Abstract T MP87: Comparison of Medicare Claims vs. Physician Adjudication for Identifying Acute Strokes in the Women's Health Initiative
Background: Many clinical trials and observational studies use medical record review for ascertaining outcomes. One large, long-running study, the Women’s Health Initiative (WHI) ascertains stroke events using participant self-report and subsequent physician review of medical records. This is resource-intensive.
Purpose: To assess whether Medicare data can reliably ascertain stroke events in the WHI.
Methods: Subjects were WHI participants with fee-for-service Medicare coverage. Stroke events from the start of WHI in 1993 through 2007 were included. Discharge diagnoses in hospitalization claims were used to create 4 stroke definitions for Medicare data. Definition 1: stroke codes in any position; Definition 2: primary position stroke codes; Definitions 3: hemorrhagic stroke codes; Definition 4: ischemic stroke codes. WHI data were randomly split into training (50%) and test sets. A concordance matrix was used to examine agreement between WHI and Medicare stroke diagnosis. Algorithm performance was optimized on the training data set. A WHI stroke and a Medicare stroke were considered a match if they occurred within +/- 7 days of each other. We excluded cases where medical records were unavailable and those without a reported WHI event within 7 days of Medicare stroke (n=239).
Results: Training data (n=24,428): There were 577 WHI strokes and 557 Medicare strokes using definition 1. Of these, 478 were a match as defined above. Algorithm performance: Specificity (SP) 99.7%; Negative Predictive Value (NPV) 99.7%; Sensitivity (SN) 82.8%; Positive Predictive Value (PPV) 85.8%; kappa 0.84. Test data set (n=24,422, see Table: Definition 1, algorithm performance was SP 99.7%, NPV 99.7, SN 82.0, PPV 84.6, kappa 0.83. Results were similar for all 4 stroke definitions.
Conclusion: Our study provides insights about the performance characteristics of Medicare data that can facilitate appropriate use of these vast data sources for outcome ascertainment in large studies.
Author Disclosures: K. Lakshminarayan: Research Grant; Modest; WHI Subcontract. J. Larson: None. B. Virnig: None. D. Burwen: None.
- © 2014 by American Heart Association, Inc.