Identifying traumatic brain injury across the lifespan in females from electronic health records

Statement of Purpose

Traumatic Brain Injury (TBI) affects 1.5–2 million people in the US annually, yet our understanding of its epidemiology is limited. Surveillance of athletes and combat veterans has resulted in potentially androcentric definitions of TBI. Electronic health records (EHR) remain under-utilized for understanding TBI in the community setting. US Centers for Disease Control (CDC) and Department of Defense/Veterans Affairs (DOD/VA) have developed TBI surveillance case definitions using ICD-9 and -10 codes. We sought to characterize females in the EHR with TBI across the lifespan, comparing and highlighting potential differences in case definitions.

Methods/Approach

We obtained data for 1,060,100 females with inpatient/outpatient Penn Medicine visits from 2010–2017. We used ICD-9/10 codes from CDC and DOD/VA definitions to identify patients with TBI, and classified TBI with the DOD/VA severity matrix (mild, moderate, severe/penetrating, unknown). We applied a decision-tree algorithm of TBI date to extract and apply most severe TBI diagnosis as severity across visits. We compared demographics among TBI-definition groups: 1) CDC-only, 2) DOD/VA-only, 3) CDC and DOD/VA, 4) Unclassified using Kruskall-Wallis and Chi-square tests. We conducted survival analysis (time to death, follow-up to 5 years) and inspected Kaplan-Meier curves with log-rank tests for differences among groups.

Results

Among 4466 patients, there were significant differences across TBI case definitions for each characteristic (age, race, injury mechanism, severity, death status, inpatient/outpatient encounter, all p<0.001), except Hispanic ethnicity (p=0.2). Survival time, on average, differed by source of case definition (p<0.0001). The largest proportion of patients surviving at five years met case definition DOD/VA_only (94.6%), followed by Unclassified (81.4%), then CDC definition-only (81.2%), then CDC and DOD/VA (72.4%).

Conclusion

Case definition can alter the cohort derived, so care must be taken extracting patients with TBI from EHRs. Future work will elucidate differences in case definitions, which may inform expanding current TBI case definitions.