Objective To investigate the individualized causal effects of regular exercise on changes in depressive mood among adolescents during the “behavior initiation” and “behavior maintenance” stages, identify the subgroups that benefit most, and provide an evidence base for precision school-based exercise interventions.

Methods This study leveraged two-year longitudinal data from 321626 adolescents, who were classified at baseline into an exercise-initiation cohort and an exercise-maintenance cohort. We applied the causal forest algorithm under the generalized random forest framework, incorporating Robinson orthogonalization to adjust for 13 high-dimensional confounders. To ensure common support overlap, we performed optimal sample trimming. Average treatment effects were estimated using doubly robust methods, and treatment effect heterogeneity was examined via feature importance analysis and best linear projection.

Results Doubly robust estimates showed that exercise had a significant antidepressant effect across both behavioral stages. The average treatment effects were －1.112 (95% CI: －1.171 to －1.053) for the initiation cohort and －1.097 (95% CI: －1.173 to －1.021) for the maintenance cohort. The conditional average treatment effects (CATE) varied widely across individuals, with 95% percentile ranges of －2.52, －0.15 and －2.20, －0.10, respectively. In the initiation cohort, effect heterogeneity was mainly attributable to baseline emotional symptoms, with depression and anxiety importance values of 0.593 and 0.151, respectively. In the maintenance cohort, baseline depression remained the dominant predictor (importance: 0.383), and the treatment effect was larger among individuals with poor family functioning (β = －0.406, 95% CI: －0.786 to －0.026).

Conclusion After adjusting for baseline covariates, both initiating and maintaining daily exercise of more than 30 minutes were associated with reduced adolescent depressive symptoms. The beneficial effects were more pronounced among individuals with greater baseline emotional symptom severity and those with poor family functioning, providing a basis for tiered or targeted delivery of school-based physical activity interventions.