Heterogeneous developmental trajectories and predictors of self-harm in early adolescence: A machine learning study.

BACKGROUND: Although prior studies have examined developmental trajectories of adolescent self-harm in terms of frequency and severity, a fundamental gap remains: we lack a dynamic, state-specific understanding of how self-harm ideation and behavior develop during early adolescence. Moreover, few studies have prospectively examined how baseline predictors relate to subsequent self-harm pathways. METHODS: Using four annual waves of data from a large-scale Chinese adolescent cohort (N = 11,366; 48.6% female; T1: Mage = 10.72 ± 0.29 years), this study used a person-centered approach to delineate distinct self-harm trajectories and interpretable machine learning methods to identify their baseline predictors. RESULTS: Five heterogeneous trajectories were identified: persistently low-risk, persistent ideation, ideation remission, behavior-to-ideation, and ideation-to-behavior. Although depressive symptoms, self-blame, family stress, and gender emerged as the most influential predictors overall, the predictors varied substantially across trajectories, indicating meaningful differences in their developmental drivers. CONCLUSIONS: These findings demonstrate the heterogeneous developmental trajectories of self-harm states in early adolescents and reveal trajectory-specific risk predictors, underscoring the importance of prevention efforts that should consider both shared and distinct factors across pathways.