Can symptom-severity phenotypes identify depression risk after mild traumatic brain injury? A cluster-based approach

Mild traumatic brain injury (mTBI) affects millions worldwide and frequently leads to secondary depression. Early identification of high-risk individuals is critical for targeted mental-health screening in this population. Data-driven phenotyping offers a promising avenue to unmask hidden symptom patterns, but few studies have combined unsupervised clustering of post-concussion profiles with established clinical and psychosocial metrics. We aimed to classify post-concussion symptom-severity profiles in adults with mTBI and to evaluate their association with secondary depression risk, adjusting for Glasgow Coma Scale (GCS) score, psychological resilience, age, sex, and time since injury.

In this cross-sectional analysis, 249 adults with mTBI (GCS 13-15) were recruited from a tertiary hospital in northern Taiwan. We performed hierarchical clustering using Ward's method with Euclidean distance (with BIC support) to derive three symptom-severity phenotypes from the Rivermead Post-Concussion Questionnaire items, then used k-means clustering to assign individuals to these by minizing within-cluster variance. Depression, defined as a Beck Depression Inventory-II ≥ 13, was modeled as an outcome in generalized linear models, adjusting for GCS and psychological resilience. Model discrimination was evaluated via area under the receiver operating characteristic curve (AUC).

Three distinct symptom clusters (mild, moderate, severe) were identified. The severe cluster was characterized by prominent visual symptoms, including light sensitivity and double vision. Compared with the mild cluster, the moderate cluster had 5.06-fold higher depression odds (95% CI [2.08-12.31]; p < .001) and the severe cluster 17.17-fold higher odds (95% CI [5.66-52.14]; p < .001). Higher resilience was independently protective (OR = 0.95, 95% CI [0.93-0.96]; p < .001), as was each additional GCS score (OR = 0.20, 95% CI [0.06-0.62]; p = .005). The full model showed excellent discrimination with an AUC of 88%, 95% CI [0.83-0.92].

Our data-driven approach shows that distinct post-concussion symptom-severity phenotypes, when integrated with GCS and resilience metrics, yields a robust tool for identifying mTBI survivors at high risks of depression. These findings support early, targeted mental-health screening and lay the groundwork for prospective validation and personalized intervention strategies.

NCT04243226. Registered on January 20. 2020.