An evaluation and comparison of traumatic brain injury data sources within the VA Million Veteran Program: A...

With the increased use and availability of electronic health record (EHR) data, population-based traumatic brain injury (TBI) research has received growing attention. However, a key challenge emerging from this research has been accurate ascertainment of TBI case/control status. While International Classification of Diseases (ICD) diagnosis codes constitute a widely used source of data, their validity may vary by clinic and provider. Examining other TBI data sources from epidemiological databases may help ascertain the value of ICD codes. Within the VA Million Veteran Program (MVP), TBI data is available from several data sources. The purpose of this study was to provide an overview of TBI data sources available within MVP and to evaluate the performance characteristics of these variables. Analyses included up to 657,790 MVP-enrolled Veterans. TBI status (i.e., "TBI" vs. "No TBI") was examined using several different variables: two variables from the MVP Baseline Survey; one variable from the MVP Lifestyle Survey; TBI ICD codes; and two variables from the VA TBI Screening and Evaluation Program. Several key findings emerged: (1) the availability of TBI data varied across data sources; (2) there were significant differences in TBI prevalence by sex, ancestry, and military service era for all TBI variables; (3) sensitivity and specificity varied considerably across TBI variables; and (4) combining TBI data sources (e.g., defining a TBI case using ICD codes and survey data) resulted in variables with stronger performance characteristics. Findings speak to the power of combining multiple data sources when retrospectively assessing TBI in the EHR.