Lippincott Williams & Wilkins



Online Supplement:Risk-Adjustment for Sepsis Mortality to Facilitate Hospital Comparisons Using CDC’s Adult Sepsis Event Criteria and Routine Electronic Clinical DataTable of ContentsSupplemental Methods. Description of Multiple Imputation Methods for Missing Severity-of-Illness CovariatesSupplemental Table 1. Study Hospital Characteristics for Primary Dataset (Cerner HealthFacts) and External Validation Dataset (HCA Healthcare)Supplemental Table 2. Centers for Disease Control and Prevention Adult Sepsis Event CriteriaSupplemental Table 3. Normal Values Imputed for Missing Physiologic CovariatesSupplemental Table 4. Characteristics of Adult Sepsis Event Patients With and Without Missing Vital Signs (Temperature, Blood Pressure, and Respiratory Rate) on Day of Sepsis OnsetSupplemental Table 5. Characteristics of Risk-Adjustment Models for In-Hospital Mortality in Primary (Cerner) and External Validation (HCA Healthcare) Datasets Supplemental Table 6. Sepsis In-Hospital Mortality Prediction Model Results Based on Administrative Data and Adult Sepsis Event Organ Dysfunction Criteria (Clinical Model 1)Supplemental Table 7. Sepsis In-Hospital Mortality Prediction Model Results Based on Administrative Data, Adult Sepsis Event Organ Dysfunction Criteria, and Extended Laboratory Data (Clinical Model 2)Supplemental Table 8. Mean/Standard Deviation and Median/Interquartile Ranges for Physiologic CovariatesSupplemental Figure 1. Proportion of Encounters Missing Physiologic Covariates within 1 Day of Sepsis Onset in Cerner and HCA Healthcare DatasetsSupplemental Figure 2. Variable Importance Plot Demonstrating Change in Model Performance by Addition or Subtraction of Covariates in Clinical Model 1Supplemental Figure 3. Calibration Plot of Clinical Models Using (A) Multiple Imputation vs (B) Normal Value Imputation for Missing Covariates in External Validation Data (HCA Healthcare)Supplemental Methods. Description of Multiple Imputation Methods for Missing Severity-of-Illness CovariatesWe used the technique named multiple imputation by chained equations (MICE), where missing variables are predicted (imputed) using regression models with other variables. Specifically, we used Predictive Mean Matching (PMM) to avoid biologically impossible predictions (e.g., outside of plausible range). The procedure sequentially cycles through every variable that contains missing values (Azur, Stuart, Frangakis, & Leaf, 2011). We generated 5 imputed datasets; for each imputation, 20 cycles were performed. The following website offers a detailed description of the steps of PMM: ():“For cases with no missing data, estimate a linear regression of?x?on?z, producing a set of coefficients?b.Make a random draw from the “posterior?predictive?distribution” of?b, producing a new set of coefficients?b*. Typically this would be a random draw from a multivariate normal distribution with?mean?b?and the estimated covariance matrix of?b?(with an additional random draw for the residual variance). This step is necessary to produce sufficient variability in the imputed values, and is common to all “proper” methods for multiple imputation.Using?b*, generate?predicted?values for?x?for?all?cases, both those with data missing predicted on?x?and those with data present.For each case with missing x, identify a set of cases with observed x whose predicted values are close to the predicted value for the case with missing data. From among those close cases, randomly choose one and assign its?observed?value to substitute for the missing value.Repeat steps 2 through 5 for each completed data set.”Supplemental Table 1. Study Hospital Characteristics for Primary Dataset (Cerner HealthFacts) and External Validation Dataset (HCA Healthcare)Hospital CharacteristicCerner HealthFacts (N=136)HCA Healthcare (N=137)Region Northeast30 (22.1%)2 (1.5%) Midwest26 (19.1%)10 (7.3%) West32 (23.5%)20 (14.6%) South 48 (35.3%)105 (76.6%)Bed Size <20088 (64.7%)49 (35.8%) 200-49939 (28.7%)78 (56.9%) >5009 (6.6%)10 (7.3%)Teaching Status Teaching49 (36.0%)34 (24.8%) Nonteaching81 (59.6%)103 (75.2%) Unknown6 (4.4%)0 (0%)Supplemental Table 2. Centers for Disease Control and Prevention Adult Sepsis Event CriteriaComponentDetailed CriteriaPresumed Serious InfectionBlood culture obtained (regardless of result), and ≥4 “Qualifying Antibiotic Days” (QAD)a – starting within +/-2 days of blood culture dayAcute Organ Dysfunction (any one of the following within +/-2 days of blood culture day)Vasopressor initiationb (norepinephrine, dopamine, epinephrine, phenylephrine, or vasopressin)Mechanical ventilation initiationbDoubling in serum creatinine or decrease by ≥50% of estimated glomerular filtration rate relative to baselinec (excluding patients with ICD-9 code for end-stage kidney disease, 585.6) Total bilirubin ≥ 2.0 mg/dL and doubling from baselinecPlatelet count <100 cells/?L and ≥ 50% decline from baselinec (baseline must be ≥100 cells/?L)Serum lactate ≥ 2.0 mmol/LSepsis = presumed serious infection + ≥1 acute organ dysfunctiona QADs start with the first “new” antibiotic (not given in the prior 2 calendar days) within the +/-2 day period surrounding the day of the blood culture draw. Subsequent QADs can be different antibiotics as long as the first dose of each is “new.” Days between administration of the same antibiotic count as QADs as long as the gap is not >1 day. At least one of the first 4 QADs must include an intravenous antibiotic. If death or discharge to another acute care hospital or hospice occurs prior to 4 days, QADs are required each day until ≤ 1 day prior to death or discharge. b Vasopressors and mechanical ventilation are considered to be “initiated” during the +/-2 day period surrounding the day of the blood culture draw if there were no vasopressors or mechanical ventilation administered on the prior calendar day. c For presumed infection present-on-admission (blood culture day or first QAD occurring on hospital day 1 or 2), baseline lab values are defined as the best values during hospitalization. For hospital-onset infection (blood culture day and first QAD occurring on hospital day ≥ 3), baseline lab values are defined as the best values during the +/-2 day period surrounding the day of the blood culture draw.Supplemental Table 3. Normal Values Imputed for Missing Physiologic CovariatesParameter“Normal” ValueUnit of MeasurementCreatinine1.0mg/dLTotal Bilirubin0.3mg/dLPlatelet Count250K/?LLactate 0.5mmol/LWBC6K/?LAlbumin4.0g/dLAnion Gap12mEq/LHematocrit40%AST30Units/LSodium140mEq/LINR1.0-SBP120mmHgTemperature98.5? FarenheightRespiratory Rate12Breaths/minuteGCS15-Abbreviations: WBC = white blood cell count, AST = asparate aminotransferase, INR = international normalized ratio, SBP = systolic blood pressure, GCS = Glasgow Coma Scale. The above values were used to impute missing values for severity-of-illness covariates in developing and validating the risk-adjustment models incorporating clinical data. In a sensitivity analyses, models were developed using multiple imputation to account for missing severity-of-illness covariates. Supplemental Table 4. Characteristics of Adult Sepsis Event Patients With and Without Missing Vital Signs (Temperature, Blood Pressure, and Respiratory Rate) on Day of Sepsis OnsetCharacteristicAdult Sepsis Event Patients with Non-Missing Vital Signs (n=50,118)Adult Sepsis Event Patients with Missing Vital Signs(n=45,036)Median Age (IQR)67 (55-79)68 (55-80)Male Sexa (N,%)24,966 (49.8%)21,311 (47.3%)Race White (N,%) Black (N,%) Other (N,%)36530 (72.9%)9,366 (18.7%)4,222 (8.4%)33,040 (73.4%)8,808 (19.6%)3,188 (7.1%)Median Elixhauser Score (IQR)12 (4-22)11 (3-19)Hospital-Onset Sepsis6,469 (12.9%)5,065 (11.3%)Organ Dysfunction Vasopressors (N,%) Mechanical Ventilation (N,%) Creatinine/eGFR (N,%) Lactateb (N,%) Bilirubin (N,%) Platelets (N,%)15,726 (31.4%)11,277 (22.5%)18,155 (36.2%)23,485 (46.9%)4,501 (9.0%)6,009 (12.0%)15,137 (33.6%)9,905 (22.0%)18,557 (41.2%)14,775 (32.8%)4,766 (10.6%)5,296 (11.8%)Positive Blood Culture b9,755 (19.5%)10,272 (22.8%)ICU Admission b22,692 (45.3%)21,182 (47.0%)Median Hospital LOS (IQR)9 (6-15)9 (6-15)In-Hospital Death (N,%)9,758 (19.5%)8,118 (18.0%)Study Period 2009-2011 2012-201510,249 (20.5%)39,869 (79.6%)36,438 (80.9%)8,598 (19.1%)aSex was missing in 10 patients.bThe higher rate of lactate dysfunction in patients with missing vital signs likely reflects the fact that missing vital signs were clustered in earlier years of the study (lactate testing rates are increasing over time).Supplemental Table 5. Characteristics of Risk-Adjustment Models for In-Hospital Mortality in Primary (Cerner) and External Validation (HCA Healthcare) DatasetsModel Components and CharacteristicsBasic Admin. ModelAdvanced Admin. ModelaClinical Model 1 (ASE Criteria)Clinical Model 2(Extended Labs)Clinical Model 3(Maximal Clinical)Predictor Category Demographics ????? Comorbidities????? Infection Site ???? Days to Sepsis Onset???? ICU at Sepsis Onset???? Coded Severity-of-Illness Markers? Adult Sepsis Event Organ Dysfunction Criteriab??? Extended Labs?? Vital Signs and GCS?Model Characteristics, Imputing Normal Values for Missing Covariatesc AUROC Internal Validation (95% CI)0.726[0.719-0.733]0.776 (0.770, 0.783)0.817 [0.811-0.823]0.826 [0.820-0.831]0.830[0.826-0.834]d AUROC External Validation (95% CI)0.710 [0.707-0.713]0.771 [0.768, 0.773]0.819 [0.816-0.821]0.827 [0.824-0.829]N/A H-L statistic (Internal Validation)10.611.110.310.110.3 p-value H-L statistic (Internal Validation)0.3280.2990.3510.3520.341 Adjusted Brier Score (Internal Validation) 10.0%14.5%24.0%25.4%26.2% Standardized Mortality Ratio (Internal Validation)0.9960.9970.9980.9970.998 Standardized Mortality Ratio (External Validation)0.8470.6310.8170.772N/AModel Characteristics, Multiple Imputation for Missing Covariatese AUROC Internal Validation (95% CI)--0.814 [0.808-0.820]0.826 [0.820-0.832]0.831 [0.825-0.837] H-L statistic (Internal Validation)--10.29.910.0 p-value H-L statistic (Internal Validation)--0.3510.3640.360 Adjusted Brier Score (Internal Validation) --23.2%25.4%26.3% Standardized Mortality Ratio (Internal Validation)--0.9980.9970.999Abbreviations: Admin = administrative, ASE = Adult Sepsis Event, ICU = intensive care unit, GCS = Glasgow Coma Scale, AUROC = area under the receiver operating curve, H-L = Hosmer-Lemeshow.a The “Advanced Administrative Model” refers to a previously published model by Ford et al. (Crit Care Med 2016, 44(2):319-327) based entirely on administrative data, including demographics (age, sex, race), comorbidities, and coded markers of severity-of-illness on admission (early/late mechanical ventilation, shock, hemodialysis, ICU care). b Both numerical values and binary yes/no Adult Sepsis Event criteria for organ dysfunctions were included in the clinical models.c The training set included two-thirds of the Cerner HealthFacts dataset, while the internal validation set included the remaining one-third (n=31,718). The external validation dataset was the HCA Healthcare dataset (n=201,997). d The results reported for clinical model 3 in the table are based on development and internal validation in Cerner imputing normal values for missing vital signs and GCS data. Due to the high amount of missing vital sign data, we also developed and internally validated this model in Cerner in patients with non-missing vital signs (n=16,714 in validation cohort) using normal value imputation for the remainder of the covariates. The AUROC of this model was 0.840 (95% CI 0.833-0.848), as reported in the main manuscript. Vital signs and GCS measurements were unavailable in the HCA Healthcare dataset and so clinical model 3 was not applied.d Multiple imputation was not performed on the HCA Healthcare dataset. Supplemental Table 6. Sepsis In-Hospital Mortality Prediction Model Results Based on Administrative Data and Adult Sepsis Event Organ Dysfunction Criteria (Clinical Model 1)CovariateEstimateStandard Errorp-valueOdds Ratio95% CI for Odds Ratio(Intercept)-4.4390.164<0.0010.010.01-0.02Age0.0230.001<0.0011.021.02-1.03Race (Referent = White) Asian0.0130.1940.9461.010.69-1.48 Black0.0660.0840.4331.070.91-1.26 Hispanic0.2170.1740.2111.240.88-1.75 Other -0.1360.0920.1400.870.73-1.05 Unknown0.2460.1560.1151.280.94-1.74Sex (Referent = Male/Unknown) Female Sex0.1550.029<0.0011.671.10-1.24Comorbiditiesa Congestive Heart Failure0.2300.048<0.0011.261.15-1.38 Pulmonary Circulatory Disease0.4320.055<0.0011.541.38-1.72 Peripheral Vascular Disease0.3270.042<0.0011.391.28-1.51 Paraplegia0.1560.0720.0301.171.02-1.35 Neurologic Disease0.2880.054<0.0011.331.20-1.48 Chronic Lung Disease0.1060.0530.0441.111.00-1.23 Renal Failure0.1720.0530.0011.191.07-1.32 Liver Disease0.5280.066<0.0011.701.49-1.93 Acquired Immunodeficiency Syndrome0.6260.180<0.0011.871.32-2.66 Lymphoma0.4240.090<0.0011.531.28-1.82 Metastatic Solid Tumor1.050.063<0.0012.862.53-3.24 Solid Tumor without Metastases0.4820.068<0.0011.621.42-1.85 Arthritis0.1900.0650.0041.211.06-1.37 Coagulopathy0.2590.055<0.0011.301.16-1.44 Weight Loss0.3130.053<0.0011.371.23-1.52 Fluid and Electrolyte Disorders0.3410.037<0.0011.411.31-1.51 Cardiac Arrhythmia0.3760.041<0.0011.461.34-1.58 Leukemia0.6510.097<0.0011.921.59-2.32 Stem Cell Transplant0.5950.155<0.0011.811.34-2.46 Square Root of Number of Comorbidities-0.5780.092<0.0010.5610.47-0.67Admission from Healthcare Facility0.2660.067<0.0011.301.15-1.49Infectious Diagnosesb Urinary Tract Infection-0.5400.040<0.0010.580.54-0.63 Intra-abdominal Infection-0.2320.036<0.0010.790.74-0.85 Skin/Soft Tissue Infection-0.4780.050<0.0010.620.56-0.68 Septicemia/Bacteremia0.3250.064<0.0011.391.22-1.57 2 or More Infections0.2000.052<0.0011.221.10-1.35Days Until Sepsis Onsetc(Referent = 0-2 Days) 3-7 Days0.6070.0491.151.841.67-2.02 >7 days1.070.058<0.0012.932.62-3.28CDC Organ Dysfunction Criteriad CDC Vasopressor Dysfunction0.5760.071<0.0011.781.55-2.05 CDC Creatinine Dysfunction-0.630.064<0.0010.530.47-0.60 CDC Bilirubin Dysfunction-0.6110.056<0.0010.540.49-0.61 CDC Platelet Dysfunction-0.2730.056<0.0010.760.68-0.85 CDC Mechanical Ventilation0.8880.062<0.0012.432.15-2.74 Square Root of Number of CDC Dysfunctional Organs0.4850.105<0.0011.621.32-1.99Physiology at Sepsis Onsete Number of Vasopressors0.1830.029<0.0011.211.13-1.27 Peak Lactate0.1350.008<0.0011.141.13-1.16 Peak Lactate squared Peak Creatinine0.2450.042<0.0011.281.18-1.39 Peak Creatinine squared-0.0170.004<0.0010.980.98-0.99 Peak Bilirubin0.1310.009<0.0011.141.12-1.16 Peak Bilirubin squared-0.0020.003<0.0011.001.00-1.00 Minimum Platelet Count-0.0020.003<0.0011.001.00-1.00 Minimum Platelet Count squared<0.001<0.001<0.0011.001.00-1.00a All comorbidities are based on the Elixhauser comorbidity index, except for leukemia.b Infectious syndromes were identified using ICD-9-CM discharge diagnosis codes: Pneumonia (480.0-480.9, 481, 482.0-482.9, 483.0-483.8, 484.1-484.8, 485, 486), Urinary (590.00, 590.01, 590.10, 590.11, 590.2, 590.3, 590.80, 590.81, 590.9, 595.0, 595.2, 595.3, 595.4, 595.89, 595.9, 597.0, 597.80, 597.89, 598.00, 598.01, 599.0), Intra-abdominal (008.45, 009.0–009.3, 540.0–540.9, 541, 542, 543.9, 562.01, 562.03, 562.11, 562.13, 567.0–567.9, 569.5, 569.61, 569.71, 569.83, 572.0–572.8, 574.00–574.91,575.0–575.9, 576.0–576.9, 614.0–614.9), Skin/Soft Tissue (680-686, 035, 376.01, 728.86), Septicemia/Bacteremia (038.0-038.9, 790.7), Central Nervous System (027.0, 036, 320.0-321.1, 321.8, 324.0), Obstetric/Gynecologic (614.0-614.5, 616.0-616.1, 616.3-616.4, 634.0, 635.0, 636.0, 637.0, 638.0, 639.0, 646.5, 646.6, 647.9, , 658.4, 659.3, 670, 675). c The day of sepsis onset is defined by the first blood culture order or the first qualifying antibiotic day in the Adult Sepsis Event window period, whichever comes first. d CDC Organ Dysfunction Criteria refers to whether Adult Sepsis Event organ dysfunction criteria were met during the sepsis event window (+/-2 calendar days from the blood culture day). e Physiology at sepsis onset refers to the worst values during a +/-1 day window around the day of sepsis onset.Supplemental Table 7. Sepsis In-Hospital Mortality Prediction Model Results Based on Administrative Data, Adult Sepsis Event Organ Dysfunction Criteria, and Extended Laboratory Data (Clinical Model 2)CovariateEstimateStandard Errorp-valueOdds Ratio95% CI for Odds Ratio(Intercept)2.8022.2980.22316.50.18-1489.6Age0.0230.001<0.011.021.02-1.03Race (Referent = White) Asian0.0380.1960.8451.040.71-1.53 Black0.0790.0720.2731.080.94-1.25 Hispanic0.2080.1600.1941.230.90-1.68 Other -0.1160.0920.2040.890.74-1.07 Unknown0.1710.1610.2901.190.86-1.63Sex (Referent = Male/Unknown) Female Sex0.1130.029<0.0011.121.06-1.18Comorbidities Congestive Heart Failure0.2630.048<0.0011.301.18-1.43 Pulmonary Circulatory Disease0.4470.057<0.0011.561.40-1.75 Peripheral Vascular Disease0.3110.041<0.0011.361.26-1.48 Paraplegia0.1380.0700.0471.151.00-1.32 Neurologic Disease0.2710.052<0.0011.311.19-1.45 Chronic Lung Disease0.1290.0510.0111.141.03-1.26 Renal Failure0.2240.050<0.0011.251.13-1.38 Liver Disease0.5290.056<0.0011.701.52-1.89 Acquired Immunodeficiency Syndrome0.6240.172<0.0011.871.33-2.61 Lymphoma0.4660.089<0.0011.591.34-1.90 Metastatic Solid Cancer1.0190.066<0.0012.772.43-3.15 Solid Tumor Without Metastases0.4850.065<0.0011.621.43-1.85 Arthritis0.1690.0690.0141.181.03-1.36 Coagulopathy0.2420.051<0.0011.271.15-1.41 Weight Loss0.2350.053<0.0011.271.14-1.40 Fluid and Electrolyte Disorders0.2900.038<0.0011.341.24-1.44 Blood Loss-0.2840.1240.0220.750.59-0.96 Anemia Deficiency-0.1260.0410.0020.880.81-0.96 Cardiac Arrhythmia0.3680.049<0.0011.441.34-1.56 Leukemia0.6320.111<0.0011.881.51-2.34 Stem Cell Transplant0.7150.148<0.0012.041.53-2.73 Square Root of Number of Comorbidities-0.5760.075<0.0010.560.49-0.65Admission from Healthcare Facility0.2390.067<0.0011.271.11-1.45Infectious Diagnoses Pneumonia Urinary Tract Infection-0.5390.040<0.0010.580.54-0.63 Intra-abdominal Infection-0.2580.034<0.0010.770.72-0.83 Skin/Soft Tissue Infection-0.4780.050<0.0010.620.56-0.68 Septicemia/Bacteremia0.3060.060<0.0011.361.21-1.53 2 or More Infections0.1910.048<0.0011.211.10-1.33Days Until Sepsis Onset (Referent = 0-2 Days) 3-7 Days0.6380.048<0.0011.891.72-2.08 >7 days1.0880.057<0.0012.972.66-3.32CDC Organ Dysfunction Criteria CDC Vasopressor Dysfunction0.5050.067<0.0011.661.45-1.89 CDC Creatinine Dysfunction-0.7190.056<0.0010.490.44-0.54 CDC Bilirubin Dysfunction-0.6840.054<0.0010.500.45-0.56 CDC Platelet Dysfunction-0.3630.048<0.0010.700.63-0.76 CDC Mechanical Ventilation0.8060.063<0.0012.241.98-2.53 Square Root of Number of CDC Dysfunctional Organs0.6060.102<0.0011.831.50-2.24Physiology at Sepsis Onset Number of Vasopressors0.1660.030<0.0011.181.11-1.25 Peak Lactate0.0530.0160.0011.051.02-1.09 Peak Lactate squared0.0030.0010.0081.001.00-1.00 Peak Creatinine0.1120.0340.0011.121.05-1.20 Peak Creatinine squared-0.0090.0030.0020.990.98-1.00 Peak Bilirubin0.0930.010<0.0011.101.08-1.12 Peak Bilirubin squared-0.001<0.001<00011.001.00-1.00 Minimum Platelet Count-0.002<0.001<0.0011.001.00-1.00 Minimum Platelet Count squared<0.001<0.001<0.0011.001.00-1.00 Peak White Blood Cell Count0.0070.002<0.0011.011.00-1.01 Peak White Blood Cell Count squared<0.001<0.001<0.0011.001.00-1.00 Peak Aspartate Aminotransferase<0.001<0.001<0.0011.001.00-1.00 Peak Aspartate Aminotransferase Squared<0.001<0.001<0.0011.001.00-1.00 Peak International Normalized Ratio0.1010.021<0.0011.111.06-1.15 Peak International Normalized Ratio squared-0.0050.0010.0011.000.99-1.00 Peak Anion Gap0.0790.015<0.0011.081.05-1.12 Peak Anion Gap squared-0.001<0.00<0.0011.001.00-1.00 Minimum Albumin-1.3210.122<0.0010.270.21-0.34 Minimum Sodium-0.0900.0320.0040.910.86-0.87 Minimum Sodium squared<0.001<0.001<0.0011.001.00-1.00Supplemental Table 8. Mean/Standard Deviation and Median/Interquartile Ranges for Physiologic CovariatesCerner HealthFactsHCA HealthcareCovariateUnitMean (SD)Median (IQR)Mean (SD)Median (IQR)Creatininemg/dL2.0 (1.9)1.4 (1.4)2.0 (1.8)1.4 (1.2)Total Bilirubinmg/dL1.4 (2.6)0.7 (0.8)1.2 (2.2)0.7 (0.7)Platelet CountK/?L232 (125)205 (144)227 (124)210 (142)Lactate mmol/L3.3 (3.1)2.4 (2.1)3.3 (2.9)2.4 (2.0)WBCK/?L14.7 (11.7)12.9 (9.6)14.7 (11.1)13.0 (9.6)Albuming/dL3.1 (0.8)3.1 (1.1)2.9 (0.7)2.9 (1.0)Anion GapmEq/L12.5 (5.5)12 (6)13.4 (5.5)13.0 (6.0)Hematocrit%34.2 (7.5)34.2 (10.3)35.2 (7.4)35.3 (10.1)ASTUnits/L134 (650)34 (46)129 (671)32 (44)SodiummEq/L137 (6.5)137 (7.0)137 (6.5)137 (7.0)INR-1.7 (1.5)1.2 (0.5)1.5 (1.1)1.2 (0.4)SBPmmHg96 (24.6)95 (31)UnavailableUnavailableTemperature? Farenheight99.7 (2)99.3 (2.6)UnavailableUnavailableRespiratory RateBreaths/minute27.9 (11.9)24 (12)UnavailableUnavailableGCS-12 (4)14 (6)UnavailableUnavailableAbbreviations: WBC = white blood cell count, AST = asparate aminotransferase, INR = international normalized ratio, SBP = systolic blood pressure, GCS = Glasgow Coma Scale. For each patient encounter, the most extreme value was taken on the day of sepsis onset. The values shown above represent the mean/medians among patients with non-missing covariates. Supplemental Figure 1. Proportion of encounters missing physiologic covariates within 1 day of sepsis onset in Cerner and HCA Healthcare datasetsNote: Vital signs and Glasgow Coma Scores were unavailable in the HCA Healthcare datasets. Missingness of vital signs in the Cerner dataset was much lower in years 2012-2015 vs 2009-2011. Supplemental Figure 2. Variable importance plot demonstrating change in model performance by addition or subtraction of covariates in clinical model 1. “Variable importance” was measured by examining the impact of omitting each variable from the model on prediction performance. For example, suppose that the training?AUROC?for clinical model 1 is A1. After removing age from the model but retaining all other variables, the "reduced model" is fit again to the training?dataset?to obtain another AUROC value (A2). The difference (A1-A2) is displayed on the y axis. A large difference indicates that the variable is “important” in that removing it leads to a large decrease in prediction accuracy.?Variables are ordered in descending order according to the rankings of the differences (A1-A2); the top 10 are plotted. Supplemental Figure 3. Calibration Plot of Clinical Models Using (A) Multiple Imputation vs (B) Normal Value Imputation for Missing Covariates in External Validation Data (HCA Healthcare)Clinical model 1 includes administrative data and basic clinical variables needed to compute Adult Sepsis Event criteria. Clinical model 2 includes the same but adds extended laboratory data. ................
................

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

Google Online Preview   Download