Introduction - Milken Institute School of Public Health



Example – Data Analysis Plan (DAP)Contents TOC \o "1-3" \h \z \u Introduction2Background2Study Objectives4Study Design and Sampling 4Study duration definitions4Definition of the Study Population5Determination of the Study Cohorts and Groups6Classification of study groups7Study Measures and Variable Definitions8Comorbid conditions/ Charlson Comorbidity Index (CCI) 12Compliance: Medication Possession Ratio (MPR)13Compliance: Proportion of days covered (PDC)14Persistence15Analysis16Study limitations17Appendix 1: Codes to identify Patient Cohort18A.Included Diagnosis18B.Included Medications19Table shells 20IntroductionThis data analysis plan (DAP) contains definitions of study period, study groups/cohorts, data elements and statistical methods for understanding the effects of an intervention to improve mediation adherence.BackgroundPoor medication adherence not only affects patients, but the entire healthcare system. Non-adherent patients are more likely to experience worse outcomes, including early death, than those who are adherent to their medication regimens.,, Poor adherence also is associated with increased utilization of healthcare resources and intensification of medical therapy, as providers want to achieve the desired clinical outcomes. According to a study, patients who were persistent with their medications incurred 12.5% lower medical costs. Indirect medical costs to employers in the form of absenteeism also exist with poor adherence. Costs incurred by employers include absenteeism due to illness and employer-sponsored health plan claims. Initiatives that improve medication adherence have demonstrated substantial cost savings, especially through reduced hospitalization and emergency department use. Few employer efforts to adjust health plans to directly affect anxiety and depression groups have been evaluated and documented specifically in regards to absenteeism. As employers become more involved in efforts to improve healthcare quality and control medical expenditures, maximizing medication adherence rates represents an important opportunity. One of the strategies successfully used by employers in improving medication adherence is lowering copayments for medications used to treat chronic conditions. Cost sharing between insurers and insured has been found to have a significant impact on medication initiation. Cost-related changes that have had an impact on adherence include: Associations between pharmacy utilization and prescription copaymentsFormulary tiersCoinsurancePharmacy benefit capsPrescription limitsFormulary restrictionsReference pricing. Increased cost-sharing for medications is associated with lower rates of initiation of prescriptions, poorer adherence among users, and more frequent discontinuation of medication. A change in the cost to evaluate the impact on adherence for those with anxiety and depression can help determine strategies to increase adherence in these challenging groups.Study ObjectivesThe aim of this study is to understand the effects of an intervention to improve medication adherence. The primary objective is Assess whether medication adherence (as measured by medication possession ratio (MPR)) among the study population was affected by the waiver of copay requirements. Secondary objectives include:Determine the relationship between the intervention and initiation of specific medications.Determine the relationship between the intervention and healthcare utilization and costs (medical and pharmacy) over time.Determine the relationship between the intervention and absenteeism due to illness.Study Design and SamplingMethodsStudy duration definitionsA historical cohort will be used to assess the effects of the intervention, which was initiated on January 1, 2012. This cohort, Control Group, will be composed of members flagged with an index event between January 1, 2010 and December 31, 2010 (ie, the control intake period). The second cohort, Intervention Group, will be composed of members flagged with one of the index events between January 1, 2012 and December 31, 2012 (ie, the intervention intake period). Following the first occurrence of the index event during either intake period, each study participant will be followed for 365 days. The two cohorts will be evaluated independently and, therefore, some members may be included in both groups. A depiction of the study timeline is presented in Table 1.Table SEQ Table \* ARABIC 1: Depiction of study timelineJul. 1, 2009Jan. 1, 2010Dec. 31, 2010Jul. 1, 2011Dec. 31, 2011Jan. 1, 2012*Dec. 31, 2012Dec. 31, 2013Control pre-index periodIntervention pre-index periodControl intake periodIntervention intake periodControl follow-up periodIntervention follow-up period*The intervention was initiated on January 1, 2012.Definition of the Study PopulationInclusion criteriaMembers with a qualifying index event during the control and/or intervention intake periods. A qualifying index event can be either 1) a medical claim that included an ICD-9 code for anxiety or depression (see Appendix 1A for a list of included codes) OR 2) a pharmaceutical claim for a medication commonly used for treatment of anxiety or depression (see Appendix 1B for a list of included medications).Members who were continuously enrolled for at least 180 days prior to the qualifying index event (pre-index period).Members who were continuously enrolled i for at least 365 days after the qualifying index event (follow-up period) the first occurrence of the index event during the control and/or intervention intake period.Exclusion criteriaPatients who do not have the diagnosis code for the relevant medical condition in the pre-index period (i.e. prior to the index prescription) or at the index prescription. Age younger than 18 years at any point during the study period.Determination of the Study Cohorts and GroupsA flow diagram of the inclusion/exclusion criteria is shown in Figure 1. Figure SEQ Figure \* ARABIC 1: Flowchart of Study CohortsClassification of study groupsIntervention Group = Members who have a qualifying index event between 1/1/2010 and 12/31/10 who are followed for 365 days prior to the waiver of copay requirements. Control Group = Members who have a qualifying index event between 1/1/2012 and 12/31/12 who are followed for 365 days after the waiver of copay requirements. In addition to classifying groups by copay waiver status, we would like to stratify the results by the following:Condition (i.e., anxiety and depression)Subscriber/dependent statusCondition ManagementCondition Management Member – member engaged in the condition management program.Non-Condition Management Member – member NOT utilizing condition management servicesStudy Measures and Variable DefinitionsPopulation/cohort demographics?VariableDefinitionTypeValue/categoryData SourceUnique IDRandomly generated unique member ID for HIPPA compliance0.Control/Intervention Group StatusControl group will be those eligible from January 1st 2010 to December 31st 2011. Intervention Group will be January 1st 2012 to December 31 2013. Member/Determine the member/dependent statusDependent statusLocation of care=KansasDetermine if the enrollee is based in Kansas city(See Appendix 1 for algorithm)Start of eligibilityStart date of eligibility during a calendar year for each member/dependent.End of eligibilityEnd date of eligibility during a calendar year for each member/dependent.AgeAge at start of study period. ?GenderThe member’s genderEnrolled in Disease managementThese variables indicate whether the patient was enrolled in DSM programs for each calendar year. Create variables for each DSM program separately??? 8 separate variables for each enrollee (see Appendix A)?? Hypertension?? Diabetes?? Weight managementDepressionCCI (Charlson Comorbid Index)Use the algorithm to determine CCI for intervention group and control groupVariableDefinitionPrescription fill dateDate of each fill for depression or anxiety medicationsNew medication userMembers who did not receive the medication during the 6-month pre-index period Total day supply (for MPR)Sum of all the day supply for each enrollee. When using a fixed time interval truncate days of utilization that falls after the interval.For example, Index date is 01/01/2009. Last refill date within 365 days falls on 12/20/2009 and patient refilled for 30 days, then only the first 12 days of the day supply fall within the follow-up period i.e. within 365 days, therefore truncate 18 days from the last day supply (i.e. 30-18=12) MPRMPR for each disease area calculated as Total day supply/ Days of follow-up?? Note that if patients refill medications before the end of the earlier supply, the MPR can exceed 1. In this scenario adjust MPR to a max of 1. MPRs that are greater than or equal to 1.5 will be excluded.Number of concomitant medicationsDefined as the number of unique medications for which each enrollee had prescriptions filled during the first 365 days of follow-up. These medications are not limited to ones included in the study and not grouped by therapeutic class (overlapping periods of medications in the same class are not considered).Onsite pharmacy user?Determine if the member uses the onsite pharmacy for all Rx fills. Patients who switch between onsite and offsite pharmacy will be excluded. ?VariableDefinitionPrescription spendTotal prescription spend Medical spendTotal Medical spend Total spendTotal prescription and medical spend Average Co-pay per visitCalculated as average expense incurred by the patient across all pharmacy encountersAbsenteeismNumber of sick daysPOSHealthcare setting, place of service (POS), where service occurred. See Appendix 1 for categorization of POS.IPCount total number of inpatient (IP) visits for each member during a calendar yearOPCount total number of outpatient (OP) visits for each member during a calendar yearERCount total number of ER visits for each member during a calendar yearPhysicianCount total number of physician visits for each member during a calendar yearOtherCount total number of other type of visits for each member during a calendar yearMissingCount total number of “missing” type of visits for each member during a calendar yearOutcome measuresCompliance: Medication Possession Ratio (MPR)One of the most common methods to measure compliance using retrospective databases is to calculate the MPR. MPR will be calculated in two ways:Refill Interval: by dividing the total day supply of medication(s) by the number of days within the refill interval. To calculate ratio, we need at least 2 fill dates (e.g., index date and at least 1 refill). The first Rx Date will be the index date; the last Rx Date will be the last refill date included in the 365 day follow-up period. MPR isSwitching between drug classes will be allowed (i.e. a day will be assumed to be covered if any drug in that therapeutic class is available). These estimates therefore represent an upper bound of the actual compliance with prescribed therapy. Fixed Interval: by dividing the total day supply of medication(s) by a set number days. One fixed interval will be included in this study of 365 days. These value will be the denominators, regardless of how long a patient was on the medication. The 365 day fixed interval will be defined as:365 Fixed Interval = (total Rx days of supply) / (fixed interval of 365 days)Switching between drug classes will be allowed (i.e. a day will be assumed to be covered if any drug in that therapeutic class is available). These estimates therefore represent an upper bound of the actual compliance with prescribed therapy. Note: MPR can exceed 1 if there are early refills. For MPRs greater than 1 and less than 1.5, truncate these to 1. For those greater than 1.5, exclude the data. Compliance: Proportion of days covered (PDC)Provides more conservative estimate of medication adherence (compared to MPR) when multiple medications are intended to be used concomitantly. PDC avoids double-counting days of medication coverage because a day is only counted if all medications are available on that day. PDC values range from 0 to 1. Because PDC will be equivalent to MPR for single medications, only calculate PDC for members with multiple medications (any medication type). PDC will be calculated only for the first 365 days of follow up. Note: PDC can exceed 1 if there are early refills. For PDCs greater than 1 and less than 1.5, truncate these to 1. For those greater than 1.5, exclude the data. Example of PDC versus MPRPersistencePersistence adds the dimension of time to the analysis and usually represents the time over which a patient continues to fill a prescription or discontinue medication. We will calculate the estimated level of persistence (ELPT) at 90 days, 180 days and 1 year (360 days) defined as the percentage of individuals remaining on therapy (persistent) at that time without a 30-day gap [see figure below]. Persistence will also be defined as days to discontinuation and measured as days from day of first claim to day of first gap that is ≥ 30 days, ≥ 45 days and ≥ 60 days. Persistence will help differentiate patients taking a medication sporadically during a defined time frame from those patients stopping the medication early during the same time. Data will be presented in results as in the following example.Note, previous early refills will not be considered to balance a refill gap. The data may also be displayed similar to a Kaplan–Meier curve, where discontinuation will be considered as a censoring event. We will also report the Percent Fill calculated as % of patients who refilled their medication within 30 days of the expected refill date. We may also consider reporting on Percent Fill within 60 or 90 days.AnalysisStatistical analysis will be conducted at two levels. Descriptive and multivariable adjusted analyses will be conducted to examine the change in medication adherence as measured as MPR from pre-prescription cost reduction and then post-reduction. This change will then be compared to a control group change from 2010 to 2011. Validation checks of the data will be run using descriptive statistics as well as graphical depictions, and data that appear missing or incorrect will be corrected when possible and otherwise excluded from analysis. Continuous variables will be expressed as the mean ± standard deviation and compared within pre-post groups using paired t tests and between groups using independent t-tests. Categorical variables will be expressed as absolute (number) and relative frequencies (percentage), with proportions compared within groups using McNemar’s test and between groups using Chi-Square test.Multivariable regression will be used to adjust for possibly measured confounding by controlling for the following covariates: age, gender, subscriber/dependent status, Charlson comorbidity index, average monthly prescription cost, and number of concomitant mediations. Primary and secondary outcomes methods will include but not are limited to:Linear multivariable regression adjusting for potential confounders (above) will be employed to measure the difference in MPR change over time for the intervention group vs. MPR change in the control group. A fixed effect for control group will be included in models to measure the 2010-2011 control group difference to the 2011-2012 intervention group difference. The dichotomous outcome of medication initiation will be modeled using multivariable modified Poisson regression and results will be represented as relative risks. The risk of initiating mediations in the control group will be compared to the risk for the intervention group using a fixed group effect while adjusting for potential confounders.Change in cost between groups will be modeled using multivariable linear regression on log-transformed cost due to the inherently skewed nature of cost data. Covariates will include potential confounders and a fixed group effect will estimate group differences. Employee absenteeism will be represented as an integer count for number of sick days and will be modeled using multivariable Poisson regression adjusting for covariates. For all statistical tests, a 2-tailed P value of <0.05 will be considered statistically significant. All statistical analyses will be performed using SAS, version 9.2 (SAS Institute Inc, ) and R (R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ).Study limitations Most limitations that may have confounded the results affected all study groups, therefore preserving the relative difference between groups. For example, the study did not account for discontinuation of medication due to improved health outcomes, pill splitting unknown to the prescriber and exclusion of patients who switched between onsite and offsite health centers. Patients with certain healthcare needs, motivations or knowledge may have had a greater tendency to seek treatment at the onsite health center and/or maintain medication adherence, which could have skewed the results. Additionally, the inclusion of only employees and their dependents may limit the generalizability of the findings.Discontinuation of medication can indicate a favorable health outcome, although many of the conditions listed are lifelong for most patients.Included patients with multiple conditions and depression, which are associated with lower medication adherence. We will account for this in the multivariate analysis thoughA substantial proportion of patients may be taking antidepressants or antianxiety medications for reasons other than depression and anxiety.Although pill splitting known to prescriber will be taken into account, pill splitting unknown to prescriber is a limitation.Due to the date shifting (Appendix 2A), responsiveness to intervention may not be reflected in the data.Will be a “prevalent population” and patients do not necessarily initiate a “new” script during the intake period; therefore, adherence rates may be lower since they diminish with time.Absenteeism due to illness can only be assessed for employees, not their dependents. Appendix 1: Codes to identify Patient CohortIncluded DiagnosisFor each of the following conditions, determination will be based on whether the condition was seen during the study period. Both primary or secondary diagnosis and procedure codes will be used.VariableDefinitionTypeValue/categoryAnxietyAny of the diagnoses below:ICD9 DxDescription293.84Anxiety disorder in conditions classified elsewhere300.0Anxiety states300.02Phobia, unspecified300.09Other309.21Separation anxiety disorder309.24Adjustment disorder with anxiety309.28Adjustment disorder with mixed anxiety and depressed moodBinary1 = Yes0 = NoDepressionAny of the diagnoses below:ICD9 DxDescription296.2Major depressive disorder, single episode296.3Major depressive disorder, recurrent episode300.4Dysthymic disorder309.10Prolonged depressive reactionBinary1 = Yes0 = NoIncluded MedicationsAnxietyDepression BUPROPIONAPLENZINBUPROPION HCLBUDEPRION SRBUPROPION XLBUDEPRION XLBUSPIRONE HCLBUPROPION HCLCITALOPRAMBUPROPION HCL SRCITALOPRAM HBRBUPROPION XLESCITALOPRAMCELEXAFLUOXETINECITALOPRAM HBRPAROXETINE HCLCYMBALTASERTRALINEEFFEXORSERTRALINE HCLEFFEXOR XRVENLAFAXINE HCLFLUOXETINE HCLVENLAFAXINE HCL XRFLUVOXAMINE MALEATEVENLAFAXINE XRLEXAPRONEFAZODONE HCLPAROXETINE HCLPAXILPAXIL CRPEXEVAPRISTIQPROZACSELFEMRASERTRALINE HCLTRANYLCYPROMINE SULFATETRAZODONE HCLVENLAFAXINE HCLVENLAFAXINE HCL ERWELLBUTRIN SRWELLBUTRIN XLZOLOFTAppendix 2: Algorithms to De-Identify DataRe-IdentificationThe sponsor has no reason to believe that the information could be used alone or in combination with other information to identify individuals once the shifting method has been applied. The algorithm used to shift the dates is known by a few qualified people who maintain the database and code. One would have to know the algorithm used and be a statistician to re-identify the shifted dates. The environment that the analysts have access to has the blinding already applied. The access to the database is managed by role, controlled by the security officer and an audit trail is maintained for all status changes. The access list is reviewed at least annually and all processes are validated by the regulatory affairs team. The algorithm used to re-identify is not released when the de-identified information is provided to 3rd parties nor is it released to anyone that is not approved by the security officer.Date Shifting AlgorithmAll data containing dates will be shifted by a random number. Zip Code AlgorithmThe initial three digits of a zip code will be used that, according to publicly available data from the Bureau of Census, include more than 20,000 people in the geographical unit formed by combining all zip codes with the initial three digits. Those geographical areas that include 20,000 or fewer people will be changed to “000”.Control and Intervention GroupingControl and Intervention groups will first be defined by the eligibility date range and be provided in the data as either a 1 or a 0 representing eligible or ineligible during the study period. Claims data to derive other inclusion criteria will be provided with an implemented data shifting algorithm. The data will include 35 days before the inclusion date range to capture changes during the period. Dates between the two groups will be shifted randomly between the Control and Intervention groups.Table shells ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download