effective use of the method. Since missing data are imputed from other variables, the full information is used in the data, as does surrogate splitters in CART. With logistic regression now able to utilise the richness of the full information in the training data set, this could explain the equity of prediction performance between the 2 methods. With these results being so similar, I would still favour one one-step approach of CART over a two-step logistic regression with multiple imputation. Multiple imputation route is more reliant on assumptions and choices such as: how to model the distribution for regression, deciding which variables are missing at random, missing completely at random and not missing at random. Although not an aim of this PhD, we did see though some interesting results with high accuracy when considering the 8 ACR20 components separately. Sensitivity was very high (range 0.85 to 0.99) for all components individually, although specificity was lower, the overall accuracy was improved across all components compared with the ACR20 composite endpoint. Notably each of the individual component’s trees varied but variables such as baseline serum creatinine, baseline CRP, baseline tender joint count were selected by at least half of the components. As described in Section 3.3.1 the ACR20 is a binary, dichotomised endpoint, calculated by a composite algorithm. This algorithm includes in its calculation swollen and tender joint counts for all patients, but then the best 3 of the 5 remaining measures for each calculation. The impact being that the precise calculation will vary from patient to patient. Add to that the convention that missing patients are considered to be non-responders, may lead to ACR20 being too blunt an instrument to predict. Dichotomised endpoints, particularly those dichotomised from continuous data have issues with efficiency ADDIN EN.CITE <EndNote><Cite><Author>Senn</Author><Year>2009</Year><RecNum>408</RecNum><DisplayText>(Senn & Julious, 2009)</DisplayText><record><rec-number>408</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1530022034">408</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Senn, S.</author><author>Julious, S.</author></authors></contributors><titles><title>Measurement in clinical trials: A neglected issue for statisticians?</title><secondary-title>Stat Med</secondary-title></titles><periodical><full-title>Stat Med</full-title></periodical><pages>3189-3209</pages><volume>28</volume><dates><year>2009</year></dates><urls></urls></record></Cite></EndNote>(Senn & Julious, 2009). In a simple setting of a single dichotomised endpoint, maximum efficiency is no more than 64%, less if the dichotomisation is not at the median value. For ACR20, the dichotomisation is not at the median but rather at an arbitrary value (20%), in addition multiple dichotomies are created, perhaps compounding the inefficiency of the endpoint. The high prediction rates for each of the components of ACR20 would seem to bear out the description of ACR20 as a blunt instrument and supporting the assertion that multiple dichotomisation compounds inefficiency. The improved prediction of the components also suggests that the methods are valuable, but the choice of classifier unyielding. On reflection it might be that the simulations conducted in Chapter 7 were more reflective of a simpler constructed response variable, than of the composite endpoint that is the ACR20. In my view the clinical relevance of the results of chapter 8 are mixed. ACR20 is a well-established, well understood and accepted endpoint for measuring improvement of RA in the context of clinical trials and development strategies for gaining regulatory approval for RA medicines. In clinical practice, in interactions between patient and physician, it is of little utility, usually their interest is of disease state (e.g. DAS28) or on a focus with a particular symptom (e.g. tender joints, pain). So in my opinion in the context of relevance in a clinic setting, the results seen in improvements in specific symptoms could be useful, and might enable a physician to predict what sort of improvements a patient might expect in 6 months. If we had seen strong prediction results for ACR20, then I would argue there is clinical relevance for clinical trial design (see enrichment discussion in next section). In chapter 8, the prediction of ACR20 is not strong, and possibly because of this weakness, the selection of predictors is something of a moving target. In the primary model, baseline tender joint count, region, joint space narrowing score, numbers of DMARDs and race were nodes on the developed tree and I judge these to be clinically feasible predictors of outcome. However, baseline tender joint count could adequately be replaced by the surrogate baseline swollen joint count and number of DMARDs replaced by baseline CRP. If we accept exploratory analyses, then by allowing study specific parameters in the dataset; then compound, drug type and population type become important predictors (with same level of prediction). So it is still a little unclear what recommendation for choice of predictor one could make to predict ACR20.9.5.Objective 3: Improvement in study design in RA clinical trials through patient selection and collection of most relevant data points This objective was approached by considering the previous objectives and the following step in section 1.2: Review current designs of clinical trials in RA and strategies for approval;In 2012, the US Food and Drug Administration (FDA) published a draft guidance on enrichment strategies for clinical trials ADDIN EN.CITE <EndNote><Cite><Author>FDA</Author><Year>2012</Year><RecNum>186</RecNum><DisplayText>(FDA, 2012)</DisplayText><record><rec-number>186</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502704051">186</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>FDA</author></authors></contributors><titles><title>Guidance for Industry: Enrichment Strategies for Clinical Trials to Support Approval of Human Drugs and Biological Products</title></titles><dates><year>2012</year></dates><urls></urls></record></Cite></EndNote>(FDA, 2012). The guidance defines enrichment as “the prospective use of any patient characteristic to select a study population in which detection of a drug effect (if one is in fact present) is more likely than it would be in an unselected population”, where characteristics can be drawn from demographic, pathophysiologic, historical, genetic or proteomic, clinical, and psychological sources. By recruiting patients with a higher likelihood of having a certain event can lead to a higher effect size, reduce heterogeneity and lead to smaller more efficient clinical trials. In a clinical study which incorporates an enrichment design, patients are screened for the presence (or absence) of a biomarker or characteristic. Only these patients are randomised into the trial. In these studies, the safety, and benefit of the treatment only within this patient segment is assessed. Where the presence of the marker or characteristic is low (<30%), great savings can be made in cost of clinical trials versus all-comer designs. The FDA guidance notes trastuzumab (Herceptin) as an example of enrichment in metastatic breast cancer. Patients with high HER 2/neu expressing tumours -- about 25% of all breast cancer patients, achieved significant survival of 5 months versus less than 2 months' survival in all-comers. “The strategy of focusing part of the trial on the specific population of patients with high HER 2/neu-expressing tumours ultimately supported use of the drug in the marker-selected population despite the significant cardiotoxicity that emerged," the guidance notes. Enrichment designs are usually implemented when there is strong evidence that patients will respond in a biomarker defined subpopulation. Enrichment designs can avoid dilution of the effect of intervention caused by patients not expected to respond. Results of these trials therefore are only applicable to this restricted population, whilst the treatment effect in the general population would remain unclear. In the context of personalised healthcare, medicines developed through enrichment strategies, might be expected to have increased certainty of response. For payers, such as the NHS in the UK, this would be appealing and economic savings may be gained by not treating patients unlikely to respond. For pharmaceutical companies a premium price could be negotiated without additional burden on healthcare systems. To take this approach however there needs to be strong belief in the biomarkers, otherwise it could be that patients not selected for trials might still have had a reduced, but still valuable, response to treatment. Due to the low predictive accuracy of prediction of ACR20 as discussed in Chapter 8, my assessment is that this objective was not achievable with the data available and in this thesis.9.6.Main Thesis Strengths and AchievementsI maintain that this research is based on sound planning and a thorough investigation into data from an RA population. Based on systematic reviews and best available knowledge, this thesis appears to be the only research attempting to predict RA patient response with respect to ACR20, and is the first research that compares CART and logistic regression in variable selection in an RA setting. As far as I am aware this is the only research that has used a dataset of this size and type to attempt to predict RA response. In general, I found little research in predicting patient outcomes as measured in RA clinical trials. The systematic review in Section 6.3 found just one article that used DAS28 to discriminate between active and quiescent disease and another that used disease activity to correlate with serum secreted phospholipase. Just one author directly attempted to predict patient outcomes from clinical biomarker data. Van der Helm et al ADDIN EN.CITE <EndNote><Cite><Author>van der Helm</Author><Year>2007</Year><RecNum>46</RecNum><DisplayText>(van der Helm, 2007)</DisplayText><record><rec-number>46</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534980">46</key></foreign-keys><ref-type name="Generic">13</ref-type><contributors><authors><author>van der Helm, A</author></authors></contributors><titles><title><style face="normal" font="Dutch801BT-Roman" size="100%">Emerging Patterns of Risk Factor Make-Up Enable Subclassification of Rheumatoid Arthritis</style></title><secondary-title>ARTHRITIS & RHEUMATISM</secondary-title></titles><pages>1728-1735</pages><volume>56</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>RA</keyword></keywords><dates><year>2007</year><pub-dates><date>2007</date></pub-dates></dates><label>47</label><urls><related-urls><url><style face="underline" font="default" size="100%">;(van der Helm, 2007) used logistic regression to predict progression to full RA from early undifferentiated arthritis using baseline characteristics, this research was not affected to the same extent by missing data. This research used a large comprehensive dataset to develop evaluation criteria for selecting the most applicable method for predicting ACR20. Whilst these criteria were specifically selected due to the nature of the available dataset, features such as missing data, mixed data types will be common to other clinical trials in other diseases.In the simulations, this thesis compared the performance of CART and logistic regression in being able to select highly correlated variables in a large and high dimensional dataset. This research developed the variable selection (VaSe) plots to enable visually the comparative performance. I generated datasets under known conditions and varied correlation structure, number of patients, types of variable and rates of missing value. Within the scope of the simulations I conducted, CART outperformed logistic regression in most scenarios. Even with enhancements to logistic regression by amending significance levels for entry and staying CART was still unbeaten. In my judgement, there was not a need to improve performance by applying boosting, bagging or random forest enhancements to CART.With respect to the final analysis of predicting ACR20 response from measures recorded at enrolment to clinical trials, the results were disappointing with prediction rates at little more than 60%. Whilst this is better than random guessing, it has little value in practical terms. Although not an aim of this thesis, but nonetheless interesting, was the ability to predict the individual components of the ACR20 with much more accuracy. Most of the components could be predicted at levels around 70 to 80%, but each had different sets of predictors. This suggests to me that the methodology employed was of utility but the chosen endpoint of ACR20 was not predictable from the available variables in the data. The ACR20 was developed as a measured that could capture the response to therapy in this multifactorial disease, the ACR20 therefore is a broad measure. On reflection it is perhaps this generalisation and bluntness of the measure that makes it so hard to predict. 9.7.Key LimitationsThe final results of this thesis are based on a patient database from patients that were recruited over a period from 1998 to 2008, during this period the treatment of RA has evolved, as observed by changes in methotrexate dosing. Patients have varying response to other treatments for RA such as inadequate response to methotrexate, anti-TNF therapies and other DMARDs. The data also represents patients from 50 countries which introduces possible heterogeneity through factors such as diet and standard of medical care. It’s possible that this data may not be representative of patients today or even of real world patients not recruited into clinical trials. Some of the radiographic variables when considered by project have higher correlations than other variables. By design, not all studies performed X-rays to capture these data. In the final analysis, joint space narrowing derived from X-rays, was one of the key predictors for the main CART model in the primary analysis.The simulations performed in this thesis have not been exhaustive, but are anchored within RA clinical development and the distributional forms and clinical parameters are informed by the clinical data. We have not explored the combination of categorical and missing values, or non-binary categorisation of explanatory variables and we have not explored correlations between explanatory variables. Neither have we explored compensating for the way SAS Proc Logistic manages missing data by nominating missing data as a class variable: in real life practice we wouldn’t be able to assume data is missing for the same reason, however this is rarely done in practice. We have also not considered nonNormal continuous data. Nonetheless, in the scenarios explored, CART has been robust in performance in both sensitivity and specificity compared to logistic regression. Whilst this staged approach intends to mimic the real patient database, it is recognised that it may be limited in expressing the nuances in such a large database and complex disease area. These simulations are anchored in the dataset I had available, but can only inform an approach to analyse the real patient data rather than definitively compare these methods in a real life data setting. As such conclusions drawn may only be valid for these situations, and caution should be exercised in attempting to draw general conclusions.Nonetheless the simulations have shown the value of undertaking simulations. For example, for large datasets we have seen that by applying a very strict entry and staying criteria for the logistic regression we can improve performance in terms of specificity and sensitivity of variable selection. 9.8.Future Related ResearchIn the conduct of this research, the thesis was limited by the available database. Some risk factors for diagnosis and progression of disease were not available, such as diet, and genetic factors. In regard to genetic factors, literature suggests a genetic component to the aetiology of RA with the HLA (human leukocyte antigen) region a major genetic risk area. A more prospective research could be considered to incorporate these other sources of data. Recently, there have been calls to make pharmaceutical clinical trial more available to researchers ADDIN EN.CITE <EndNote><Cite><Author>Godlee</Author><Year>2012</Year><RecNum>179</RecNum><DisplayText>(Godlee, 2012)</DisplayText><record><rec-number>179</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1494869579">179</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Godlee, F</author></authors></contributors><titles><title>The new BMJ policy on sharing data from drug and device trials</title><secondary-title>BMJ</secondary-title></titles><periodical><full-title>BMJ</full-title></periodical><volume>345</volume><number>e7888</number><dates><year>2012</year></dates><urls></urls></record></Cite></EndNote>(Godlee, 2012). Extension of the database to RA trials conducted by other organisations would be a useful project for future research.9.9.Overall Conclusions To the best of my knowledge this is the only research that thoroughly explores the prediction of RA response through baseline characteristics in clinical trials. During the course of this thesis we have developed some useful variable selection (VaSe) graphics to compare performance of methods.This research found the application of CART to predict patient ACR20 response is justified over logistic regression in data of this type, specifically with respect to management of missing data. In this respect, the simulations conducted on the variety of scenarios were successful in identifying the better method. Multiple imputation was useful in improving performance of logistic regression, but no more than CART performance. Unfortunately, the final primary results of the analysis have not been of sufficient level of predictability to be able to impact either individual patient prediction or overall RA clinical trial design. The primary model was able to identify non-responders 76% of the time, but responders to treatment just 41% of the time. Application of enhancements to CART gave little improvement to the prediction models. The results from this research could be beneficial for helping to predict patient response to specific components of the ACR20, such as swollen and tender joint counts. For both these measures baseline serum creatinine and CRP seemed to be important measures in models that identify over 96% of responders and over 54% of non-responders.The methods, results and evaluation process described in this research could be used for other clinical developments that seek to exploit biomarkers and baseline characteristics in study design, not just for RA but also for other disease areas. It is recommended that for research of this type, a systematic review of predictive markers is performed as well as a systematic review of methodology. In addition, it is recommended to perform simulations based on expected data structure to assess the performance of methods against a ‘known truth’. A protocol for simulations would increase validity, and finally apply the optimal method to retrospective data. For the research question addressed in this dissertation, we chose to predict the regulatory accepted endpoint of ACR20. This research, through its large clinical database illuminates that this is too tall an expectation to be able to predict ACR20. Nonetheless when one considers individual components that include pain, global health, joint status, this research was able to predict with high accuracy. In the context of personalised health care, prediction of the course specific symptoms may be more relevant to the patient than a complex composite score.References ADDIN EN.REFLIST Abdel Nasser, A., Rasker, J., & Valkenberg, H. (1997). Epidemiological and Clinical Aspects Relating to the Variability of Rheumatoid Arthritis. Seminars in Arthritis and Rheumatism, 27(2), 123-140. Alamanos, Y., & Drosos, A. (2005). Epidemiology of adult rheumatoid arthritis. Autoimmunity reviews, 4(3), 130-136. Aletaha, D. (2010). 2010 Rheumatoid Arthritis Classification Criteria. In ARTHRITIS & RHEUMATISMAn American College of Rheumatology/European League Against Rheumatism Collaborative Initiative (Vol. 62, pp. 2569-2581). (Reprinted from: Not in File).Aletaha, D., Landewe, R., Karonitsch, T., Bathon, J., Boers, M., Bombardier, C., . . . Acr. (2008). Reporting disease activity in clinical trials of patients with rheumatoid arthritis: EULAR/ACR collaborative recommendations. Arthritis Rheum, 59(10), 1371-1377. doi:10.1002/art.24123Aletaha, D., Smolen, J., & Ward, M. (2006). Measuring function in rheumatoid arthritis: Identifying reversible and irreversible components. Arthritis and Rheumatism, 54(9), 2784-2792. Alfaro, E., Gamez, M., & Garcia, N. (2015). adabag: Applies Multiclass AdaBoost.M1, SAMME and Bagging: R package version 4.1. Retrieved from , D. G., & Bland, J. M. (1994). Diagnostic Tests 1: sensitivity and specificity. BMJ, 308, 1552. Arnett, F. C., Edworthy, S. M., Bloch, D. A., McShane, D. J., Fries, J. F., Cooper, N. S., . . . . (1988). The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum, 31(3), 315-324. Ates, A., Kinikli, G., Turgay, M., Akay, G., & Tokgoz, G. (2007). Effects of rheumatoid factor isotypes on disease activity and severity in patients with rheumatoid arthritis: a comparative study. Clin Rheumatol, 26(4), 538-545. doi:10.1007/s10067-006-0343-xBarlow, J. H., & Wright, C. C. (1998). Dimensions of the Center of Epidemiological Studies-Depression Scale for people with arthritis from the UK. Psychological reports 83(3), 915-919. Baron, G., Boutron, I., Giraudeau, B., & Ravaud, P. (2007). Reporting of radiographic methods in randomised controlled trials assessing structural outcomes in rheumatoid arthritis. Annals of the rheumatic diseases, 66(5), 651-657. Breiman, L. (1996). Bagging Predictors. Machine Learning, 24(2), 123-140. Breiman, L., Friedman, J., Olshen, R. A., & Stone, C. J. (1984). Classification and Regression Trees: Chapman and Hall.Broeder, d., Joosten, L. A. B., Saxne, T., Heinegard, D., Fenner, H., Miltenburg, A. M. M., . . . Barrera, P. (2002). Long term anti-tumour necrosis factor alpha monotherapy in rheumatoid arthritis: Effect on radiological course and prognostic value of markers of cartilage turnover and endothelial activation. Annals of the Rheumatic Diseases, 61(4), 311-318. Burckhardt, C. S., & Bjelle, A. A. (1994). Swedish version of the short-form McGill Pain Questionnaire. Scandinavian journal of rheumatology 23(2), 77-81. Burton A., A. D. G., Royston P., Holder, R.L. . (2006). The design of simulation studies in medical statistics. Statistics in medicine, 25, 4279-4292. Carano, R. A. D., Lynch, J. A., Redei, J., Ostrowitzki, S., Miaux, Y., Zaim, S., . . . H.K., G. (2004). Multispectral analysis of bone lesions in the hands of patients with rheumatoid arthritis. Magnetic resonance imaging 22(4), 505-514. Casey, A. T., Bland, J. M., & Crockard, H. A. (1996). Development of a functional scoring system for rheumatoid arthritis patients with cervical myelopathy. Annals of the Rheumatic Diseases, 55(12), 901-906. Cella, D., Yount, S., Sorensen, M., Chartash, E., Sengupta, N., & Grober, J. (2005). Validation of the Functional Assessment of Chronic Illness Therapy Fatigue Scale relative to other instrumentation in patients with rheumatoid arthritis. The Journal of rheumatology, 32(5), 811-819. Chiaravalloti, N. D., Christodoulou, C., Demaree, H. A., & DeLuca, J. (2003). Differentiating simple versus complex processing speed: influence on new learning and memory performance. Journal of clinical and experimental neuropsychology, 25(4), 489-501. Choy, E. H., & Panayi, G. S. (2001). Cytokine pathways and joint inflammation in rheumatoid arthritis. The New England Journal of Medicine, 344(12), 907-916. Cole, J. C., Motivala, S. J., Khanna, D., Lee, J. Y., Paulus, H. E., & Irwin, M. R. (2005). Validation of single-factor structure and scoring protocol for the Health Assessment Questionnaire-Disability Index. Arthritis Rheum, 53(4), 536-542. Cooper, N. J. (2000). Economic burden of rheumatoid arthritis: a systematic review. Rheumatology. (Oxford), 39(1), 28-33. Cortes, C., & Vapnik, V. N. (1995). Support-Vector Networks. Machine Learning, 20(3), 273-297. Cross, M., Smith, E., Hoy, D., Carmona, L., Wolfe, F., Vos, T., . . . March, L. (2014). The global burden of rheumatoid arthritis: estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis, 0, 1-7. Curtis, R., Groarke, A., Coughlan, R., & Gsel, A. (2005). Psychological stress as a predictor of psychological adjustment and health status in patients with rheumatoid arthritis. Patient Education & Counseling, 59(2), 192-198. Dessein, P. H., Joffe, B. I., Veller, M. G., Stevens, B. A., Tobias, M., Reddi, K., & Stanwix, A. E. (2005). Traditional and Nontraditional Cardiovascular Risk Factors Are Associated with Atherosclerosis in Rheumatoid Arthritis. Journal of Rheumatology, 32(3), 435-442. Doran, M., Pond, G., Crowson, C., Fallon, W., & Gabriel, S. (2002). Trends in incidence and mortality in rheumatoid arthritis in Rochester, Minnesota, over a forty-year period. Arthritis and rheumatism, 46(3), 625-631. Edwards, C. J., Feldman, J. L., Beech, J., Shields, K. M., Stover, J. A., Trepicchio, W. L., . . . Pittman, D. D. (2007). Molecular profile of peripheral blood mononuclear cells from patients with rheumatoid arthritis. Molecular medicine 13(1-2), 40-58. EMA. (2003). Points to Consider on the Clinical Investigation of Medicinal Products other than Nsaids in Rheumatoid Arthritis.FDA. (2010). Guidance for Industry Clinical Development Programs for Drugs, Devices, and Biological Products for the Treatment of Rheumatoid Arthritis (RA). In. (Reprinted from: Not in File).FDA. (2012). Guidance for Industry: Enrichment Strategies for Clinical Trials to Support Approval of Human Drugs and Biological Products. FDA. (2016). Study Data Standards: What you need to know. Retrieved from . (2017). Briefing Document Arthritis Advisory Committee PLIVENSIA? (sirukumab). Retrieved from , D. T., Anderson, J. J., Boers, M., Bombardier, C., Furst, D., Goldsmith, C., . . . Strand, V. (1995). American College of Rheumatology Preliminary definition of improvement in rheumatoid arthritis. Arthritis and Rheumatism, 38(6), 727-735. Fieller, N. (2011). Multivariate Data Analysis course notes. University of Sheffield. Firestein, G. (2003). Evolving concepts of rheumatoid arthritis. Nature, 423(6937), 356-361. Fransen, J., & van Riel, P. (2009). The Disease Activity Score and the EULAR response criteria. Rheumatic diseases clinics of North America, 35(4), 745-757. Freund, Y., & Schapire, R. E. (1999). A Short Introduction to Boosting. Journal of Japanese Society for Artificial Intelligence, 14(5), 771-780. Fries, J. F., Spitz, P., Kraines, R. G., & Holman, H. R. (1980). Measurement of patient outcome in arthritis. Arthritis Rheum, 23(2), 137-145. Frison, L. J., & Pocock, S. J. (1997). Linearly divergent treatment effects in clinical trials with repeated measures: efficient analysis using summary statistics. Stat Med, 16, 2855-2872. Gaffo, A., Saag, K. G., & Curtis, J. R. (2006). Treatment of rheumatoid arthritis. Am. J. Health Syst. Pharm, 63(24), 2451-2465. doi:10.2146/ajhp050514Gao, P., Zhou, X., Wang, Z.-n., Song, Y.-x., Tong, L.-l., Xu, Y.-y., . . . Xu, H.-m. (2012). Which Is a More Accurate Predictor in Colorectal Survival Analysis? Nine Data Mining Algorithms vs. the TNM Staging System. PloS one, 7(7), e42015. doi:doi:10.1371/journal.poneGenant, H. K. (1983). Methods of assessing radiographic change in rheumatoid arthritis. The American journal of medicine, 75(6A), 35-47. Girard, F., Guillemin, F., Novella, J. L., Valckenaere, I., Krzanowska, K., Vitry, F., . . . Le Lo?t, X. (2002). Health-care use by rheumatoid arthritis patients compared with non-arthritic subjects. Rheumatology 41(2), 167-175. Godlee, F. (2012). The new BMJ policy on sharing data from drug and device trials. BMJ, 345(e7888). Goeppinger, J., Doyle, M. A., Charlton, S. L., & Lorig, K. (1998). A nursing perspective on the assessment of function in persons with arthritis. Research in nursing & health 11(5), 321-331. Gotzsche, P. C., & Johansen, H. K. (1998). Meta-analysis of short-term low dose prednisolone versus placebo and non-steroidal anti-inflammatory drugs in rheumatoid arthritis. BMJ (Clinical research ed. ), 316(7134), 811-818. Gourraud, P., Dieude, P., Boyer, J., Nogueira, L., Cambon, T., Mazieres, B., . . . Constantin, A. (2007). A new classification of HLA-DRB1 alleles differentiates predisposing and protective alleles for autoantibody production in rheumatoid arthritis. Arthritis research & therapy, 9(2), R27. Guillemin, F., Brian?on, S., & Pourel, J. (1992). Validity and discriminant ability of the HAQ Functional Index in early rheumatoid arthritis. Disability and rehabilitation 14(2), 71-77. Hastie, T., Tibshirani, R., & Friedman, J. (2011). The Elements of Statistical Learning (Second ed.): Springer He, H., & Garcia, E. A. (2009). Learning from Imbalanced Data. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 21(9), 1263-1284. Heijde, v. d. (2002). Structural damage in rheumatoid arthritis as visualized through radiographs. Arthritis research, 4, S29-S33. Hochberg, M. C., Chang, R. W., Dwosh, I., Lindsey, S., Pincus, T., & Wolfe, F. (1992). The American College of Rheumatology 1991 revised criteria for the classification of global functional status in rheumatoid arthritis. Arthritis and Rheumatism, 35(5), 498-502. Hutchinson, D., Shepstone, L., Moots, R., Lear, J. T., & Lynch, M. P. (2001). Heavy cigarette smoking is strongly associated with rheumatoid arthritis RA, particularly in patients without a family history of RA. Annals of the Rheumatic Diseases, 60(3), 223-227. Isaacs, J., & Ferraccioli, G. (2011). The need for personalised medicine for rheumatoid arthritis. Annals of the rheumatic diseases, 70(1), 4-7. Ishwaran, H., & Kogalur, U. (2017). randomForestSRC: Random Forests for Survival, Regression, and Classification (RF-SRC). R package version 2.5.0. Iversen, M. D., Eaton, H. M., & Daltroy, L. H. (2004). How rheumatologists and patients with rheumatoid arthritis discuss exercise and the influence of discussions on exercise prescriptions. Arthritis and Rheumatism, 51(1), 63-72. James, K. E., White, R. F., & Kraemer, H. C. (2005). Repeated split sample validation to assess logistic regression and recursive partitioning: An application to the prediction of cognitive impairment. Stat Med, 24(19), 3019-3035. Jenkins, M., Flynn, A., Smart, T., Harbron, C., Sabin, T.., Ratnayake, J., Delmar, P., Herath, A., Jarvis, P., Matcham, J. (2011). A statistician’s perspective on biomarkers in drug development. Pharmaceutical Statistics, 10, 494-507. Jones, P. W., Ziade, M. F., Davis, M. J., & Dawes, P. T. (1993). An index of disease activity in rheumatoid arthritis. Statistics in medicine 12(12), 1171-1181. Juang, Y. T., Peoples, C., Kafri, R., Kyttaris, V. C., Sunahori, K., Kis-Toth, K., . . . Tsokos, G. C. (2011). A systemic lupus erythematosus gene expression array in disease diagnosis and classification: a preliminary report. . Lupus, 20(3), 243-249. Julià, A., Erra, A., Palacio, C., Tomas, C., Sans, X., Barceló, P., & Marsal , S. (2009). An eight-gene blood expression profile predicts the response to infliximab in rheumatoid arthritis. PloS one, 4(10), e7556. Julious, S. (2000). Letter to the Editor Repeated measures in clinical trials: analysis using mean summary statistics and its implications for design. Stat Med, 19, 3133-3135. Kabacoff, R. I. (2015). R In Action: Manning.Karlson, E. W., Lee, I. M., Cook, N. R., Manson, J. E., Buring, J. E., & Hennekens, C. H. (1999). A retrospective cohort study of cigarette smoking and risk of rheumatoid arthritis in female health professionals. Arthritis and Rheumatism, 42(5), 910-917. Kitsantas, P., Hollander, M., & Li, L. (2006). Using classification trees to assess low birth weight outcomes. Artif Intell Med, 38(3), 275-289. doi:10.1016/j.artmed.2006.03.008Kohavi, R. (1995). A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection. Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence 2(12), 1137-1143. Korman, B. D., Seldin, M. F., Taylor, K. E., Le, J. M., Lee, A. T., Plenge, R. M., . . . Remmers, E. F. (2009). The chromosome 7q region association with rheumatoid arthritis in females in a British population is not replicated in a North American case-control series. Arthritis and Rheumatism, 60(1), 47-52. Kuhnert, P. M., Do, K. A., & McClure, R. (2000). Combining non-parametric models with logistic regression : an application to motor vehicle injury data. Computational Statistics & Data Analysis 34, 371-386. Kvien, T. K. (2004). Epidemiology and Burden of Illness of Rheumatoid Arthritis. PharmacoEconomics, 22 Suppl(1), 1-12. Landewe, R., & van der Heijde, D. (2005). Presentation and analysis of radiographic data in clinical trials and observational studies. Annals of the rheumatic diseases, 64, iv48-iv51. Lard, L. R., Visser, H., Speyer, I., vander, H., Zwinderman, A., Breedveld, F. C., & Hazes, J. M. (2001). Early versus delayed treatment in patients with recent-onset rheumatoid arthritis: comparison of two cohorts who received different treatment strategies. The American journal of medicine, 111(6), 446-451. Larsen, A., Dale, K., & Eek, M. (1977). Radiographic evaluation of rheumatoid arthritis and related conditions by standard reference films. Acta Radiol. Diagn. (Stockh), 18(4), 481-491. Lin, M. K., Farewell, V., Vadas, P., Bookman, A. A., Keystone, E. C., & Pruzanski, W. (1996). Secretory phospholipase A2 as an index of disease activity in rheumatoid arthritis. Prospective double blind study of 212 patients. The Journal of rheumatology, 23(7), 1162-1166. Lindqvist, E., Jonsson, K., Saxne, T., & Eberhardt, K. (2003). Course of radiographic damage over 10 years in a cohort with early rheumatoid arthritis. Annals of the Rheumatic Diseases, 62(7), 611-616. Liu, W., Li, X., Ding, F., & Li, Y. (2008). Using SELDI-TOF MS to identify serum biomarkers of rheumatoid arthritis. Scandinavian journal of rheumatology 37(2), 94-102. Metsios, G. S., Stavropoulos-Kalinoglou, A., Panoulas, V. F., Koutedakis, Y., Nevill, A. M., K.M., D., . . . Kitas, G. D. (2008). New resting energy expenditure prediction equations for patients with rheumatoid arthritis Rheumatology, 47(4), 500-506. Mielenz, T., Jackson, E., Currey, S., DeVellis, R., & Callahan, L. F. (2006). Psychometric properties of the Centers for Disease Control and Prevention Health-Related Quality of Life (CDC HRQOL) items in adults with arthritis. Health and quality of life outcomes, 4(66), 1-8. Moher, D., Cook, D. J., Eastwood, S., Olkin, I., Rennie, D., & Stroup, D. F. (1999). Improving the quality of reports of meta-analyses of randomised controlled trials: the QUOROM statement Quality of Reporting of Meta- analyses. Lancet, 354(9193), 1896-1900. NCCfCC. (2015). Rheumatoid arthritis: national clinical guideline for management and treatment in adults: London: Royal College of Physicians.Nell, V. P. K., Machold, K. P., Eberl, G., Stamm, T. A., Uffmann, M., & Smolen, J. S. (2004). Benefit of very early referral and very early therapy with disease- modifying anti-rheumatic drugs in patients with early rheumatoid arthritis. Rheumatology (Oxford England), 43(7), 906-914. NICE. (2009). Rheumatoid Arthritis National clinical guideline for management and treatment in adults. In. (Reprinted from: Not in File).NICE. (2015, December 2015). Rheumatoid arthritis in adults: management. Retrieved from 'Dell, J. (2004). Therapeutic strategies for rheumatoid arthritis. The New England journal of medicine, 350(25), 2591-2602. O'Hanlon, T. P., Li, Z., Gan, L., Gourley, M. F., Rider, L. G., & Miller, F. W. (2011). Plasma proteomic profiles from disease-discordant monozygotic twins suggest that molecular pathways are shared in multiple systemic autoimmune diseases. Arthritis research & therapy, 13(6), R181. Park, G. S., Wong, W. K., Oh, M., Khanna, D., Gold, R. H., Sharp, J. T., & Paulus, H. E. (2007). Classifying radiographic progression status in early rheumatoid arthritis patients using propensity scores to adjust for baseline differences. Statistical methods in medical research 16(1), 13-29. Pasare, C., & Medzhitov, R. (2004). Toll-like receptors: linking innate and adaptive immunity. Microbes and Infection, 6(15), 1382-1387. Pearson, K. (1901). On lines and planes closest fit to a system of points in space. Philos. Mag., 2, 559-572. Pedersen, M., Jacobsen, S. r., Klarlund, M., Pedersen, B., V, Wiik, A., Wohlfahrt, J., & Frisch, M. (2006). Environmental risk factors differ between rheumatoid arthritis with and without auto-antibodies against cyclic citrullinated peptides. Arthritis research & therapy, 8(4), R133. Philbin, E. F., Ries, M. D., & French, T. S. (1995). Feasibility of maximal cardiopulmonary exercise testing in patients with end-stage arthritis of the hip and knee prior to total joint arthroplasty. Chest, 108(1), 174-181. Pinals, R. S., Masi, A. T., & Larsen, R. A. (1981). Preliminary criteria for clinical remission in rheumatoid arthritis. Arthritis Rheum, 24(10), 1308-1315. Ranganath, V., Elashoff, D., Khanna, D., Park, G., Peter, J., Paulus, H., & rhuemotologists., W. c. o. p. (2005). Age adjustment corrects for apparent differences in erythrocyte sedimentation rate and C-reactive protein values at the onset of seropositive rheumatoid arthritis in younger and older patients. The Journal of rheumatology, 32(6), 1040-1042. Rindfleisch, J. A., & Muller, D. (2005). Diagnosis and management of rheumatoid arthritis. Am. Fam. Physician, 72(6), 1037-1047. Rioja, I., Hughes, F., Sharp, C., Warnock, L., Montgomery, D., Akil, M., . . . Dickson, M. C. (2008). Potential novel biomarkers of disease activity in rheumatoid arthritis patients: CXCL13, CCL23, transforming growth factor alpha, tumor necrosis factor receptor superfamily member 9, and macrophage colony-stimulating factor. . Arthritis Rheum, 58(8), 2257-2267. Rousseau, R., Govaerts, B., Verleysen, M., & Boulanger, B. (2008). Comparison of some chemometric tools for metabonomics biomarker identification. Chemometrics and Intelligent Laboratory Systems, 91(1), 54-66. doi:10.1016/j.chemolab.2007.06.008Rousseau, R., Govaerts, B., Verleysen, M., & Boulanger, B. (2008). Comparison of some chemometric tools for metabonomics biomarker identification. Chemometrics and Intelligent Laboratory Systems, 91, 54-66. Ruffing, V., & Clifton, O. B. (2016, 13 January 2016). Rheumatoid arthritis – clinical presentation Retrieved from , K. G., Koehnke, R., Caldwell, J. R., Brasington, R., Burmeister, L. F., Zimmerman, B., . . . Furst, D. E. (1994). Low dose long-term corticosteroid therapy in rheumatoid arthritis: an analysis of serious adverse events. The American journal of medicine, 96(2), 115-123. Scott, D. L. (2000). Prognostic factors in early rheumatoid arthritis. Rheumatology (Oxford England), 39(suppl. 1), 6. Senn, S., & Julious, S. (2009). Measurement in clinical trials: A neglected issue for statisticians? Stat Med, 28, 3189-3209. Sertkaya, A., Wong, H.,Jessup, A., Beleche, T. (2016). Key cost drivers of pharmaceutical clinical trials in the United States. Clinical Trials, 13(2), 117-126. Sharp, J. T., Young, D. Y., Bluhm, G. B., Brook, A., Brower, A. C., Corbett, M., . . . Goodman, N. (1985). How many joints in the hands and wrists should be included in a score of radiologic abnormalities used to assess rheumatoid arthritis? Arthritis and Rheumatism, 28(12), 1326-1335. Sihvonen, S., Korpela, M., Laippala, P., Mustonen, J., & Pasternack, A. (2004). Death rates and causes of death in patients with rheumatoid arthritis: a population-based study. Scandinavian journal of rheumatology, 33(4), 221-227. Silman, A. J., & Pearson, J. E. (2002). Epidemiology and genetics of rheumatoid arthritis. Arthritis Res, 4 Suppl 3, S265-S272. Silveira, I. G., Burlingame, R. W., von, M. h., Bender, A. L., & Staub, H. L. (2007). Anti-CCP antibodies have more diagnostic impact than rheumatoid factor RF in a population tested for RF. Clinical rheumatology, 26(11), 1883-1889. Simon, R., & Maitoutnam, A. (2004). Evaluating the Efficiency of Targeted Designs for Randomized Clinical Trials. Clinical Cancer Research, 10, 6759-6763. Skoldstam, L., Hagfors, L., & Johansson, G. (2003). An experimental study of a Mediterranean diet intervention for patients with rheumatoid arthritis. Annals of the Rheumatic Diseases, 62(3), 208-214. Smith, M. K., Marshall, A. (2010). Importance of protocols for simulation studies in clinical drug development. Statistical Methods in Medical Research, 00, 1-12. Smolen, J., Aletaha, D., Grisar, J., Redlich, K., Steiner, G., & Wagner, O. (2008). The need for prognosticators in rheumatoid arthritis. Biological and clinical markers: where are we now? . Arthritis Res Ther, 10(3), 208-219. Smolen, J., & Steiner, G. (2003). Therapeutic Strategies for Rheumatoid Arthritis. In Nature Reviews Drug Discovery (Vol. 2, pp. 473-488). (Reprinted from: Not in File).Sokka, T., & Pincus, T. (2005). Quantitative joint assessment in rheumatoid arthritis. Clinical and Experimental Rheumatology, 23(S39), S58-S62. Sokoloff, L., & Varma, A. A. (1988). Chondrocalcinosis in surgically resected joints. Arthritis and Rheumatism, 1988(31), 6. Sugiyama, D., Nishimura, K., Tamaki, K., Tsuji, G., Nakazawa, T., Morinobu, A., & Kumagai, S. (2010). Impact of smoking as a risk factor for developing rheumatoid arthritis: a meta-analysis of observational studies. Annals of the rheumatic diseases, 69(1), 70-81. Suurmeijer, T. P., Doeglas, D. M., Moum, T., Brian?on, S., Krol, B., Sanderman, R., . . . van den Heuvel, W. J. (1994). The Groningen Activity Restriction Scale for measuring disability: its utility in international comparisons. American journal of public health, 84(8), 1270-1273. Symmons, D. P., Bankhead, C. R., Harrison, B. J., Brennan, P., Barrett, E. M., Scott, D. G., & Silman, A. J. (1997). Blood transfusion, smoking, and obesity as risk factors for the development of rheumatoid arthritis: results from a primary care-based incident case-control study in Norfolk, England. Arthritis and Rheumatism, 40(11), 1955-1961. Symmons, D. P., Barrett, E. M., Bankhead, C. R., Scott, D. G., & Silman, A. J. (1994). The incidence of rheumatoid arthritis in the United Kingdom: results from the Norfolk Arthritis Register. British journal of rheumatology, 33(8), 735-739. Szilasiová, A., Macejová, Z., Nagyová, I., ., Kovarová, M., Béresová, A., & J., S. (2002). Reliability and validation of the Slovak modified version of the Stanford Health Assessment Questionnaire using the functional disability index in patients with rheumatoid arthritis Vnitr?ní lékar?ství 48(1), 8-16. Therneau, T., Atkinson, B., & Ripley, B. (2017). rpart: Recursive Partitioning and Regression Trees. R package version 4.1-11. Retrieved from , R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society, Series B, 58(1), 267-288. Uhlig, T., Hagen, K. B., & Kvien, T. K. (1999). Current tobacco smoking, formal education, and the risk of rheumatoid arthritis. The Journal of rheumatology, 26(1), 47-54. van der Heijde, D., Hof, M. A., P., v. R., Theunisse, L. A., Lubberts, E. W., van Leeuwen, M. A., . . . van de Putte, L. (1990). Judging disease activity in clinical practice in rheumatoid arthritis: first step in the development of a disease activity score. Annals of the Rheumatic Diseases, 49(11), 916-920. van der Helm, A. (2007). Emerging Patterns of Risk Factor Make-Up Enable Subclassification of Rheumatoid Arthritis. In ARTHRITIS & RHEUMATISM (Vol. 56, pp. 1728-1735). (Reprinted from: Not in File).van Gestel, A. M., Prevoo, M. L., van T Hof, M. A., van Rijswijk, M. H., van de Putte, L. B. A., & van Riel, P. L. C. M. (1996). Development and validation of the European League Against Rheumatism response criteria for rheumatoid arthritis Comparison with the preliminary American College of Rheumatology and the World Health Organization/International League Against Rheumatism Criteria. Arthritis and Rheumatism, 39(1), 34-40. Voll, R., & Burkhardt, H. (2000). Prospective multicenter observational study of early rheumatoid arthritis--prognostic factors and predictors of disease course. Zeitschrift fnr Rheumatologie, 59(2), 113-116. Voll, R., & Kalden, J. (2005). Do we need new treatment that goes beyond tumor necrosis factor blockers for rheumatoid arthritis? Annals of the New York Academy of Sciences, 1051, 799-810. Ward, M. (2004). Clinical and Laboratory Measures. In E. W. St Clair (Ed.), Rheumatoid Arthiritis. USA: Lippincott Williams and Wilkins.Ware, J. E., Keller, S. D., Hatoum, H. T., & Kong, S. X. (1999). The SF-36 Arthritis-Specific Health Index (ASHI): I. Development and cross-validation of scoring algorithms. Medical care 37(5), MS40-50. . Watkins, K. W., Shifren, K., Park, D. C., & Morrell, R. W. (1999). Age, pain, and coping with rheumatoid arthritis. Pain, 82(3), 217-228. WHO. (2008). The global burden of disease (updated 2008):. In. (Reprinted from: Not in File).Wild, N., Karl, J., Grunert, V., Schmitt, R. I., Garczarek, U., Krause, F., . . . Zolg, W. (2008). Diagnosis of rheumatoid arthritis: Multivariate analysis of biomarkers. Biomarkers, 13(1), 88-105. Wild, N., Karl, J., Grunert, V. P., Schmitt, R. I., Garczarek, U., Krause, F., . . . Zolg, W. (2008). Diagnosis of rheumatoid arthritis: multivariate analysis of biomarkers. Biomarkers, 13(1), 88-105. doi:10.1080/13547500701669410Wiles, N., Dunn, G., Barrett, E., Silman, A., & Symmons, D. (2000). Associations between demographic and disease-related variables and disability over the first five years of inflammatory polyarthritis: a longitudinal analysis using generalized estimating equations. . Journal of Clinical Epidemiology, 53(10), 988-996. Wilson, K., & Goldsmith, C. H. (1999). Does smoking cause rheumatoid arthritis? The Journal of rheumatology, 26(1), 1-3. Wolfe, F., & Michaud, K. (2009). Predicting depression in rheumatoid arthritis: the signal importance of pain extent and fatigue, and comorbidity. Arthritis and Rheumatism, 61(5), 667-673. Wolfe, F., Pincus, T., & O'Dell, J. (2001). Evaluation and documentation of rheumatoid arthritis disease status in the clinic: which variables best predict change in therapy. The Journal of rheumatology 28(7), 1712-1717. Woolf, A., & Pfleger, B. (2010). Burden of major musculoskeletal conditions. In Bulletin of the World Health Organization. (Reprinted from: Not in File).Yellen, S. B., Cella, D. F., Webster, K., Blendowski, C., & Kaplan, E. (1997). Measuring fatigue and other anemia-related symptoms with the Functional Assessment of Cancer Therapy FACT measurement system. Journal of pain and symptom management, 13(2), 63-74. Yokoyama, E., & Muto, M. (2006). Adult variant of self-healing papular mucinosis in a patient with rheumatoid arthritis: predominant proliferation of dermal dendritic cells expressing CD34 or factor XIIIa in association with dermal deposition of mucin. The Journal of dermatology 33(1), 30-35. Appendices Appendix A: Systematic ReviewsA.1.Multivariate Methods Systematic ReviewsThere are various risk factors known to predict an individual’s propensity to develop Rheumatoid Arthritis (RA). Alamanos ADDIN EN.CITE <EndNote><Cite><Author>Alamanos</Author><Year>2005</Year><RecNum>14</RecNum><DisplayText>(Alamanos & Drosos, 2005)</DisplayText><record><rec-number>14</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1443534719">14</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Alamanos, Yannis</author><author>Drosos, Alexandros</author></authors></contributors><auth-address>Department of Hygiene and Epidemiology, Medical School, University of Ioannina, Ioannina, Greece</auth-address><titles><title>Epidemiology of adult rheumatoid arthritis</title><secondary-title>Autoimmunity reviews</secondary-title></titles><periodical><full-title>Autoimmunity reviews</full-title></periodical><pages>130-136</pages><volume>4</volume><number>3</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>ARTHRITIS-RHEUMATOID/*EP (epidemiology),*ET (etiology)</keyword><keyword>HUMANS</keyword><keyword>INCIDENCE</keyword><keyword>PREVALENCE</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RISK-FACTORS</keyword></keywords><dates><year>2005</year><pub-dates><date>3/2005</date></pub-dates></dates><label>15</label><urls></urls><access-date>2004</access-date></record></Cite></EndNote>(Alamanos & Drosos, 2005) discusses the factors smoking history, age, gender, genetics, socioeconomics, geography and ethnicity. Many papers discuss these factors in a univariate fashion, however very few, if any, discuss these factors in a multivariate way. A more sophisticated understanding of the influences of RA may lead to a better approach to personalised healthcare for patients with RA.These systematic reviews seek to identify through available literature which biological factors influence progression of the disease and to summarise methods and results that are assessed in a multivariate analysis.The review considered sources relating to an adult RA population, patients with juvenile variations of RA (e.g. systemic juvenile idiopathic arthritis (sJIA)) and related inflammatory diseases (e.g. Lupus, Castleman’s disease, psoriatic arthritis) were excluded.A.1.1.Medline Systematic ReviewAimsTo identify patient characteristics considered to influence progression of RATo review methods used to draw out these conclusionsTo identify where this has been analysed in a non-univariate mannerMethodsA systematic review was undertaken according to the general principles outlined in the QUOROM statement ADDIN EN.CITE <EndNote><Cite><Author>Moher</Author><Year>1999</Year><RecNum>120</RecNum><DisplayText>(Moher et al., 1999)</DisplayText><record><rec-number>120</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">120</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Moher, D.</author><author>Cook, D.J.</author><author>Eastwood, S.</author><author>Olkin, I.</author><author>Rennie, D.</author><author>Stroup, D.F.</author></authors></contributors><auth-address>University of Ottawa, Thomas C Chalmers Centre for Systematic Reviews, Ontario, Canada. dmoher@uottawa.ca</auth-address><titles><title>Improving the quality of reports of meta-analyses of randomised controlled trials: the QUOROM statement Quality of Reporting of Meta- analyses</title><secondary-title>Lancet</secondary-title></titles><pages>1896-1900</pages><volume>354</volume><number>9193</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>AUTHORSHIP</keyword><keyword>Canada</keyword><keyword>GUIDELINES-AS-TOPIC</keyword><keyword>HUMANS</keyword><keyword>META-ANALYSIS-AS-TOPIC/*</keyword><keyword>methods</keyword><keyword>RANDOMIZED-CONTROLLED-TRIALS-AS-TOPIC</keyword><keyword>REVIEW</keyword><keyword>SOFTWARE-DESIGN</keyword><keyword>standards</keyword></keywords><dates><year>1999</year><pub-dates><date>11/27/1999</date></pub-dates></dates><label>190</label><urls></urls><access-date>2002</access-date></record></Cite></EndNote>(Moher et al., 1999). The QUOROM Statement is specifically aimed at systematic reviews of randomised clinical trials. Principles will be applied in this more general context.Search StrategyA search of electronic databases including MEDLINE, EMBASE. Inclusion criteria Adults diagnosed with RAStudies investigating response in subgroups of patientsStudy TypesSystematic ReviewsRandomised controlled trials Cohort studiesCase control studiesCross sectional surveysExclusion CriteriaJuvenile patients Non RA diseases such as Lupus, Castleman’s disease, psoriatic arthritis Non English referencesResults Medline REF _Ref461277539 \h \* MERGEFORMAT Table A0.1 shows the results of a search performed using Medline. This search identified 113 articles.Table A STYLEREF 1 \s 0. SEQ Table \* ARABIC \s 1 1 Medline Search for statistical methods used to analyse risk factors in RASearch TermResults 1Rheumatoid Arthritis.mp. or Arthritis, Rheumatoid889312limit 1 to ("all adult (19 plus years)" or "young adult (19 to 24 years)" or "adult (19 to 44 years)" or "young adult and adult (19-24 and 19-44)" or "middle age (45 to 64 years)" or "middle aged (45 plus years)" or "all aged (65 and over)" or "aged (80 and over)")403363cluster analysis/ or analysis/ or factor analysis, statistical/494434cart.tw237953 or 45178262 and 5113The 113 articles were reviewed against the following criteria:Are the participants adult?Are the participants diagnosed with Rheumatoid Arthritis (RA)?Is diagnosis of RA explicit or implicit?What endpoints are analysed in the paper?Are baseline characteristics or clinical biomarkers specified?Are Multivariate methods appliedFigure A SEQ Figure \* ARABIC \s 1 1 Results of Medline Review -66675301625Results from Medline (n=113)Not adult(n= 8)Not RA(n= 37)No bio/ clinical markers (n= 90)Adult RA Marker analysis (n= 13)For further discussion (n= 5)Univariate Methods(n=8)00Results from Medline (n=113)Not adult(n= 8)Not RA(n= 37)No bio/ clinical markers (n= 90)Adult RA Marker analysis (n= 13)For further discussion (n= 5)Univariate Methods(n=8)Full articles were obtained and reviewed since some of the required information was not typically available in the article abstracts. Despite the search terms, the review of the 113 articles identified 37 articles not based on RA patients and 8 articles not based on data from adults. Ninety articles did not explore the impact of clinical or biomarker on RA outcomes. Of the remaining 13 articles, 8 only dealt with simple univariate analyses ( REF _Ref461280148 \h \* MERGEFORMAT Figure A1).Discussion of Medline results The final 5 articles and their titles identified by the Medline systematic review were:Psychological stress as a predictor of psychological adjustment and health status in patients with rheumatoid arthritis. ADDIN EN.CITE <EndNote><Cite><Author>Curtis</Author><Year>2005</Year><RecNum>138</RecNum><DisplayText>(Curtis et al., 2005)</DisplayText><record><rec-number>138</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473702730">138</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Curtis, R.</author><author>Groarke, A.</author><author>Coughlan, R. </author><author>Gsel, A. </author></authors></contributors><titles><title>Psychological stress as a predictor of psychological adjustment and health status in patients with rheumatoid arthritis</title><secondary-title>Patient Education & Counseling</secondary-title></titles><periodical><full-title>Patient Education & Counseling</full-title></periodical><pages>192-8</pages><volume>59</volume><number>2</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(Curtis et al., 2005)Traditional and non-traditional cardiovascular risk factors are associated with atherosclerosis in rheumatoid arthritis. ADDIN EN.CITE <EndNote><Cite><Author>Dessein</Author><Year>2005</Year><RecNum>139</RecNum><DisplayText>(Dessein et al., 2005)</DisplayText><record><rec-number>139</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">139</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Dessein, P.H.</author><author>Joffe, B.I.</author><author>Veller, M.G.</author><author>Stevens, B.A.</author><author>Tobias, M.</author><author>Reddi, K</author><author>Stanwix, A.E.</author></authors></contributors><titles><title>Traditional and Nontraditional Cardiovascular Risk Factors Are Associated with Atherosclerosis in Rheumatoid Arthritis</title><secondary-title>Journal of Rheumatology</secondary-title></titles><pages>435-42</pages><volume>32</volume><number>3</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(Dessein et al., 2005)Evaluation and documentation of rheumatoid arthritis disease status in the clinic: which variables best predict change in therapy? ADDIN EN.CITE <EndNote><Cite><Author>Wolfe</Author><Year>2001</Year><RecNum>140</RecNum><DisplayText>(Wolfe et al., 2001)</DisplayText><record><rec-number>140</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">140</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Wolfe, F.</author><author>Pincus, T.</author><author>O'Dell, J.</author></authors></contributors><titles><title>Evaluation and documentation of rheumatoid arthritis disease status in the clinic: which variables best predict change in therapy</title><secondary-title>The Journal of rheumatology </secondary-title></titles><pages>1712-7</pages><volume>28</volume><number>7</number><dates><year>2001</year></dates><urls></urls></record></Cite></EndNote>(Wolfe et al., 2001)Associations between demographic and disease-related variables and disability over the first five years of inflammatory polyarthritis: a longitudinal analysis using generalized estimating equations. ADDIN EN.CITE <EndNote><Cite><Author>Wiles</Author><Year>2000</Year><RecNum>141</RecNum><DisplayText>(Wiles et al., 2000)</DisplayText><record><rec-number>141</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">141</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Wiles, N..</author><author>Dunn, G.,</author><author>Barrett, E.,</author><author>Silman, A.,</author><author>Symmons, D. </author></authors></contributors><titles><title>Associations between demographic and disease-related variables and disability over the first five years of inflammatory polyarthritis: a longitudinal analysis using generalized estimating equations. </title><secondary-title>Journal of Clinical Epidemiology</secondary-title></titles><pages>988-96</pages><volume>53</volume><number>10</number><dates><year>2000</year></dates><urls></urls></record></Cite></EndNote>(Wiles et al., 2000)Age, pain, and coping with rheumatoid arthritis. ADDIN EN.CITE <EndNote><Cite><Author>Watkins</Author><Year>1999</Year><RecNum>142</RecNum><DisplayText>(Watkins et al., 1999)</DisplayText><record><rec-number>142</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">142</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Watkins, K.W.</author><author>Shifren, K.</author><author>Park, D.C.</author><author>Morrell, R.W. </author></authors></contributors><titles><title>Age, pain, and coping with rheumatoid arthritis.</title><secondary-title>Pain</secondary-title></titles><pages>217-28</pages><volume>82</volume><number>3</number><dates><year>1999</year></dates><urls></urls></record></Cite></EndNote>(Watkins et al., 1999)Curtis et al studied the extent to which perceived stress, social support, coping and clinical disease indicators predicted adjustment to disease in patients with RA. Whilst this research involved adults with RA and considered markers such as ESR and joint counts, it focused on predicting stress as an outcome rather than disease diagnosis or progression. Nonetheless the methods used merit further consideration. In the first step Pearson correlations were used to assess the relationship between disease indicators and psychological variables. Secondly hierarchical regression analyses identified sets of variables significantly related to outcome. Finally, a principal components analysis was conducted to reduce the number of predictors in the final model. Curtis et al claimed to demonstrate predictors of stress but did not examine how those predictors interacted. The sample size of this study was also quite small (n=59), making the use of techniques such as principal component analysis of limited utility. Dessein et al aimed to determine the association between cardiovascular risk factors and atherosclerosis in patients with RA. Their research included 74 adult patients with RA and considered clinical and laboratory disease markers. In their article Dessein et al compared characteristics of patients with and without atherosclerosis and with or without plaque using Mann Whitney U tests or Chi-Square tests. Associations were examined using simple logistic regression. Variables with the strongest univariate associations were considered in multi-variable models. Alternate models were explored using classification and regression tree (CART) analysis. Finally, factor analysis with varimax rotation was applied to see whether predictors associated with atherosclerosis or plaque were associated with the same factored RA characteristic. Dessein et al’s study presented some useful methods for looking at disease risk factors although, as with Curtis et al, the sample size is small given the statistical methods applied. There is some consideration to the relationship between predictor variables through the use of Factor analysis, however the exploration and discussion of such relationships was quite limited. Wolfe, Pincus and O’Dell studied a large number of adult RA patients (n=1905) to determine which disease status variables were most associated with a change in Disease-Modifying Anti-Rheumatic Drug (DMARD) therapy. Aside from the outcome measure there were close parallels to the intent of the research of this thesis. The strength of association between individual clinical predictor variables and outcome (treatment changes) were analysed using generalised estimating equations (GEE). GEE is a method that enables, amongst other things, analysis of categorical repeated measures. Wolfe, Pincus and O’Dell claimed to look at multivariate associations using GEE in a logistic regression model although any such association is only displayed in a 1st order manner (no interaction terms). In addition to GEE, CART analyses were performed to explore non-linear predictor variables. This study achieved its aim to identify variables associated with change in therapy but despite the use of potentially sophisticated methods, applied them in a basic manner.Wiles et al studied a population which included a subset of RA adult patients and used demographic and disease related variables to investigate their relationship with disability during the first five years of disease. Aware of the high degree of correlation between certain variables, the researchers first applied a principal components analysis to eliminate collinearity problems and to reduce the number of variables included in later models. The research discussed the importance of examining factors over time rather than at arbitrary points in time. Wiles et al proposed the use of GEE to fully utilise the full potential of longitudinal data. An odds ratio was calculated for each variable and assumed to be constant over time, therefore after modelling, the interaction between clinical variables and time was assessed by introducing a multiplicative term into the GEE model. Again this research answered a different question to the proposed aims of this thesis, but the methodology and approach may useful to replicate.Watkins et al examined the impact of age and pain severity in predicting coping strategies in a modest sample (n=121) of adult RA patients. Although the outcome of coping strategy was not of interest in this thesis, there were other aspects of this study of interest. Watkins et al analysed dependent measures in a “doubly multivariate analysis of variance” using age as a between-subjects factor and pain severity as a within-subjects factor. This allowed the researchers to study age and pain severity interaction in predicting outcomes. Possible confounding variables such as disease duration were also analysed. Lastly hierarchical regression analyses were conducted to determine the extent to which outcomes were associated with demographic factors and coping strategies. A.1.2.Embase Systematic ReviewFollowing systematic review using Medline above, a second systematic review was performed using Embase. An exploratory run of the same Medline criteria on Embase generated 178 hits instead of 113. In the Medline review the ratio of initial hits to final shortlist was high (113: 5), therefore the criteria were refined to focus on biological markers. REF _Ref461277601 \h \* MERGEFORMAT Table A0.2 shows the results of a search performed using Embase. This search identified 7 articles. Table A STYLEREF 1 \s 0. SEQ Table \* ARABIC \s 1 2 Embase Search for statistical methods used to analyse risk factors in RASearch TermResults 1Rheumatoid-Arthritis.de.1196162Adult.w..de or middle-aged.w..de423957631 and 2340544Multivariate-Analysis.de591345Classification.w..de 2231846Discriminant-Analysis.de823174 or 5 or 62883768Biological-Marker.de7294093 and 7 and 87Discussion of Embase ResultsThe 7 articles with their titles identified by the Embase systematic review were:The need for personalised medicine for rheumatoid arthritis. PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Jc2FhY3M8L0F1dGhvcj48WWVhcj4yMDExPC9ZZWFyPjxS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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Jc2FhY3M8L0F1dGhvcj48WWVhcj4yMDExPC9ZZWFyPjxS
ZWNOdW0+MTA4PC9SZWNOdW0+PERpc3BsYXlUZXh0PihJc2FhY3MgJmFtcDsgRmVycmFjY2lvbGks
IDIwMTEpPC9EaXNwbGF5VGV4dD48cmVjb3JkPjxyZWMtbnVtYmVyPjEwODwvcmVjLW51bWJlcj48
Zm9yZWlnbi1rZXlzPjxrZXkgYXBwPSJFTiIgZGItaWQ9IjA5cHY1YXpwamV2dmRoZTVhczF2dnN6
eXdmZnNyZnd0ZTVlNSIgdGltZXN0YW1wPSIxNDQzNTM0OTgwIj4xMDg8L2tleT48L2ZvcmVpZ24t
a2V5cz48cmVmLXR5cGUgbmFtZT0iSm91cm5hbCBBcnRpY2xlIj4xNzwvcmVmLXR5cGU+PGNvbnRy
aWJ1dG9ycz48YXV0aG9ycz48YXV0aG9yPklzYWFjcywgSm9objwvYXV0aG9yPjxhdXRob3I+RmVy
cmFjY2lvbGksIEdpYW5mcmFuY288L2F1dGhvcj48L2F1dGhvcnM+PC9jb250cmlidXRvcnM+PGF1
dGgtYWRkcmVzcz5KLiBELiBJc2FhY3M6IEluc3RpdHV0ZSBvZiBDZWxsdWxhciBNZWRpY2luZSwg
TXVzY3Vsb3NrZWxldGFsIFJlc2VhcmNoIEdyb3VwLCBNZWRpY2FsIFNjaG9vbCwgRnJhbWxpbmd0
b24gUGxhY2UsIE5ld2Nhc3RsZSBVcG9uIFR5bmUgTkUyIDRISCwgVUsuIEVtYWlsOiBqLmQuaXNh
YWNzQG5jbC5hYy51azwvYXV0aC1hZGRyZXNzPjx0aXRsZXM+PHRpdGxlPlRoZSBuZWVkIGZvciBw
ZXJzb25hbGlzZWQgbWVkaWNpbmUgZm9yIHJoZXVtYXRvaWQgYXJ0aHJpdGlzPC90aXRsZT48c2Vj
b25kYXJ5LXRpdGxlPkFubmFscyBvZiB0aGUgcmhldW1hdGljIGRpc2Vhc2VzPC9zZWNvbmRhcnkt
dGl0bGU+PC90aXRsZXM+PHBlcmlvZGljYWw+PGZ1bGwtdGl0bGU+QW5uYWxzIG9mIHRoZSByaGV1
bWF0aWMgZGlzZWFzZXM8L2Z1bGwtdGl0bGU+PC9wZXJpb2RpY2FsPjxwYWdlcz40LTc8L3BhZ2Vz
Pjx2b2x1bWU+NzA8L3ZvbHVtZT48bnVtYmVyPjE8L251bWJlcj48cmVwcmludC1lZGl0aW9uPk5v
dCBpbiBGaWxlPC9yZXByaW50LWVkaXRpb24+PGtleXdvcmRzPjxrZXl3b3JkPjE1NDc1LTU2LTYg
NTktMDUtMiA3NDEzLTM0LTUgKG1ldGhvdHJleGF0ZSk8L2tleXdvcmQ+PGtleXdvcmQ+MTcwMjc3
LTMxLTMgKGluZmxpeGltYWIpPC9rZXl3b3JkPjxrZXl3b3JkPjE3NDcyMi0zMS03IChyaXR1eGlt
YWIpPC9rZXl3b3JkPjxrZXl3b3JkPjE4NTI0My02OS0wIDIwMDAxMy04Ni0xIChldGFuZXJjZXB0
KTwva2V5d29yZD48a2V5d29yZD4zMzE3MzEtMTgtMSAoYWRhbGltdW1hYik8L2tleXdvcmQ+PGtl
eXdvcmQ+Mzc1ODIzLTQxLTkgKHRvY2lsaXp1bWFiKTwva2V5d29yZD48a2V5d29yZD41MC0yNC04
IChQcmVkbmlzb2xvbmUpPC9rZXl3b3JkPjxrZXl3b3JkPjkwMDktNzktNCAocmhldW1hdG9pZCBm
YWN0b3IpPC9rZXl3b3JkPjxrZXl3b3JkPjk3Nzk0LTI3LTkgKGltbXVub2dsb2J1bGluIEcpPC9r
ZXl3b3JkPjxrZXl3b3JkPkFEQUxJTVVNQUIqL0RUIChkcnVnIHRoZXJhcHkpPC9rZXl3b3JkPjxr
ZXl3b3JkPkFEVUxUPC9rZXl3b3JkPjxrZXl3b3JkPkFOVElCT0RZLVBST0RVQ1RJT048L2tleXdv
cmQ+PGtleXdvcmQ+QU5USUJPRFktVElURVI8L2tleXdvcmQ+PGtleXdvcmQ+QXJ0aHJpdGlzLWFu
ZC1SaGV1bWF0aXNtPC9rZXl3b3JkPjxrZXl3b3JkPkFVVE9BTlRJQk9EWS9FQyAoZW5kb2dlbm91
cyBjb21wb3VuZCk8L2tleXdvcmQ+PGtleXdvcmQ+Qi1MWU1QSE9DWVRFPC9rZXl3b3JkPjxrZXl3
b3JkPkJJT0xPR0lDQUwtTUFSS0VSL0VDIChlbmRvZ2Vub3VzIGNvbXBvdW5kKTwva2V5d29yZD48
a2V5d29yZD5DSFJPTklDLURJU0VBU0UvRE0gKGRpc2Vhc2UgbWFuYWdlbWVudCk8L2tleXdvcmQ+
PGtleXdvcmQ+Q09TVC1DT05UUk9MPC9rZXl3b3JkPjxrZXl3b3JkPkNZQ0xJQy1DSVRSVUxMSU5B
VEVELVBFUFRJREUtQU5USUJPRFkvRUMgKGVuZG9nZW5vdXMgY29tcG91bmQpPC9rZXl3b3JkPjxr
ZXl3b3JkPkRpc2Vhc2U8L2tleXdvcmQ+PGtleXdvcmQ+RElTRUFTRS1DT1VSU0U8L2tleXdvcmQ+
PGtleXdvcmQ+RElTRUFTRS1NT0RJRllJTkctQU5USVJIRVVNQVRJQy1EUlVHL0RUIChkcnVnIHRo
ZXJhcHkpPC9rZXl3b3JkPjxrZXl3b3JkPkRydWctTGl0ZXJhdHVyZS1JbmRleDwva2V5d29yZD48
a2V5d29yZD5FVEFORVJDRVBUL0RUIChkcnVnIHRoZXJhcHkpPC9rZXl3b3JkPjxrZXl3b3JkPkZF
TUFMRTwva2V5d29yZD48a2V5d29yZD5IRUFMVEgtQVNTRVNTTUVOVC1RVUVTVElPTk5BSVJFPC9r
ZXl3b3JkPjxrZXl3b3JkPkhFQUxUSC1DQVJFLUNPU1Q8L2tleXdvcmQ+PGtleXdvcmQ+SGVhbHRo
LVBvbGljeS1FY29ub21pY3MtYW5kLU1hbmFnZW1lbnQ8L2tleXdvcmQ+PGtleXdvcmQ+SFVNQU48
L2tleXdvcmQ+PGtleXdvcmQ+SU1NVU5FLVNZU1RFTTwva2V5d29yZD48a2V5d29yZD5JTU1VTk9H
TE9CVUxJTi1HL0VDIChlbmRvZ2Vub3VzIGNvbXBvdW5kKTwva2V5d29yZD48a2V5d29yZD5JbW11
bm9sb2d5LVNlcm9sb2d5LWFuZC1UcmFuc3BsYW50YXRpb248L2tleXdvcmQ+PGtleXdvcmQ+SU5G
TElYSU1BQiovRFQgKGRydWcgdGhlcmFweSk8L2tleXdvcmQ+PGtleXdvcmQ+SU5OQVRFLUlNTVVO
SVRZPC9rZXl3b3JkPjxrZXl3b3JkPkludGVybmFsLU1lZGljaW5lPC9rZXl3b3JkPjxrZXl3b3Jk
Pk1BQ1JPUEhBR0U8L2tleXdvcmQ+PGtleXdvcmQ+TUFKT1ItQ0xJTklDQUwtU1RVRFk8L2tleXdv
cmQ+PGtleXdvcmQ+TUFMRTwva2V5d29yZD48a2V5d29yZD5NRVRIT1RSRVhBVEUvRFQgKGRydWcg
dGhlcmFweSk8L2tleXdvcmQ+PGtleXdvcmQ+TUlDUk9BUlJBWS1BTkFMWVNJUzwva2V5d29yZD48
a2V5d29yZD5NVUxUSVZBUklBVEUtQU5BTFlTSVM8L2tleXdvcmQ+PGtleXdvcmQ+T1VUQ09NRS1B
U1NFU1NNRU5UPC9rZXl3b3JkPjxrZXl3b3JkPlBFUlNPTkFMSVpFRC1NRURJQ0lORSo8L2tleXdv
cmQ+PGtleXdvcmQ+UEhBU0UtMy1DTElOSUNBTC1UUklBTDwva2V5d29yZD48a2V5d29yZD5QTEFT
TUEtQ0VMTDwva2V5d29yZD48a2V5d29yZD5QUkVESUNUSU9OPC9rZXl3b3JkPjxrZXl3b3JkPlBS
RUROSVNPTE9ORS9EVCAoZHJ1ZyB0aGVyYXB5KTwva2V5d29yZD48a2V5d29yZD5QUklPUklUWS1K
T1VSTkFMPC9rZXl3b3JkPjxrZXl3b3JkPlJFTUlTU0lPTjwva2V5d29yZD48a2V5d29yZD5SRVZJ
RVc8L2tleXdvcmQ+PGtleXdvcmQ+UmhldW1hdG9pZCBhcnRocml0aXM8L2tleXdvcmQ+PGtleXdv
cmQ+UkhFVU1BVE9JRC1BUlRIUklUSVMqL0RNIChkaXNlYXNlIG1hbmFnZW1lbnQpLERUIChkcnVn
IHRoZXJhcHkpPC9rZXl3b3JkPjxrZXl3b3JkPlJIRVVNQVRPSUQtRkFDVE9SL0VDIChlbmRvZ2Vu
b3VzIGNvbXBvdW5kKTwva2V5d29yZD48a2V5d29yZD5SSVNLLUJFTkVGSVQtQU5BTFlTSVM8L2tl
eXdvcmQ+PGtleXdvcmQ+UklUVVhJTUFCKi9EVCAoZHJ1ZyB0aGVyYXB5KTwva2V5d29yZD48a2V5
d29yZD5TRVJVTTwva2V5d29yZD48a2V5d29yZD5TWU5PVklVTTwva2V5d29yZD48a2V5d29yZD5U
LUxZTVBIT0NZVEU8L2tleXdvcmQ+PGtleXdvcmQ+VE9DSUxJWlVNQUIqL0RUIChkcnVnIHRoZXJh
cHkpPC9rZXl3b3JkPjxrZXl3b3JkPlRSRUFUTUVOVC1SRVNQT05TRTwva2V5d29yZD48a2V5d29y
ZD5UVU1PUi1ORUNST1NJUy1GQUNUT1ItQUxQSEEtSU5ISUJJVE9SKi9EVCAoZHJ1ZyB0aGVyYXB5
KSxQRSAocGhhcm1hY29lY29ub21pY3MpPC9rZXl3b3JkPjxrZXl3b3JkPlRVTU9SLU5FQ1JPU0lT
LUZBQ1RPUi1SRUNFUFRPUi0yL0VDIChlbmRvZ2Vub3VzIGNvbXBvdW5kKTwva2V5d29yZD48a2V5
d29yZD5UVU1PUi1ORUNST1NJUy1GQUNUT1IvRUMgKGVuZG9nZW5vdXMgY29tcG91bmQpPC9rZXl3
b3JkPjwva2V5d29yZHM+PGRhdGVzPjx5ZWFyPjIwMTE8L3llYXI+PHB1Yi1kYXRlcz48ZGF0ZT4x
LzIwMTE8L2RhdGU+PC9wdWItZGF0ZXM+PC9kYXRlcz48bGFiZWw+MTc4PC9sYWJlbD48dXJscz48
L3VybHM+PC9yZWNvcmQ+PC9DaXRlPjwvRW5kTm90ZT4A
ADDIN EN.CITE.DATA (Isaacs & Ferraccioli, 2011)Potential novel biomarkers of disease activity in rheumatoid arthritis patients: CXCL13, CCL23, transforming growth factor alpha, tumour necrosis factor receptor superfamily member 9, and macrophage colony-stimulating factor. ADDIN EN.CITE <EndNote><Cite><Author>Rioja</Author><Year>2008</Year><RecNum>144</RecNum><DisplayText>(Rioja et al., 2008)</DisplayText><record><rec-number>144</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473704487">144</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rioja, I.</author><author>Hughes, F.</author><author>Sharp, C.</author><author>Warnock, L.</author><author>Montgomery, D.</author><author>Akil, M</author><author>Wilson, A.G.</author><author>Binks,M.H.</author><author>Dickson, M.C.</author></authors></contributors><titles><title>Potential novel biomarkers of disease activity in rheumatoid arthritis patients: CXCL13, CCL23, transforming growth factor alpha, tumor necrosis factor receptor superfamily member 9, and macrophage colony-stimulating factor. </title><secondary-title>Arthritis Rheum</secondary-title></titles><periodical><full-title>Arthritis Rheum</full-title></periodical><pages>2257-2267</pages><volume>58</volume><number>8</number><dates><year>2008</year></dates><urls></urls></record></Cite></EndNote>(Rioja et al., 2008)Diagnosis of rheumatoid arthritis: Multivariate analysis of biomarkers. PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5XaWxkPC9BdXRob3I+PFllYXI+MjAwODwvWWVhcj48UmVj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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5XaWxkPC9BdXRob3I+PFllYXI+MjAwODwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Wild et al., 2008)A new classification of HLA-DRB1 alleles differentiates predisposing and protective alleles for autoantibody production in rheumatoid arthritis. PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Hb3VycmF1ZDwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Hb3VycmF1ZDwvQXV0aG9yPjxZZWFyPjIwMDc8L1llYXI+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ADDIN EN.CITE.DATA (Gourraud et al., 2007)Long term anti-tumour necrosis factor alpha monotherapy in rheumatoid arthritis: Effect on radiological course and prognostic value of markers of cartilage turnover and endothelial activation. PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Ccm9lZGVyPC9BdXRob3I+PFllYXI+MjAwMjwvWWVhcj48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 EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Ccm9lZGVyPC9BdXRob3I+PFllYXI+MjAwMjwvWWVhcj48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 EN.CITE.DATA (Broeder et al., 2002)Prospective multicentre observational study of early rheumatoid arthritis--prognostic factors and predictors of disease course. ADDIN EN.CITE <EndNote><Cite><Author>Voll</Author><Year>2000</Year><RecNum>113</RecNum><DisplayText>(Voll & Burkhardt, 2000)</DisplayText><record><rec-number>113</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">113</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Voll, R.</author><author>Burkhardt, H.</author></authors></contributors><auth-address><style face="normal" font="default" size="100%">R. Voll: Medizinische Klinik III mit Poliklinik und Institut fnr Klinische Immunologie Friedrich-Alexander-Universit</style><style face="normal" font="Symbol" charset="2" size="100%">S</style><style face="normal" font="default" size="100%">t Erlangen- Nnrnberg.</style></auth-address><titles><title>Prospective multicenter observational study of early rheumatoid arthritis--prognostic factors and predictors of disease course</title><secondary-title>Zeitschrift fnr Rheumatologie</secondary-title></titles><pages>113-116</pages><volume>59</volume><number>2</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>ADULT</keyword><keyword>AGED</keyword><keyword>ARTICLE</keyword><keyword>BIOLOGICAL-MARKER/AN (drug analysis)</keyword><keyword>classification</keyword><keyword>CLINICAL-TRIAL</keyword><keyword>Disease</keyword><keyword>DISEASE-COURSE</keyword><keyword>FEMALE</keyword><keyword>GENERAL-PRACTITIONER</keyword><keyword>HUMAN</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>MULTICENTER-STUDY</keyword><keyword>PATIENT-CARE</keyword><keyword>PREDICTION</keyword><keyword>PROSPECTIVE-STUDY</keyword><keyword>Rheumatoid arthritis</keyword><keyword>RHEUMATOID-ARTHRITIS*/DI (diagnosis)</keyword><keyword>TOTAL-QUALITY-MANAGEMENT*</keyword></keywords><dates><year>2000</year><pub-dates><date>4/2000</date></pub-dates></dates><label>183</label><urls></urls></record></Cite></EndNote>(Voll & Burkhardt, 2000)Secretory phospholipase A2 as an index of disease activity in rheumatoid arthritis. Prospective double blind study of 212 patients. PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MaW48L0F1dGhvcj48WWVhcj4xOTk2PC9ZZWFyPjxSZWNO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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MaW48L0F1dGhvcj48WWVhcj4xOTk2PC9ZZWFyPjxSZWNO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ADDIN EN.CITE.DATA (Lin et al., 1996)Isaacs and Ferraccioli publication is a general discussion of personalised medicine. Whilst some predictors for response are discussed they are mentioned mostly in a univariate sense. A multivariate analysis from a paper on Rituximab is mentioned but not in detail. Interestingly the authors conclude that ‘biomarkers may be of limited value and should ideally be studied in combination with one another’ yet little is done here to examine this question.In the research by Rioja et al, the authors attempted to identify proteins that may distinguish between 'active disease’ and ‘quiescent disease’. They identified 5 proteins through partial least squares discriminant analysis that are expressed differently in the 2 disease states and found that levels of these proteins correlate well with disease activity measures such as DAS28. This research was based on 163 plasma proteins and a relatively small number of subjects (44 patients & 16 controls), in concluding the authors recommended further work to validate their findings. Wild et al investigated how combinations of biomarkers could improve the diagnosis of RA. The authors used information from a large number of subjects (n=1448) and a modest number of biomarkers (n=32) to develop models for diagnosing RA. The authors used 4 different mathematical models and ‘regularised discriminant analysis’. According to the authors this allows for combination of marker values and their interaction whilst controlling overfitting. With little detail the authors also comment that they applied and compared the performance of other statistical methods with regularised discriminant analysis such as Bayes Logistic Regression, Bayes Linear Support Vector Machine and Bayes Kernel Support Vector Machine. Gourraud et al considered a few genetic alleles for predisposition to RA. Whilst this is a study with adult RA patients and includes the biomarker Rheumatoid Factor in the statistical analysis, very simple statistics, such as Pearson’s Chi Square test, Fisher’s Exact test and odds ratios were used to test association between genetic markers and presence of disease. This paper was selected by EMBASE search due to its new classification of HLA-DRB1 SE+ alleles.In the paper by den-Broeder et al, the authors consider the effect of baseline characteristics on radiological outcome for a small number (n=36) RA patients. Whilst the statistical methods state that ‘potential interactions between independent variables’ were to be tested, this is not apparent or mentioned again in results or elsewhere in the publication. In the logistic regression that was conducted just 3 independent variables were included in the model.The paper by Voll and Burkhardt mentions no statistical methods. Lin et al correlated serum secreted phospholipase with disease activity using data from 212 patients in a retrospective study. Lin et al performed regressions through generalised estimating equations methodology. In all the models employed main effects of baseline characteristics were included but interactions were not considered. Table A STYLEREF 1 \s 0. SEQ Table \* ARABIC \s 1 3 Summary of Medline and Embase Systematic ReviewsReferenceOutcome Sample SizeStatistical MethodsExplanatory VariablesCurtis et alDisease Adjustment (longitudinal dichotomous assessment)59Pearson CorrelationHierarchical regression Principal ComponentsStressSocial SupportCopingAgeIllness DurationESR Joint Count Dessein et alAtherosclerosis (dichotomous assessment)74Mann Whitney UChi SquareLogistic regressionClassification & Regression Tree (CART)Factor analysis with Varimax rotationAgeGender RaceSmoking HistoryDisease duration, Disease severityHAQESR, CRPJoint Count, Joint space narrowingDrug therapyWaist HypertensionSystolic blood pressureTriglyceridesUric acidHypothyroidismPolymorphonuclear countGlomerular filtration rateRisk of Cardiovascular eventWolfe, Pincus and O’DellChange in therapy (longitudinal dichotomous assessment)1905Generalised Estimating Equations (GEE)Logistic regression CARTJoint countPainGlobal severityESR Grip strengthHAQDepressionAnxiety Morning stiffnessFatigueSleep DisturbanceWiles et alDisability (longitudinal dichotomous assessment)684 (325 RA)Principal ComponentsGEEAgeGenderDelay to presentationMorning StiffnessJoint countRheumatoid factor statusPresence of nodulesWatkins et alCoping (multiple outcomes)121Multivariate Analysis of Variance (MANOVA)Hierarchical regressionAge Pain Isaacs and FerraccioliDiscussion of biomarkers-None-Rioja et alActive vs Quiescent RA using DAS28(continuous assessment)163Partial least squares discriminant analysis Age GenderCRP, ESR, RFJoint countsDASConcomitant medicationWild et alDiagnosis of RA (dichotomous assessment)1448(906 RA)Receiver Operating Curves ADDIN EN.CITE <EndNote><Cite ExcludeYear="1"><Author>Casey</Author><Year>1996</Year><RecNum>155</RecNum><DisplayText>(Casey et al.)</DisplayText><record><rec-number>155</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">155</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Casey, A.T.</author><author>Bland, J.M.</author><author>Crockard, H.A. </author></authors></contributors><titles><title>Development of a functional scoring system for rheumatoid arthritis patients with cervical myelopathy</title><secondary-title>Annals of the rheumatic diseases </secondary-title></titles><periodical><full-title>Annals of the Rheumatic Diseases</full-title></periodical><pages>901-6</pages><volume>55</volume><number>12</number><dates><year>1996</year></dates><urls></urls></record></Cite></EndNote>(Casey et al.)Regularised discriminant analysis Anti-CCPGourraud et alGenetic association160Chi-Square, Fishers exact testRF, Genetic Allelesden-Broeder et alRadiological outcomes (dichotomous assessment)47t-test, Wilcoxon, Pearson correlation, logistic regression.Markers of cartilage and synovium turnoverMarkers of endothelial activationTreatment statusImmunoglobulinsRheumatoid factor Baseline DAS & Sharp scoreVoll and BurkhardtErosive disease400NoneUnknownLin et alRA Disease Activity (longitudinal continuous assessment)212GEEserum secreted phospholipaseLansbury index, EffusionsJoint Counts ESRPlatelet count, Haemoglobin A.1.3.Medline – revised criteriaDue to the differences in the search results of the Medline and Embase searches, the Embase criteria were run on Medline database. Figure A SEQ Figure \* ARABIC \s 1 2 Results of Revised Medline ReviewResults from Medline (n=37)Not adult(n= 1)Not RA(n= 6)No bio/ clinical markers (n= 6)Adult RA Marker analysis (n= 27)For further discussion (n= 25)Univariate Methods(n=2)Results from Medline (n=37)Not adult(n= 1)Not RA(n= 6)No bio/ clinical markers (n= 6)Adult RA Marker analysis (n= 27)For further discussion (n= 25)Univariate Methods(n=2)The revised Medline search yielded 37 results ( REF _Ref461280512 \h \* MERGEFORMAT Figure A2), of which just three (PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5XaWxkPC9BdXRob3I+PFllYXI+MjAwODwvWWVhcj48UmVj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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5XaWxkPC9BdXRob3I+PFllYXI+MjAwODwvWWVhcj48UmVj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ADDIN EN.CITE.DATA (Wild et al., 2008), PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Ccm9lZGVyPC9BdXRob3I+PFllYXI+MjAwMjwvWWVhcj48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 EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5Ccm9lZGVyPC9BdXRob3I+PFllYXI+MjAwMjwvWWVhcj48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 EN.CITE.DATA (Broeder et al., 2002), PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MaW48L0F1dGhvcj48WWVhcj4xOTk2PC9ZZWFyPjxSZWNO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ADDIN EN.CITE PEVuZE5vdGU+PENpdGU+PEF1dGhvcj5MaW48L0F1dGhvcj48WWVhcj4xOTk2PC9ZZWFyPjxSZWNO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ADDIN EN.CITE.DATA (Lin et al., 1996)) appeared in the Embase results.After filtering out non-RA, non-adult studies, those which did not investigate clinical biomarkers and those not utilising multivariate methods, 25 articles remained. From these 25 articles, 23 articles translated multivariate methods as linear or logistic regression but without any examination of interactions. The other 2 were Wild et al and Lin et al who used discriminant analysis and GEE respectively.A.1.4.Systematic Review of Data Mining MethodologiesAdditional systematic reviews were conducted on specific data mining terms to identify their use with RA data. The data mining terms searched on were: CART, Cross Validation, Nearest Neighbour, Lasso, Bagging, Kernel, Principal Components, Ridge Regression and Support Vector Machines. Details of the systematic review can be found in Appendix A.ResultsA summary of results is shown in REF _Ref490055713 \h \* MERGEFORMAT Table A0.4. Even though over 40 articles were identified that applied data mining methodology to an RA setting, just one applied these methods to predict RA outcomes from biomarkers.Table A STYLEREF 1 \s 0. SEQ Table \* ARABIC \s 1 4 Data Mining Systematic ReviewData Mining/ Statistical TermArticlesPrediction of RA outcomes in Adults using clinical biomarkersCART40Bagging00Boosting00Random Forest20Nearest Neighbour30Kernel20Support Vector Machines10Ridge Regression00Lasso00Principal Components140Cross Validation61A.1.5.Systematic Review of logistic regression and CART in the context of variable selection Search criteria: CART AND Logistic Regression AND "variable selection"Databases:BIOSIS Previews?Current Contents? SearchDerwent Drug FileEmbase?Embase? AlertGale Group PROMT?International Pharmaceutical AbstractsMEDLINE?SciSearch?: A Cited Reference Science DatabaseResults60007528575Figure A SEQ Figure \* ARABIC \s 1 3 Results of logistic regression and CART variable selection systematic reviewFigure A3 Results of logistic regression and CART variable selection systematic review600075485775Results databases (n=64)Not Variable Selection(n= 59)Comparison using ROC (n=2)Variable selection discussed (n=5)Comparison of variable selection (n= 3)Comparison of variable selection in RA (n= 0)Results databases (n=64)Not Variable Selection(n= 59)Comparison using ROC (n=2)Variable selection discussed (n=5)Comparison of variable selection (n= 3)Comparison of variable selection in RA (n= 0)64 Publications reviewed59 did not compare CART vs logistic regression in variable selection. Some did not compare methods at all, some used CART for variable selection and then used logistic regression for model building. Others did compare CART and logistic regression but in terms of prediction in propensity scoring. 2 indirectly compared CART vs logistic regression in variable bining non-parametric models with logistic regression: an application to motor vehicle injury data. ADDIN EN.CITE <EndNote><Cite><Author>Kuhnert</Author><Year>2000</Year><RecNum>172</RecNum><DisplayText>(Kuhnert et al., 2000)</DisplayText><record><rec-number>172</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">172</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kuhnert, P.M. </author><author>Do, K.A., </author><author>McClure, R.</author></authors></contributors><titles><title>Combining non-parametric models with logistic regression : an application to motor vehicle injury data</title><secondary-title>Computational Statistics & Data Analysis </secondary-title></titles><pages>371-386</pages><volume>34</volume><dates><year>2000</year></dates><urls></urls></record></Cite></EndNote>(Kuhnert et al., 2000)Which Is a More Accurate Predictor in Colorectal Survival Analysis? Nine Data Mining Algorithms vs. the TNM Staging System. ADDIN EN.CITE <EndNote><Cite><Author>Gao</Author><Year>2012</Year><RecNum>173</RecNum><DisplayText>(Gao et al., 2012)</DisplayText><record><rec-number>173</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473716834">173</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Gao, P.</author><author>Zhou, X.</author><author>Wang, Z-n.</author><author>Song, Y-x.</author><author>Tong, L-l.</author><author>Xu, Y-y. </author><author>Yue, Z-y.</author><author>Xu, H-m</author></authors></contributors><titles><title>Which Is a More Accurate Predictor in Colorectal Survival Analysis? Nine Data Mining Algorithms vs. the TNM Staging System</title><secondary-title>PLoS ONE</secondary-title></titles><periodical><full-title>PloS one</full-title></periodical><pages>e42015</pages><volume>7</volume><number>7</number><dates><year>2012</year></dates><urls></urls><electronic-resource-num>doi:10.1371/journal.pone</electronic-resource-num></record></Cite></EndNote>(Gao et al., 2012).3 directly compared CART vs logistic regression in variable parison of some chemometric tools for metabonomics biomarker identification. ADDIN EN.CITE <EndNote><Cite><Author>Rousseau</Author><Year>2008</Year><RecNum>174</RecNum><DisplayText>(Rousseau et al., 2008)</DisplayText><record><rec-number>174</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473717023">174</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rousseau, R., </author><author>Govaerts, B. </author><author>Verleysen, M. </author><author>Boulanger, B. </author></authors></contributors><titles><title>Comparison of some chemometric tools for metabonomics biomarker identification</title><secondary-title>Chemometrics and Intelligent Laboratory Systems </secondary-title></titles><periodical><full-title>Chemometrics and Intelligent Laboratory Systems</full-title></periodical><pages>54-66</pages><volume>91</volume><dates><year>2008</year></dates><urls></urls></record></Cite></EndNote>(Rousseau et al., 2008).Using classification trees to assess low birth weight outcomes. ADDIN EN.CITE <EndNote><Cite><Author>Kitsantas</Author><Year>2006</Year><RecNum>288</RecNum><DisplayText>(Kitsantas et al., 2006)</DisplayText><record><rec-number>288</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1519904301">288</key><key app="ENWeb" db-id="">0</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kitsantas, P.</author><author>Hollander, M.</author><author>Li, L.</author></authors></contributors><auth-address>George Mason University, Department of Health Administration and Policy, The College of Health and Human Services, 4400 University Drive, Fairfax, VA 22030, USA. kitsantap@ecu.edu</auth-address><titles><title>Using classification trees to assess low birth weight outcomes</title><secondary-title>Artif Intell Med</secondary-title></titles><periodical><full-title>Artif Intell Med</full-title></periodical><pages>275-89</pages><volume>38</volume><number>3</number><edition>2006/05/30</edition><keywords><keyword>Adolescent</keyword><keyword>Adult</keyword><keyword>Female</keyword><keyword>Florida</keyword><keyword>Humans</keyword><keyword>*Infant, Low Birth Weight</keyword><keyword>Infant, Newborn</keyword><keyword>Mothers</keyword><keyword>Pregnancy</keyword><keyword>*Pregnancy Outcome</keyword><keyword>Risk Factors</keyword></keywords><dates><year>2006</year><pub-dates><date>Nov</date></pub-dates></dates><isbn>0933-3657 (Print)
0933-3657 (Linking)</isbn><accession-num>16730961</accession-num><urls><related-urls><url>;(Kitsantas et al., 2006).Repeated split sample validation to assess logistic regression and recursive partitioning: An application to the prediction of cognitive impairment. ADDIN EN.CITE <EndNote><Cite><Author>James</Author><Year>2005</Year><RecNum>176</RecNum><DisplayText>(James et al., 2005)</DisplayText><record><rec-number>176</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473717423">176</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>James, K.E. </author><author>White, R.F. </author><author>Kraemer, H.C.</author></authors></contributors><titles><title>Repeated split sample validation to assess logistic regression and recursive partitioning: An application to the prediction of cognitive impairment</title><secondary-title>Stat Med</secondary-title></titles><periodical><full-title>Stat Med</full-title></periodical><pages>3019-35</pages><volume>24</volume><number>19</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(James et al., 2005)DiscussionKuhnert et al ADDIN EN.CITE <EndNote><Cite><Author>Kuhnert</Author><Year>2000</Year><RecNum>172</RecNum><DisplayText>(Kuhnert et al., 2000)</DisplayText><record><rec-number>172</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">172</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kuhnert, P.M. </author><author>Do, K.A., </author><author>McClure, R.</author></authors></contributors><titles><title>Combining non-parametric models with logistic regression : an application to motor vehicle injury data</title><secondary-title>Computational Statistics & Data Analysis </secondary-title></titles><pages>371-386</pages><volume>34</volume><dates><year>2000</year></dates><urls></urls></record></Cite></EndNote>(Kuhnert et al., 2000), compare multivariate adaptive regression splines (MARS) modelling with logistic regression and CART in an epidemiological setting of injuries resulting from motor vehicle accidents. The data consist of 55 cases and 663 controls and a small number of explanatory variables (less than 20). The paper compares the variables selected by each method. The authors found that strengths of each method counteract the weaknesses of another but conclude that MARS outperforms Logistic Regression and CART on this particular dataset based on sensitivity and specificity. The authors recognise that this performance may not be generalisable outside of this data. Gao et al ADDIN EN.CITE <EndNote><Cite><Author>Gao</Author><Year>2012</Year><RecNum>173</RecNum><DisplayText>(Gao et al., 2012)</DisplayText><record><rec-number>173</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473716834">173</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Gao, P.</author><author>Zhou, X.</author><author>Wang, Z-n.</author><author>Song, Y-x.</author><author>Tong, L-l.</author><author>Xu, Y-y. </author><author>Yue, Z-y.</author><author>Xu, H-m</author></authors></contributors><titles><title>Which Is a More Accurate Predictor in Colorectal Survival Analysis? Nine Data Mining Algorithms vs. the TNM Staging System</title><secondary-title>PLoS ONE</secondary-title></titles><periodical><full-title>PloS one</full-title></periodical><pages>e42015</pages><volume>7</volume><number>7</number><dates><year>2012</year></dates><urls></urls><electronic-resource-num>doi:10.1371/journal.pone</electronic-resource-num></record></Cite></EndNote>(Gao et al., 2012), compared a selection of data mining methods and their performance in predicting 5 year survival rate of colorectal cancer. The methods included CART and logistic regression. They used 2 datasets: a large American dataset of 10,000 cases and twenty variables, and a smaller Chinese origin dataset of 760 cases and 20 variables. Each method was compared using AUC of the ROC curve, but little comparison of variable selection. Here the authors claim logistic regression outperformed CART, but AUCs were very close at 81% and 80% respectively with mostly overlapping confidence intervals.Rousseau et al ADDIN EN.CITE <EndNote><Cite><Author>Rousseau</Author><Year>2008</Year><RecNum>174</RecNum><DisplayText>(Rousseau et al., 2008)</DisplayText><record><rec-number>174</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473717023">174</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Rousseau, R., </author><author>Govaerts, B. </author><author>Verleysen, M. </author><author>Boulanger, B. </author></authors></contributors><titles><title>Comparison of some chemometric tools for metabonomics biomarker identification</title><secondary-title>Chemometrics and Intelligent Laboratory Systems </secondary-title></titles><periodical><full-title>Chemometrics and Intelligent Laboratory Systems</full-title></periodical><pages>54-66</pages><volume>91</volume><dates><year>2008</year></dates><urls></urls></record></Cite></EndNote>(Rousseau et al., 2008), explore the use of 6 multivariate methods for biomarker identification. The methods include logistic regression and CART. The authors constructed a “semi-artificial” dataset of 800 spectra representing rat urine, and 46 potential biomarkers. Each multivariate method was employed to identify the 6 independent biomarkers. In their analysis Logistic regression identified 5 of 6 biomarkers and CART just 3. The authors concluded that ‘basic CART’ can be efficient in situations with low noise but not able to find signals where there is high noise. Further the authors concluded that once a biomarker is found with logistic regression, others dependent on the first tend to be ignored. Similarly, they claim CART only identifies independent biomarkers.James et al ADDIN EN.CITE <EndNote><Cite><Author>James</Author><Year>2005</Year><RecNum>176</RecNum><DisplayText>(James et al., 2005)</DisplayText><record><rec-number>176</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1473717423">176</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>James, K.E. </author><author>White, R.F. </author><author>Kraemer, H.C.</author></authors></contributors><titles><title>Repeated split sample validation to assess logistic regression and recursive partitioning: An application to the prediction of cognitive impairment</title><secondary-title>Stat Med</secondary-title></titles><periodical><full-title>Stat Med</full-title></periodical><pages>3019-35</pages><volume>24</volume><number>19</number><dates><year>2005</year></dates><urls></urls></record></Cite></EndNote>(James et al., 2005), aim to select predictors for cognitive impairment. The authors aimed to predict illness in 252 patients from 5 to 7 tests from an available 22 tests. The ROC curves for logistic regression and CART were almost identical. CART had slightly better sensitivity whilst logistic regression produced somewhat better specificity. The authors found methods were comparable in performance, but found CART easier for clinician to use. Kitsantas et al ADDIN EN.CITE <EndNote><Cite><Author>Kitsantas</Author><Year>2006</Year><RecNum>288</RecNum><DisplayText>(Kitsantas et al., 2006)</DisplayText><record><rec-number>288</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="1519904301">288</key><key app="ENWeb" db-id="">0</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Kitsantas, P.</author><author>Hollander, M.</author><author>Li, L.</author></authors></contributors><auth-address>George Mason University, Department of Health Administration and Policy, The College of Health and Human Services, 4400 University Drive, Fairfax, VA 22030, USA. kitsantap@ecu.edu</auth-address><titles><title>Using classification trees to assess low birth weight outcomes</title><secondary-title>Artif Intell Med</secondary-title></titles><periodical><full-title>Artif Intell Med</full-title></periodical><pages>275-89</pages><volume>38</volume><number>3</number><edition>2006/05/30</edition><keywords><keyword>Adolescent</keyword><keyword>Adult</keyword><keyword>Female</keyword><keyword>Florida</keyword><keyword>Humans</keyword><keyword>*Infant, Low Birth Weight</keyword><keyword>Infant, Newborn</keyword><keyword>Mothers</keyword><keyword>Pregnancy</keyword><keyword>*Pregnancy Outcome</keyword><keyword>Risk Factors</keyword></keywords><dates><year>2006</year><pub-dates><date>Nov</date></pub-dates></dates><isbn>0933-3657 (Print)
0933-3657 (Linking)</isbn><accession-num>16730961</accession-num><urls><related-urls><url>;(Kitsantas et al., 2006), research low birth weights in Florida based on over 180,000 births. The authors compared CART and logistic regression using ROC curves and sensitivity and specificity. Ten explanatory variables were considered in the models. For their data, Kitsantas et al, found that logistic regression and CART revealed similar risk factors with some slight variations across regions. Sensitivity and specificity were found to be comparable.Appendix B: Clinical Trial DatasetB.1.Detail of variables The CONTENTS Procedure Data Set Name PHDLIB.ALLALL Observations 11705 Member Type DATA Variables 178 Engine V9 Indexes 0 Created 03/13/2017 20:25:43 Observation Length 1487 Last Modified 03/13/2017 20:25:43 Deleted Observations 0 Engine/Host Dependent Information Data Set Page Size 65536 Number of Data Set Pages 153 Number of Data Set Repairs 0 Filename /opt/BIOSTAT/home/mahonep1/phd/libraries/allall.sas7bdat Release Created 9.0401M3 Host Created Linux Inode Number 1771330 Access Permission rw-rw-r-- Owner Name mahonep1 File Size 10MB File Size (bytes) 10092544 Alphabetic List of Variables and Attributes (1) # Variable Type Label26 ACR20 Num ACR2027 ACR50 Num ACR5028 ACR70 Num ACR7056 ACR90 Num ACR9029 ACRN Num ACRN3 AGE Num Age (years) at screening65 APH Num Alkaline Phosphatase U/L66 BASO Num Basophils 10**9/L30 BCRP Num Baseline CRP55 BDAS Num Baseline DAS31 BESR Num Baseline ESR32 BHAQ Num Baseline HAQ67 BILT Num Total Bilirubin umol/L154 BLOOD Num Hx Blood and Lymphatic Disorders133 BP_Z Num SF36 Bodily Pain Std Score33 BSJC Num Baseline Swollen Joint Count22 BSTEROID Char Steroid use at Baseline34 BTJC Num Baseline Tender Joint Count35 BVASPAIN Num Baseline Pain VAS36 BVASPHGS Num Baseline Physician Global Score VAS37 BVASPTGS Num Baseline Patient Global Score VAS68 CA Num Calcium mmol/L155 CARDIAC Num Hx Cardiac Disorders69 CHL Num Chloride mmol/L70 CHOL Num Cholesterol mmol/L170 CNS Num Hx Nervous System and Neurological14 COMPOUND Char Drug Compound58 CRP20 Num CRP Response (20%)46 CRP24 Num CRP Week 242 CRTN Num Clinical Research Task #25 DASCW24 Num DAS CRP Week 2424 DASEW24 Num DAS ESR Week 2454 DASW24 Num DAS Week 24101 DBP Num Diastolic Blood Pressure23 DMARDNUM Num Number of previous DMARDS18 DRUGTYPE Char Treatment type8 DSETTYPE Char Data Domain Alphabetic List of Variables and Attributes (2) 158 EAR Num Hx Ear and Labyrinth Disorders162 ENDOCRIN Num Hx Endocrine Disorders71 EOS Num Eosinophils 10**9/L151 ERNSCORE Num Erosion Score107 EROSION Num Erosion Score59 ESR20 Num ESR Response (20%)47 ESR24 Num ESR Week 24159 EYE Num Hx Eye Disorders163 GASTRO Num Hx Gastrointestinal Disorders164 GENERAL Num Hx General Disorders156 GENETIC Num Hx Congenital and Genetic Disorders134 GH_Z Num SF36 General Health Std Score72 GLU Num Fasting Glucose mmol/L57 HAQ20 Num HAQ Response (20%)48 HAQ24 Num HAQ Week 2473 HCT Num Haematocrit fraction11 HEIGHT Num Height165 HEPBIL Num Hx Hepato-Biliary Disorders74 HGB Num Haemoglobin g/L160 IMMUNE Num Hx Immune System Disorders166 INFECT Num Hx Infestations and Infestations167 INJURY Num Hx Injury and Poisoning168 INVEST Num Hx Investigations117 JSNFEET Num Joint Space Narrowing - Feet118 JSNHANDS Num Joint Space Narrowing - Hands119 JSNSCORE Num Joint Space Narrowing Total Score75 LDH Num Lactic dehydrogenase U/L76 LYM Num Lymphocytes 10**9/L121 MCS Num SF36 Mental Component Score157 METABOL Num Hx Metabolism and Nutrition Disorders135 MH_Z Num SF36 Mental Health Std Score77 MONO Num Monocytes 10**9/L169 MUSCULO Num Hx Musculoskeletal and Connective Tissue Disorders153 NEOPLASM Num Hx Neoplasms78 NTT Num Neutrophils 10**9/L38 PCRP Num Percent change in CRP126 PCS Num SF36 Physical Component Score39 PESR Num Percent change in ESR136 PF_Z Num SF36 Physical Functioning Std Score40 PHAQ Num Percent change in HAQ79 PHOS Num Phosphate mmol/L80 PLT Num Platelets 10**9/L16 POPIR Char Inadequate Response Population15 POPRA Char RA Population81 POT Num Potassium mmol/L171 PREG Num Hx Pregnancy Conditions82 PROT Num Total Protein g/L1 PROTO Char Protocol (in upper case)41 PSJC Num Percent change in SJC172 PSYCH Num Hx Psychiatric Disorders12 PT Num Patient #42 PTJC Num Percent change in TJC115 PULSE Num Heart Rate43 PVASPAIN Num Percent change in Pain44 PVASPHGS Num Percent change in Phys Global Score45 PVASPTGS Num Percent change in Patient Global Score5 RACE Char Race7 RADUR Num Duration of RA (years)83 RBC Num RBC 10**12/L6 REGION Char GEOGRAPHIC REGION173 RENAL Num Hx Renal and Urinary Disorders161 REPRO Num Hx Reproductive and Breast Disorders174 RESP Num Hx Respiratory Thoracic Mediastinal Disorders137 RE_Z Num SF36 Role Limitation - Emotional Std Score84 RF Num RF U/mL20 RFCAT Char RF positive negative21 RFNUM Num RF value138 RH_Z Num SF36 RH Std Score139 RP_Z Num SF36 Role Limitation - Physical Std Score85 SALB Num Albumin g/L132 SBP Num Systolic Blood Pressure86 SCRT Num Creatinine umol/L4 SEX Char Sex113 SF0101 Num SF36 General Health Alphabetic List of Variables and Attributes (3) 111 SF0201 Num SF36 Health Compared to 1 year ago145 SF0301 Num SF36 Vigorous Activity122 SF0302 Num SF36 Moderate Activity120 SF0303 Num SF36 Lifting100 SF0304 Num SF36 Climbing several flights stairs99 SF0305 Num SF36 Climbing one flight stairs96 SF0306 Num SF36 Bending146 SF0307 Num SF36 Walking more than a mile148 SF0308 Num SF36 Walking several blocks147 SF0309 Num SF36 Walking one block95 SF0310 Num SF36 Bathing and Dressing127 SF0401 Num SF36 Physical: Cut time on work or activities125 SF0402 Num SF36 Physical: Accomplished Less130 SF0403 Num SF36 Physical: Limited in work or activities128 SF0404 Num SF36 Physical: Difficulty in work or activities105 SF0501 Num SF36 Emotional: Cut time on work or activities104 SF0502 Num SF36 Emotional: Accomplished Less103 SF0503 Num SF36 Emotional: not as careful131 SF0601 Num SF36 Interference with normal social activities98 SF0701 Num SF36 Bodily Pain124 SF0801 Num SF36 Pain interfere with work108 SF0901 Num SF36 Feel full of pep123 SF0902 Num SF36 Been nervous102 SF0903 Num SF36 Down in the dumps129 SF0904 Num SF36 Felt calm and peaceful106 SF0905 Num SF36 Lot of energy97 SF0906 Num SF36 Felt downhearted and blue150 SF0907 Num SF36 Felt wornout109 SF0908 Num SF36 Been happy144 SF0909 Num SF36 Felt tired116 SF1001 Num SF36 Amount of time interfered with social143 SF1101 Num SF36 Get sick easier110 SF1102 Num SF36 As healthy as anybody114 SF1103 Num SF36 Expect health to get worse112 SF1104 Num SF36 Health is excellent140 SF_Z Num SF36 Social Functioning Std Score87 SGOT Num ASAT (SGOT) U/L88 SGPT Num ALAT (SGPT) U/L142 SHARP Num Sharp Genant Score152 SHPSCORE Num Sharp Score60 SJC20 Num SJC Response (20%)49 SJC24 Num Swollen Joints Week 24175 SKIN Num Hx Skin and Subcutaneous Disorders9 SMOKHIS Char Smoking history10 SMOKYR Num Years since stopped smoking (derived)176 SOCIAL Num Hx Social Circumstances89 SOD Num Sodium mmol/L177 SURGERY Num Hx Surgical and Medical Procedures61 TJC20 Num TJC Response (20%)50 TJC24 Num Tender Joints Week 2413 TRTMENT Char Treatment assigned90 UAC Num Uric Acid umol/L91 UBLD_QLT Num Haematuria 0 to 4+92 UGLU_QLT Num Glycosuria 0 to 4+93 UREA Num BUN mmol/L178 VASCULAR Num Hx Vascular Disorders62 VPAIN20 Num VAS Pain Response (20%)51 VPAIN24 Num Pain Score Week 2463 VPHGS20 Num VAS Phys Global Score Response (20%)52 VPHGS24 Num Physician Global Score Week 2464 VPTGS20 Num VAS Patient Global Score Response (20%)53 VPTGS24 Num Patient Global Score Week 24141 VT_Z Num SF36 Vitality Std Score94 WBC Num White blood cell (WBC) 10**9/L19 WEIGHT Num Weight kg149 WITHERN Num Number of joints with erosion17 YOE Num Year of first enrolment B.2.Summary statistics Figure B SEQ Figure \* ARABIC \s 1 1 Histogram, Boxplot and Normal Q-Q plot for continuous variablesFigure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B1 Histogram, Boxplot and Normal Q-Q plot for continuous variables (continued)Figure B SEQ Figure \* ARABIC \s 1 2 Categorical variables Figure B2 Categorical variables (continued) Figure B SEQ Figure \* ARABIC \s 1 3 Medical History variables Figure B3 Medical History variables (continued)Figure B3 Medical History variables (continued)Figure B3 Medical History variables (continued)Figure B3 Medical History variables (continued)Figure B3 Medical History variables (continued)Figure B3 Medical History variables (continued)Appendix C: SimulationsC.1.Simulation ProtocolCharacterising Outcomes in Rheumatoid ArthritisPaul Mahoney ScHARR, University of Sheffield11/1/2011Overall objectives The objective of this simulation is to identify how CART and logistic regression perform in predicting patient ACR20, ACR50 and ACR90 response in an RA setting under differing scenarios, and to select a shortlist of methods to later apply to available real life RA clinical trial database. Simulation proceduresA series of simulated datasets will be created to probe how CART and logistic regression perform under changes in data structure. Simulated datasets will be generated independently in SAS (Version 9.4), with different seeds for each simulation. CART analysis will be performed in R (package rpart ADDIN EN.CITE <EndNote><Cite><Author>Therneau</Author><Year>2017</Year><RecNum>183</RecNum><DisplayText>(Therneau et al., 2017)</DisplayText><record><rec-number>183</rec-number><foreign-keys><key app="EN" db-id="09pv5azpjevvdhe5as1vvszywffsrfwte5e5" timestamp="1502290933">183</key></foreign-keys><ref-type name="Web Page">12</ref-type><contributors><authors><author>Therneau, T. </author><author>Atkinson, B.</author><author>Ripley, B.</author></authors></contributors><titles><title>rpart: Recursive Partitioning and Regression Trees. R package version 4.1-11</title></titles><number>Last Accessed 30th June 2017</number><dates><year>2017</year></dates><urls><related-urls><url>;(Therneau et al., 2017)), whilst logistic regression will be performed in SAS (Version 9.4).Data generation methods The clinical trial database represents over 11,000 adult patients diagnosed with RA from 18 randomised clinical trials in 4 development drug projects enrolled between 1998 and 2008. Simulations will be generated that mimic this real life database. Two types of variable will be generated in the simulated datasets: continuous and binary. In this simulation, continuous variables will be assumed to be normally distributed or assumed that they can be approximated by a Normal distribution due to large sample size. Simulated datasets will be generated in stages. In the first stage the simulation will generate a binary response followed by 10 random Normal variables highly correlated with this response, and 90 random Normal variables with low correlation with this response. Since around 100 variables are available in the clinical database, this is sufficient and manageable in the context of multiple simulations. Simulations of size 100, 1,000 and 10,000 will be generated. For both continuous and binary data, a seed will be specified to enable reproducibility.In the second stage, the impact of dichotomisation of explanatory variables will be evaluated, with one half of the correlated and uncorrelated variables transformed to binary variables. The third stage will explore how missing values affect the performance of CART and logistic regression. Two types of missing value will be generated: missing completely at random and missing not at random. For missing not at random, missing values will be assigned depending on the value of the response variable. That is, a higher probability of being missing will be set higher for non-responders. If homogeneity of outcomes is observed in the first stage, a reduced set of simulations may be applied to subsequent stages. Statistical methods to be investigatedFollowing generation of the datasets, logistic regression models will be fitted, to ascertain which correlated variables are selected as significant predictors of outcome. Stepwise logistic regression will be performed in SAS with the significance level for entry PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+V2lsZDwvQXV0aG9yPjxZZWFy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ADDIN EN.CITE PEVuZE5vdGU+PENpdGUgRXhjbHVkZVllYXI9IjEiPjxBdXRob3I+V2lsZDwvQXV0aG9yPjxZZWFy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ADDIN EN.CITE.DATA (Wild et al.) and significance level for staying (SLS) in the model set to the same value. The CART analysis will be performed using rpart package of the R software application to see if high correlated variables are selected in the final tree. Parameters applied include: equal weights, anova modelling, retention of observations if one or more predictors are missing, equal costs, and a minimum of 20 observations per node in order for a split to be attempted. Scenarios to be investigatedThe initial scenarios will assume no missing values in the data. The following overall adaptations will be implemented:Sample Size: 100, 1,000, 10,000High Correlation: 0.90, 0.80, 0.70, 0.60 Low Correlation: 0, 0.10, 0.20, 0.30Response 0.25, 0.50, 0.90 The Logistic Regression analysis will follow the stepwise procedure with parameters SLENTRY and SLSTAY determining the significance levels for entry and removal of effects from the model. SLENTRY and SLSTAY will be adapted for values: 0.30, 0.20, 0.10, 0.05, 0.01, 0.001For simulations evaluating the impact of dichotomisation and missing values, a subset of scenarios may be applied, depending on any homogeneity of results observed from the initial stage. Number of Simulations to be performedEach combination of the parameters of the simulation will be performed 100 times, leading to 14,400 analyses on CART and 86,400 analyses on logistic regression. The number of simulations is subject to system processing limitations, particularly for the 10,000 subject simulations. Performance criteriaFor each analysis, the rate of highly correlated variables selected by each method (sensitivity) and the rate of low correlated variables not selected by each method (specificity) will be calculated. These paired values will be presented in Sensitivity and Specificity (VaSe) plots. VaSe plots will be developed based on ROC curves, where points are plotted on scale where the x-axis displays 1-specificity and the y-axis displays sensitivity. When the method selects only the 10 higher correlated variables and none of the lower correlated variables the sensitivity and specificity will be 100%, visually this will be in the top left corner of the VaSe plots.Presentation of resultsSensitivity and inverse specificity will be plotted on VaSe plots. CART and logistic regression results will be plotted together for each simulation scenario, so that performance can easily be compared. VaSe plots curves will also be produced for logistic regression only displaying together the varying SLENTRY and SLSTAY values so that the optimal value can be identified. VaSe plots comparing dichotomisation and non-dichotomisation will be produced together as will VaSe plots for non-missing, missing at random and missing not at random.SummaryThis protocol outlines the planned simulation strategy for evaluating the performance of CART and logistic regression for simulated RA data. The results of the simulations are intended to enable selection of an optimal method to analyse the real life clinical database and to identify the highest performing analysis parameters. C.2.Simulation codeThe code for simulations was generated as depicted in figure C2.1 and described in Section 7.4.2.Figure C SEQ Figure \* ARABIC \s 1 1 Generation of Simulated DatasetsC.2.1.R Code simrpart.RSummary: Reads in simulated dataset from SAS (sim.csv). Generates CART using package rpartIdentifies nodes of the created treeOutputs a text file of nodes to be read by SAS program (rpart.txt)# load packageslibrary("rpart", lib.loc="/usr/lib64/R/library")# simulation with n=100sima <- read.csv("./sim.csv", header=T)# cart on sima100fit<- rpart(RESP~ ., data=sima)summary(fit)# what are components of fitnames(fit)# frame seems to contain nodes#nodes<-fit$frame[fit$frame.var!=" <leaf>",1]nodes<-fit$variable.importance# output this to a file that SAS can pick updput(nodes, file="rpart.txt")C.2.2.SAS Code simmacro.sas/******************************************************************************** Create a series of simulated datasets simulation a: constant variance shift in means 80% correlation for 10 x's, 0 correlation for 90 x's response 50/50 repeat for n=100 (sima100) n=1000 (sima1000) n=9000 (sima9000)*********************************************************************************/%macro sim(dset,s,p,corr1,corr2,seed);data &dset; do study=1 to &s; between = 5*rannor(&seed); %do high=1 %to 9; if &corr2 gt 0 then total0&high = 5+(between+sqrt(((1-&corr2)/&corr2))*5*rannor(&seed+&high)); %end; if &corr2 gt 0 then total10 =10+(between+sqrt(((1-&corr2)/&corr2))*5*rannor(&seed+10)); %do low=11 %to 100; if &corr1 = 0 then total&low = 5+10*rannor(&seed+&low); else total&low = 5+(between+sqrt(((1-&corr1)/&corr1))*10*rannor(&seed+&low)); %end; **** between determines split of response and non response ****; * *** cut off 0 gives 50/50 ***; *** use N(0,1) table to find other splits ***; if between >=0 then resp=1; else resp=0; output; end;run;options nosource nonotes;*turn off export code;proc export data=&dset outfile="sim.csv" dbms=csv replace;run;options source notes;title "Correlation= &corr1 &corr2";%mend sim;C.2.3.SAS Code simmacro.sas/****************************************************************************Creates a series of simulated datasets simulationFor a specific number of simulations. each simulated dataset is used by R rpart procedure - variables used in tree are reimported into SAS dataset created to summarise each rpart. use this to generate ROC info*****************************************************************************/options symbolgen mprint mlogic;%include 'environ.sas';** Get macro ;%inc 'simmacro0.sas';* parameters of simmacro ;* dset = dataset to be created ;* s = number of patients ; * also s used to name output dataset;* p = ;* corr1 = correlation for top 10 vars ;* corr2 = correlation for 90 vars ;* seed = seed for rannor ;%macro cartroc(howmany,dset,s,corr1,corr2,sle,sls); * create empty dataset for appending later cartrocs; data cartroc; length truepos 8 falspos 8 trueneg 8 falsneg 8 iter 8; label truepos ='No. of highly correlated variables identified by CART as predictive' falspos ='No. of lowly correlated variables identified by CART as predictive' trueneg ='No. of lowly correlated variables not identified by CART as predictive' falsneg ='No. of highly correlated variables not identified by CART as predictive' iter ='Seed'; stop; run; * create empty dataset for appending later lrrocs; data lrroc; length truepos 8 falspos 8 trueneg 8 falsneg 8 iter 8; label truepos ='No. of highly correlated variables identified by LR as predictive' falspos ='No. of lowly correlated variables identified by LR as predictive' trueneg ='No. of lowly correlated variables not identified by LR as predictive' falsneg ='No. of highly correlated variables not identified by LR as predictive' iter ='Seed'; stop; run; %do i=1 %to &howmany; ** create simulation dataset; %sim(&dset, &s, 3, &corr1, &corr2, &i); ** run the CART on each dataset *******************************************************************; ** Execute R that runs CART analysis and returns nodes of tree ; x 'rr simrpart.R'; data cartpick; infile "./rpart.txt" firstobs=1 delimiter="," dsd scanover; input @'TOTAL' value1 $3. @@; valuen=input(compress(value1,'("'),6.0); if valuen le 10 then truepos=1; else truepos =0; * cart rightly predict hi correlates; if valuen gt 10 then falspos=1; else falspos =0; * cart wrongly predict lo correlates; run; proc summary data=cartpick; var truepos falspos; output out=cartsum sum=truepsum falspsum; run; data cartroc&i(rename=(truepsum=truepos falspsum=falspos) drop=_FREQ_ _TYPE_); set cartsum; falsneg=10-truepsum; *cart wrongly didnt pick out these hi correlates; trueneg=90-falspsum; *cart rightly didnt pick out these lo correlates; iter=&i; run; proc datasets nolist; append base=cartroc data=cartroc&i force; run; proc datasets; delete cartroc&i; run; ** run the logistic regression on each dataset *******************************************************************; * for same dataset as rpart run a logistic regression ; title "Stepwise Regression &dset"; proc logistic data=&dset outest=betas covout; model resp(event='1')=total01--total100 / selection=stepwise slentry=&sle slstay=&sls details lackfit; output out=pred p=phat lower=lcl upper=ucl predprob=(individual crossvalidate); run; proc print data=betas; title2 'Parameter Estimates and Covariance Matrix'; run cancel; proc print data=pred; title2 'Predicted Probabilities and 95% Confidence Limits'; run cancel; data lrpick(keep = _name_ intercept valuen truepos falspos); set betas; where intercept ne . and _name_ not in ("RESP","Intercept"); valuen=input(compress(_name_,'TOTAL'),6.0); if valuen le 10 then truepos=1; else truepos =0; * cart rightly predict hi correlates; if valuen gt 10 then falspos=1; else falspos =0; * cart wrongly predict lo correlates; run; proc summary data=lrpick; var truepos falspos; output out=lrsum sum=truepsum falspsum; run; data lrroc&i(rename=(truepsum=truepos falspsum=falspos) drop=_FREQ_ _TYPE_); set lrsum; falsneg=10-truepsum; *cart wrongly didnt pick out these hi correlates; trueneg=90-falspsum; *cart rightly didnt pick out these lo correlates; iter=&i; run; proc datasets nolist; append base=lrroc data=lrroc&i force; run; proc datasets; delete lrroc&i; run; %end; * end i to howmany loop ; data phdlib.c&dset; set cartroc; label sens ='Sensitivity' specinv ='1 - Specificity'; utt=&s;type='CART';parm="&corr2 corr"; sens=truepos/(truepos+falsneg); specinv=1-(trueneg/(trueneg+falspos)); run; data phdlib.l&dset; set lrroc; label sens ='Sensitivity' specinv ='1 - Specificity'; utt=&s;type='LR';parm="&corr2 corr"; sens=truepos/(truepos+falsneg); specinv=1-(trueneg/(trueneg+falspos)); run;%mend cartroc;************************************************** zero correlation on low correlated variables ***;** simulations on 100 patients *****%cartroc(100,sa100,100,0,0.90,0.30,0.30);%cartroc(100,sb100,100,0,0.80,0.30,0.30);%cartroc(100,sc100,100,0,0.70,0.30,0.30);%cartroc(100,sd100,100,0,0.60,0.30,0.30);** simulations on 1000 patients *****;%cartroc(100,sa1000,1000,0,0.90,0.30,0.30);%cartroc(100,sb1000,1000,0,0.80,0.30,0.30);%cartroc(100,sc1000,1000,0,0.70,0.30,0.30);%cartroc(100,sd1000,1000,0,0.60,0.30,0.30)**** simulations on 10000 patients *****/%cartroc(100,sa10000,10000,0,0.90,0.30,0.30);%cartroc(100,sb10000,10000,0,0.80,0.30,0.30%cartroc(100,sc10000,10000,0,0.70,0.30,0.30);%cartroc(100,sd10000,10000,0,0.60,0.30,0.30);** 0.10 correlation on low correlated variables ***;** simulations on 100 patients *****;%cartroc(100,se100,100,0.1,0.90,0.30,0.30);%cartroc(100,sf100,100,0.1,0.80,0.30,0.30);%cartroc(100,sg100,100,0.1,0.70,0.30,0.30);%cartroc(100,sh100,100,0.1,0.60,0.30,0.30);** simulations on 1000 patients *****;%cartroc(100,se1000,1000,0.1,0.90,0.30,0.30);%cartroc(100,sf1000,1000,0.1,0.80,0.30,0.30);%cartroc(100,sg1000,1000,0.1,0.70,0.30,0.30);%cartroc(100,sh1000,1000,0.1,0.60,0.30,0.30);** simulations on 10000 patients *****;%cartroc(100,se10000,10000,0.1,0.90,0.30,0.30);%cartroc(100,sf10000,10000,0.1,0.80,0.30,0.30);%cartroc(100,sg10000,10000,0.1,0.70,0.30,0.30);%cartroc(100,sh10000,10000,0.1,0.60,0.30,0.30);** 0.20 correlation on low correlated variables ***;** simulations on 100 patients *****;%cartroc(100,si100,100,0.2,0.90,0.30,0.30);%cartroc(100,sj100,100,0.2,0.80,0.30,0.30);%cartroc(100,sk100,100,0.2,0.70,0.30,0.30);%cartroc(100,sl100,100,0.2,0.60,0.30,0.30);** simulations on 1000 patients *****;%cartroc(100,si1000,1000,0.2,0.90,0.30,0.30);%cartroc(100,sj1000,1000,0.2,0.80,0.30,0.30);%cartroc(100,sk1000,1000,0.2,0.70,0.30,0.30);%cartroc(100,sl1000,1000,0.2,0.60,0.30,0.30);** simulations on 10000 patients *****;%cartroc(100,si10000,10000,0.2,0.90,0.30,0.30);%cartroc(100,sj10000,10000,0.2,0.80,0.30,0.30);%cartroc(100,sk10000,10000,0.2,0.70,0.30,0.30);%cartroc(100,sl10000,10000,0.2,0.60,0.30,0.30);** 0.30 correlation on low correlated variables ***;** simulations on 100 patients *****;%cartroc(100,sm100,100,0.3,0.90,0.30,0.30);%cartroc(100,sn100,100,0.3,0.80,0.30,0.30);%cartroc(100,so100,100,0.3,0.70,0.30,0.30);%cartroc(100,sp100,100,0.3,0.60,0.30,0.30);** simulations on 1000 patients *****;%cartroc(100,sm1000,1000,0.3,0.90,0.30,0.30);%cartroc(100,sn1000,1000,0.3,0.80,0.30,0.30);%cartroc(100,so1000,1000,0.3,0.70,0.30,0.30);%cartroc(100,sp1000,1000,0.3,0.60,0.30,0.30);* simulations on 10000 patients *****;%cartroc(100,sm10000,10000,0.3,0.90,0.30,0.30);%cartroc(100,sn10000,10000,0.3,0.80,0.30,0.30);%cartroc(100,so10000,10000,0.3,0.70,0.30,0.30);%cartroc(100,sp10000,10000,0.3,0.60,0.30,0.30);C.3.Simulation ScenariosTable C STYLEREF 1 \s 0. SEQ Table \* ARABIC \s 1 1 Summary of number of datasets created for main simulationsMost cells in this table represent 3 datasets of size 100, 1000, 10000 (n=3). Cells with n=8, represent additional datasets for missing values and dichotomisation, which were not created on all scenarios.Table C STYLEREF 1 \s 0. SEQ Table \* ARABIC \s 1 2 Summary of number of data points generated in simulationsSummary of simulation scenarios createdWith stats on number of datasets and patients C.4.Simulation PosterAppendix D: Analyses D.1.Primary CART Analysis > # clear datasets> rm(list=ls())> > # read data from BCE> allall <-read_bce('/opt/BIOSTAT/home_ext2/mahonep1/phd/libraries/allall.sas7bdat')> > ########################################> # a function written to calculate performance > ########################################> performance <- function(table, n=2){ if(!all(dim(table) == c(2,2)))+ stop("Must be a 2 x 2 table")+ tn = table[1,1]+ fp = table[1,2]+ fn = table[2,1]+ tp = table[2,2]+ sensitivity = tp/(tp+fn)+ specificity = tn/(tn+fp)+ ppp = tp/(tp+fp)+ npp = tn/(tn+fn)+ hitrate = (tp+tn)/(tp+tn+fp+fn)+ result <- paste("Sensitivity = ", round(sensitivity, n) ,+ "\nSpecificity = ", round(specificity, n),+ "\nPositive Predictive Value = ", round(ppp, n),+ "\nNegative Predictive Value = ", round(npp, n),+ "\nAccuracy = ", round(hitrate, n), "\n", sep="")+ cat(result)+ }> > ########################################> # all patients and biomarkers only> ########################################> > ########################################> # read in mega dataset> ########################################> setwd("~/phd")> #setwd("/opt/BIOSTAT/home_ext2/mahonep1/phd")> #allall <- read.csv("./allall.csv", header=T)> # make ACR20 a factor variable for Random Forest> allallc1 <-transform(allall, ACR20=as.factor(ACR20))> allallc <- subset(allallc1, ACR20!="NA")> > # make region binary and drop region in drops3> allallc$REGIONNA <- as.factor(ifelse(allallc$REGION=="NORTH AMERICA", "North America", "Non North America"))> > drops3 <-c("ACRN","ACR50","ACR70","ACR90","CRTN","PT","DSETTYPE","DASEW24","DASCW24","DASW24"+ ,"COMPOUND","DRUGTYPE","POPIR","POPRA","SMOKHIS","YOE","REGION", "SJC24","TJC24"+ ,"CRP24","ESR24","VPAIN24","HAQ24","VPHGS24","VPTGS24")> allallred2 <- allallc[,!(names(allallc) %in% drops3)]> > # drop some study type variables> drops4 <-c("TRTMENT","PROTO")> allallred3 <- allallred2[,!(names(allallred2) %in% drops4)]> > ########################################> # Create train and test datasets> ########################################> > ## 75% of the sample size> ## 75% of the sample size> smp_size <- floor(0.75 * nrow(allallred3))> > ## set the seed to make partition reproducible> set.seed(1234)> train_ind1 <- sample(seq_len(nrow(allallred3)), size = smp_size)> > trainall1 <- allallred3[train_ind1, ]> testall1 <- allallred3[-train_ind1,]> > ########################################> # CART Analysis on all> ########################################> library(rpart)> library(rpart.plot)> > trainfitall1<- rpart(ACR20~ ., data=trainall1, method="class")> trainfitall1$cptable CP nsplit rel error xerror xstd1 0.03312136 0 1.0000000 1.0000000 0.012161512 0.02287582 2 0.9337573 0.9401166 0.012067973 0.01000000 5 0.8651298 0.9054054 0.01199742> plotcp(trainfitall1)> > trainfitall1.pruned <- prune(trainfitall1, cp=0.01343166)> trainfitall1.prunedn= 8542 node), split, n, loss, yval, (yprob) * denotes terminal node 1) root 8542 3774 0 (0.5581831 0.4418169) 2) BTJC< 7.5 815 166 0 (0.7963190 0.2036810) * 3) BTJC>=7.5 7727 3608 0 (0.5330659 0.4669341) 6) REGIONNA=North America 3085 1162 0 (0.6233387 0.3766613) * 7) REGIONNA=Non North America 4642 2196 1 (0.4730720 0.5269280) 14) JSNSCORE>=0.4903846 3993 1977 0 (0.5048835 0.4951165) 28) DMARDNUM>=3.5 473 175 0 (0.6300211 0.3699789) * 29) DMARDNUM< 3.5 3520 1718 1 (0.4880682 0.5119318) 58) RACE=ASIAN,OTHER,UNKNOWN 1508 686 0 (0.5450928 0.4549072) * 59) RACE=AMERICAN INDIAN OR ALASKA NATIVE,BLACK,HISPANIC,MULTIPLE,NATIVE HAWAIIAN/OTHER PACIFIC ISLANDER,WHITE 2012 896 1 (0.4453280 0.5546720) * 15) JSNSCORE< 0.4903846 649 180 1 (0.2773498 0.7226502) *> summary(trainfitall1.pruned)Call:rpart(formula = ACR20 ~ ., data = trainall1, method = "class") n= 8542 CP nsplit rel error xerror xstd1 0.03312136 0 1.0000000 1.0000000 0.012161512 0.02287582 2 0.9337573 0.9401166 0.012067973 0.01000000 5 0.8651298 0.9054054 0.01199742Variable importance BTJC REGIONNA JSNSCORE DMARDNUM PSYCH RACE IMMUNE WEIGHT GENERAL CNS 27 22 14 6 5 5 3 3 3 3 RADUR BSJC BCRP MCS PCS RBC SF0101 2 2 1 1 1 1 1 Node number 1: 8542 observations, complexity param=0.03312136 predicted class=0 expected loss=0.4418169 P(node) =1 class counts: 4768 3774 probabilities: 0.558 0.442 left son=2 (815 obs) right son=3 (7727 obs) Primary splits: BTJC < 7.5 to the left, improve=102.27610, (2 missing) SHARP < 5.5 to the right, improve= 90.41307, (4307 missing) JSNSCORE < 0.5 to the right, improve= 83.54683, (3703 missing) REGIONNA splits as RL, improve= 79.86411, (0 missing) EROSION < 4.5 to the right, improve= 69.53788, (4307 missing) Surrogate splits: BSJC < 5.5 to the left, agree=0.912, adj=0.081, (0 split)Node number 2: 815 observations predicted class=0 expected loss=0.203681 P(node) =0.09541091 class counts: 649 166 probabilities: 0.796 0.204 Node number 3: 7727 observations, complexity param=0.03312136 predicted class=0 expected loss=0.4669341 P(node) =0.9045891 class counts: 4119 3608 probabilities: 0.533 0.467 left son=6 (3085 obs) right son=7 (4642 obs) Primary splits: REGIONNA splits as RL, improve=83.69613, (0 missing) SHARP < 4.5 to the right, improve=63.54933, (4279 missing) RADUR < 0.939863 to the right, improve=57.45138, (0 missing) JSNSCORE < 0.5 to the right, improve=54.34351, (3675 missing) EROSION < 7.5 to the right, improve=52.58031, (4279 missing) Surrogate splits: PSYCH < 0.5 to the right, agree=0.684, adj=0.209, (0 split) IMMUNE < 0.5 to the right, agree=0.663, adj=0.155, (0 split) WEIGHT < 85.25 to the right, agree=0.662, adj=0.153, (0 split) GENERAL < 0.5 to the right, agree=0.653, adj=0.131, (0 split) CNS < 0.5 to the right, agree=0.650, adj=0.124, (0 split)Node number 6: 3085 observations predicted class=0 expected loss=0.3766613 P(node) =0.3611566 class counts: 1923 1162 probabilities: 0.623 0.377 Node number 7: 4642 observations, complexity param=0.02287582 predicted class=1 expected loss=0.473072 P(node) =0.5434325 class counts: 2196 2446 probabilities: 0.473 0.527 left son=14 (3993 obs) right son=15 (649 obs) Primary splits: JSNSCORE < 0.5 to the right, improve=54.29425, (2247 missing) SHARP < 5.5 to the right, improve=45.59904, (2709 missing) RADUR < 3.278329 to the right, improve=40.66651, (0 missing) EROSION < 7.5 to the right, improve=39.38122, (2709 missing) DMARDNUM < 3.5 to the right, improve=33.62045, (512 missing) Surrogate splits: RADUR < 0.517631 to the right, agree=0.797, adj=0.159, (2247 split) AGE < 18.5 to the right, agree=0.760, adj=0.007, (0 split) RACE splits as LLL-LRLLL, agree=0.759, adj=0.002, (0 split)Node number 14: 3993 observations, complexity param=0.02287582 predicted class=0 expected loss=0.4951165 P(node) =0.4674549 class counts: 2016 1977 probabilities: 0.505 0.495 left son=28 (473 obs) right son=29 (3520 obs) Primary splits: DMARDNUM < 3.5 to the right, improve=24.40744, (504 missing) BSJC < 11.5 to the left, improve=16.42077, (0 missing) PROT < 69.15 to the left, improve=14.11071, (33 missing) SF1103 < 2.5 to the left, improve=13.46342, (422 missing) BCRP < 7.815 to the right, improve=12.78504, (1 missing) Surrogate splits: BCRP < 107.4 to the right, agree=0.866, adj=0.004, (503 split)Node number 15: 649 observations predicted class=1 expected loss=0.2773498 P(node) =0.07597752 class counts: 180 469 probabilities: 0.277 0.723 Node number 28: 473 observations predicted class=0 expected loss=0.3699789 P(node) =0.05537345 class counts: 298 175 probabilities: 0.630 0.370 Node number 29: 3520 observations, complexity param=0.02287582 predicted class=1 expected loss=0.4880682 P(node) =0.4120815 class counts: 1718 1802 probabilities: 0.488 0.512 left son=58 (1508 obs) right son=59 (2012 obs) Primary splits: RACE splits as RLRRRRLLR, improve=17.15819, (0 missing) BCRP < 8.54 to the right, improve=14.78012, (1 missing) PROT < 73.25 to the left, improve=14.46744, (31 missing) SF0101 < 2.5 to the left, improve=13.50631, (141 missing) BSJC < 11.5 to the left, improve=13.25879, (0 missing) Surrogate splits: BCRP < 4.995 to the right, agree=0.691, adj=0.280, (0 split) MCS < 16.5 to the right, agree=0.654, adj=0.192, (0 split) PCS < 4.5 to the right, agree=0.653, adj=0.189, (0 split) RBC < 4.335 to the left, agree=0.651, adj=0.186, (0 split) SF0101 < 2.5 to the left, agree=0.641, adj=0.161, (0 split)Node number 58: 1508 observations predicted class=0 expected loss=0.4549072 P(node) =0.1765395 class counts: 822 686 probabilities: 0.545 0.455 Node number 59: 2012 observations predicted class=1 expected loss=0.445328 P(node) =0.235542 class counts: 896 1116 probabilities: 0.445 0.555 > > prp(trainfitall1.pruned, type = 2, extra = 104,+ fallen.leaves = TRUE, main="Decision Tree All Data")> > fitall1.pred <- predict(trainfitall1.pruned, testall1, type="class")> fitall1.perf <- table(testall1$ACR20, fitall1.pred, + dnn=c("Actual", "Predicted"))> fitall1.perf PredictedActual 0 1 0 1200 369 1 757 522> performance(fitall1.perf)Sensitivity = 0.41Specificity = 0.76Positive Predictive Value = 0.59Negative Predictive Value = 0.61Accuracy = 0.6######################################### randomforest using randomForestSRC########################################> library(randomForestSRC)> set.seed(1234)> rfsrcfit1 <- rfsrc(ACR20~.,data=rftrain, na.action="na.impute", importance=TRUE)> rfsrcfit1 Sample size: 8542 Frequency of class labels: 4768, 3774 Was data imputed: yes Number of trees: 1000 Minimum terminal node size: 1 Average no. of terminal nodes: 1810.558No. of variables tried at each split: 12 Total no. of variables: 135 Analysis: RF-C Family: class Splitting rule: gini Normalized Brier score: 92.11 Error rate: 0.39, 0.19, 0.64Confusion matrix: predicted observed 0 1 class.error 0 3845 923 0.1936 1 2390 1384 0.6333Overall error rate: 38.76% all 0 1JSNSCORE 0.0036 0.0019 0.0005SHARP 0.0036 0.0019 0.0005BTJC 0.0030 0.0016 0.0005EROSION 0.0028 0.0013 0.0007REGIONNA 0.0028 0.0009 0.0012RADUR 0.0019 0.0002 0.0013BVASPHGS 0.0017 0.0010 0.0001RACE 0.0016 0.0010 0.0000MCS 0.0015 0.0008 0.0003BILT 0.0014 0.0007 0.0002SCRT 0.0013 0.0007 0.0002WITHERN 0.0012 0.0012 -0.0005BVASPTGS 0.0012 0.0007 0.0001RE_Z 0.0012 0.0007 0.0001DMARDNUM 0.0012 0.0004 0.0005SF0101 0.0010 0.0005 0.0002PCS 0.0010 0.0007 0.0000BSJC 0.0009 0.0006 0.0001RH_Z 0.0009 0.0004 0.0002SF_Z 0.0009 0.0004 0.0002SF0908 0.0009 0.0006 0.0000JSNHANDS 0.0009 0.0008 -0.0003PSYCH 0.0008 0.0004 0.0001BCRP 0.0008 0.0006 0.0000BESR 0.0008 0.0005 0.0001SF0601 0.0008 0.0004 0.0002> plot(rfsrcfit1)> rftest <-transform(testall1, SEX=as.factor(SEX), RACE=as.factor ADDIN EN.CITE <EndNote><Cite ExcludeYear="1"><Author>Ranganath</Author><Year>2005</Year><RecNum>60</RecNum><DisplayText>(Ranganath et al.)</DisplayText><record><rec-number>60</rec-number><foreign-keys><key app="EN" db-id="5a99wdaazwdxt3ex5pfppae3fspswvtfrxrd" timestamp="0">60</key></foreign-keys><ref-type name="Journal Article">17</ref-type><contributors><authors><author>Ranganath, Veena</author><author>Elashoff, David</author><author>Khanna, Dinesh</author><author>Park, Grace</author><author>Peter, James</author><author>Paulus, Harold</author><author>Western consortium of practicing rhuemotologists.</author></authors></contributors><auth-address>Division of Rheumatology, Department of Biostatistics and School of Public Health, Geffen School of Medicine at UCLA, Los Angeles, California 90095-1670, USA</auth-address><titles><title>Age adjustment corrects for apparent differences in erythrocyte sedimentation rate and C-reactive protein values at the onset of seropositive rheumatoid arthritis in younger and older patients</title><secondary-title>The Journal of rheumatology</secondary-title></titles><pages>1040-1042</pages><volume>32</volume><number>6</number><reprint-edition>Not in File</reprint-edition><keywords><keyword>9007-41-4 (C-Reactive-Protein)</keyword><keyword>ADULT</keyword><keyword>AGE-FACTORS</keyword><keyword>AGED</keyword><keyword>ARTHRITIS-RHEUMATOID/*BL (blood),PP (physiopathology)</keyword><keyword>BLOOD-SEDIMENTATION/*</keyword><keyword>C-REACTIVE-PROTEIN/*AN (analysis)</keyword><keyword>COHORT-STUDIES</keyword><keyword>FEMALE</keyword><keyword>HEALTH-STATUS</keyword><keyword>HUMANS</keyword><keyword>MALE</keyword><keyword>MIDDLE-AGED</keyword><keyword>PROSPECTIVE-STUDIES</keyword><keyword>QUESTIONNAIRES</keyword><keyword>RA</keyword><keyword>Rheumatoid arthritis</keyword><keyword>SEVERITY-OF-ILLNESS-INDEX</keyword></keywords><dates><year>2005</year><pub-dates><date>6/2005</date></pub-dates></dates><label>61</label><urls></urls><access-date>2006</access-date></record></Cite></EndNote>(Ranganath et al.),+ RFCAT=as.factor(RFCAT), BSTEROID=as.factor(BSTEROID))> rfsrc1.pred <- predict(rfsrcfit1, rftest, na.action="na.impute")> rfsrc1.pred Sample size of test (predict) data: 2848 Was test data imputed: yes Number of grow trees: 1000 Average no. of grow terminal nodes: 1810.558 Total no. of grow variables: 135 Analysis: RF-C Family: class Test set Normalized Brier score: 92.38 Test set error rate: 0.38, 0.17, 0.64Confusion matrix: predicted observed 0 1 class.error 0 1304 265 0.1689 1 811 468 0.6341Overall error rate: 37.82% ADDIN ADDIN ................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.
Related searches
- baking soda rheumatoid arthritis recipe
- injections for rheumatoid arthritis treatment
- rheumatoid arthritis shots
- injections rheumatoid arthritis treatment
- rheumatoid arthritis injection medication
- rheumatoid arthritis treatment options
- rheumatoid arthritis shot treatments
- rheumatoid arthritis injections
- medicines for rheumatoid arthritis pain
- medicine for rheumatoid arthritis treatment
- rheumatoid arthritis otc medications
- signs of rheumatoid arthritis in women