Predicting sub-clinical risk of complex PTSD in adolescents using multi-informant ecological data: A machine learning approach.

BACKGROUND: Complex Post-Traumatic Stress Disorder (CPTSD) is a significant mental health concern in adolescents, however traditional diagnostic tools often fail to capture early risk patterns. Adolescence is a unique developmental window marked by neurobiological remodeling and heightened interpersonal sensitivity. This study aims to move beyond binary clinical diagnosis toward a subclinical prevention framework by predicting continuous CPTSD symptom severity using multidimensional ecological data.

METHODS: The data were collected from a triad of 4177 Chinese adolescents and their parents. A two-layer stacking ensemble machine learning model was employed to integrate 118 variables across individual, family, and school domains. SHapley Additive exPlanations (SHAP) analysis was utilized to provide interpretability and identify the predictive contribution of key ecological factors.

RESULTS: The stacking ensemble model demonstrated higher predictive accuracy in identifying complex ecological risk patterns compared to standalone base learners. School climate functioned as an asymmetrical indicator where deterioration significantly predicted symptom elevation while a positive climate offered a limited protective association. Peer victimisation behaviors, maternal anxiety, and paternal coparenting conflict were identified as core components of the risk patterns.

CONCLUSION: An interpretable stacking ensemble model is developed to estimate concurrent CPTSD symptom severity in adolescents using multiple informant ecological data. Peer victimisation behaviors, maternal anxiety, and school climate are identified as the most informative indicators. A temporal inference is precluded by the cross sectional design. Therefore, this framework is viewed as an exploratory approach to recognizing adolescents with an elevated concurrent symptom burden.