BACKGROUND: Standard outcome validation processes can differ between organizations and data sources, which challenges harmonization for multi-database studies. The data sources used in this study were: The HealthCore Integrated Research Database (HIRD), a US commercial medical insurance claims database; Medicare, a US federal medical insurance program mainly for individuals age 65 and older; and the Clinical Practice Research Datalink (CPRD), a UK primary care electronic medical records (MR) database.
OBJECTIVES: To describe the outcome validation process, source verification retrieval rate, and ability to determine a final case status from two ongoing multidatabase studies with focus on acute kidney injury (AKI) and breast cancer.
METHODS: Possible cases diagnosed in 2012-2018 were first identified using an electronic algorithm, and validation was conducted using pre-defined clinical case definitions and information from redacted MRs (HIRD), MR abstraction forms (Medicare), clinical patient profiles and general practitioner questionnaires (GPQx) (CPRD). To limit variability, the study team aimed to align the following validation components for each outcome across the data sources: uniform adjudication trainings, designation of adjudication committee members, and case definitions.
RESULTS: The overall retrieval rate for MRs was 50.4% (HIRD), 56.4% (Medicare), and 64.4% for GPQx (CPRD). The adjudication committee assigned a definitive case status for 79% (HIRD), 80% (Medicare), and 63% (CPRD) of AKI cases and 98% (HIRD), 86% (Medicare), and 90% (CPRD) of breast cancer cases. The proportion of reviewed cases with insufficient information in the available records for assigning a definitive case status (and/or due to lack of response from GPQx in CPRD): 21% (HIRD), 20% (Medicare), and 38% (CPRD) for AKI (case definition required information on laboratory values) and 2% (HIRD), 14% (Medicare), 10% (CPRD) for breast cancer (case definition required evidence of diagnosis and either a cancer-specific therapy or encounter).
CONCLUSIONS: A higher proportion of reviewed breast cancer cases were assigned a final case status compared to AKI cases. The AKI case definition required laboratory values which were often incomplete or missing from available records. Although the adjudication process in this study was aligned across data sources to the extent possible, with similar results in the two US claims-based data sources, some reviewed cases were not adjudicated. Harmonization of validation approaches is important to minimize potential variation in safety study results.
