Brain age gap as a diffusion MRI-based marker of traumatic brain injury-related brain changes and associated outcomes.

Traumatic brain injury is a common neurological disorder and a leading cause of long-term disability, presenting with heterogeneous cognitive, emotional, and functional impairments. A critical clinical challenge is the early identification of patients at risk for persistent symptoms. The Brain Age Gap (BAG), the difference between an individual's predicted brain age from imaging data and their chronological age, has emerged as a potential marker of injury-related brain changes. Here, we aim to evaluate diffusion-MRI-derived BAG for identifying traumatic brain injury patients (Glasgow Coma Scale 13-15) at risk for persistent symptoms. For this, we trained a normative age-prediction model on >13 000 healthy controls, and applied it to traumatic brain injury patients. We associated BAG clinical scores that evaluate processing speed and executive function (Trail Making Test Parts A and B), verbal memory (Rey Auditory Verbal Learning Test), general processing speed (Wechsler Adult Intelligence Scale), post-concussion symptoms (Rivermead Post-Concussion Symptoms Questionnaire), psychological distress (Brief Symptom Inventory-18) and insomnia severity (Insomnia Severity Index). Analyses included: (i) cross-sectional comparisons across three BAG-based subgroups: BAG+, BAGn and BAG-, representing patients with higher, neutral and lower BAGs relative to healthy controls, (ii) longitudinal linear mixed-effects models evaluating BAG measured at 2-week post-injury (BAG2wk) in symptom trajectories and (iii) prognostic logistic regression for prespecified poor 12-month outcomes. All statistical analyses were performed using Python libraries, including SciPy and Statsmodels. The age-prediction model demonstrated high accuracy and reliability (mean absolute error = 3.05 ± 3.67 years; intraclass correlation = 0.93). Cross-sectionally, higher BAG was associated with greater symptom burden. Compared with BAG- patients, those in the BAG+ subgroup reported higher Insomnia Severity Index (d = 0.411; P = 0.038) and Rivermead Post-Concussion Symptoms Questionnaire scores (d = 0.415; P = 0.038). Similarly, BAGn patients showed higher Brief Symptom Inventory (d = -0.419; P = 0.040), Insomnia Severity Index (d = -0.382; P = 0.038) and Rivermead Post-Concussion Symptoms Questionnaire (d = -0.525; P = 0.002) scores relative to BAG-. Longitudinally, higher BAG2wk was associated with worse Rivermead Post-Concussion Symptoms Questionnaire (β = 0.195, 95% CI[0.093, 0.297]; partial R2 = 0.017; P = 0.0016) and worse Insomnia Severity Index (β = 0.107, 95% CI[0.033, 0.181]; partial R2 = 0.010; P = 0.0196) without time interaction. BAG2wk modified change over time in Trail Making Test Parts A (β = 0.017, 95% CI 0.002-0.056; partial R2 = 0.043; P = 0.016). Finally, adding BAG2wk into the prognostic model yielded a significant improvement in 12-month outcome prediction (likelihood-ratio test = 6.40; P = 0.011). Together, these findings indicate that diffusion-MRI-derived BAG is a robust and reproducible biomarker that captures clinically meaningful heterogeneity as early as two-week post-injury.