Estimating Repeat Traumatic Brain Injury in the U.S. Military, 2015-2017

Abstract

ABSTRACT Introduction Traumatic brain injury (TBI) is a significant health issue among deployed and non-deployed U.S. military service members (SMs). Since 2000, an estimated 413,858 SMs have been diagnosed with at least one TBI. Due to the difficulty in distinguishing new incident TBIs from follow-up TBI-related medical encounters in the Military Health System (MHS), the official TBI case definition also includes an incidence rule considering an individual an incident case only once per lifetime. We sought to examine patterns in medical records of SMs with at least one TBI encounter, in an effort to identify repeat TBIs in individual SMs and to estimate the incidence of repeat TBIs within the study cohort as a whole. Materials and Methods Using the official DoD TBI case definition, we obtained a list of SMs who sustained their first active duty TBI between October 1, 2015, and September 30, 2017. We identified the SM’s diagnosing encounter (index TBI). Subsequently, we identified patterns associated with diagnosing medical encounters, as opposed to encounters associated with follow-up TBI care. We flagged external cause of injury records and the presence of TBI-related symptom codes at the diagnosing encounter. Traumatic brain injury–related symptoms included memory issues, alteration of cognition, hearing loss, vertigo, headache, anxiety, depression, emotional lability, weakness, insomnia, and vision disturbance. Data discovery results were shared with a group of clinicians at the Defense and Veterans Brain Injury Center, and the list of variables was further refined based on clinical expertise. Subsequently, we conducted stepwise logistic regression, and best fitting model was used to create a probability score to be applied to all TBI-related medical encounters. To validate the accuracy of the model-derived probability score, a stratified random sample of medical records was reviewed by trained clinician. At the 0.5 probability cutoff point, the model had an area under the curve of 0.69. We applied the final model portability scores to all identified TBI encounters to estimate the incidence of repeat TBI within the cohort. Results Between October 1, 2015, and September 30, 2017, we identified 36,440 SMs and their first lifetime TBI encounter. Study follow-up period was 2 years. Predictors of repeat TBI (rTBI) encounters included the presence of TBI diagnosis extender codes “A” (odds ratio [OR] = 4.67, 95% CI 2.15-10.12); W and V series codes (OR = 4.05, 95% CI 2.05-7.95 and OR = 2.86, 95% CI 1.40-5.83, respectively); patient’s disposition at home/quarters; and admission or immediate referral (OR = 3.67, 95% CI 1.79-7.51). Number of diagnosis codes in patient’s medical record was inversely associated with a repeat TBI encounter (OR = 0.84, 95% CI 0.76-0.96). Applying model-derived probability score onto identified medical records, we estimate that 804 unique SMs sustained an rTBI during the follow-up period, yielding a rate of 260 rTBIs per 10,000 person-years or approximately 2.32% of SMs annually. Conclusion Probability scores based on statistical modeling can provide reasonable estimates of repeat incidences of TBI using medical billing data when formerly only the first TBI was thought to be measurable. With 100% sensitivity and 69% specificity, application of these models can inform estimates of repeat TBI across the MHS. This effort shows initial success if estimating repeat TBI, and further modeling work is encouraged to increase the predictive characteristics of the models as these efforts show promise in estimating repeat TBI across the MHS.