Longitudinal Trends of Depression in Traumatic Brain Injury: The Role of Individual Heterogeneity in Clinical Prediction

Background Depression affects approximately 30% of individuals after traumatic brain injury (TBI), yet long-term depression trends and their determinants are poorly understood. This study aimed to model depression trajectories over ten years post-TBI, compare the predictive performance of population-level only versus both population and subject-level effects, and assess the model's clinical utility for predicting depression in unseen and existing patients. Methods Data were obtained from the Traumatic Brain Injury Model System (TBIMS) National Data Bank. Depression was measured using the Patient Health Questionnaire-9 (PHQ-9) and collected at 1, 2, 5 and 10 years after injury. Covariates included age, sex, race, employment, education, functional measures, injury severity, pre-injury mental health, and substance use. Linear mixed modelling was used to identify depression trends and factors associated with depression. Predictive performance was evaluated using mean squared error, coverage, and precision. Results The sample comprised 19,397 individuals (mean age 43). Depression scores showed a small decrease over time among those with pre-injury mental health treatment history, but this change was not clinical meaningful. Significant predictors of Year 1 depression included pre-injury mental health treatment (β=1.6), female sex (β=0.86), and prior head injuries (β=0.4). When predicting depression for existing patients using early depression scores, the model achieved precision of 3.7 points, whereas for new patients, the model's precision was 6 points. Conditional predictions outperformed marginal predictions. Conclusion Depression trajectories following TBI exhibit substantial individual heterogeneity. Population-level models alone inadequately capture this complexity, while models incorporating both population and subject-level variations significantly improve predictive performance. This modeling approach demonstrates the potential for predicting depression trajectories in clinical settings, thereby facilitating individualized assessment and intervention.