Accelerometer-derived movement phenotypes and adolescent mental health in pediatric obesity: A causal machine-learning study.

Background: Children with overweight or obesity are at elevated risk for later-life mental health challenges, but the role of accelerometer-derived waking movement behaviors remains uncertain. Thus, we examined longitudinal associations of sedentary behavior (SB), light-intensity physical activity (LPA), and moderate-to-vigorous-intensity physical activity (MVPA) at age 7 with later-life internalizing and externalizing problems. Method: We used data from the UK Millennium Cohort Study, focusing on children with overweight or obesity at age 7 who had valid accelerometer data and a complete set of predefined analytic variables (N = 858). Exposures were average daily minutes of SB, LPA, and MVPA. Outcomes were parent-reported Strengths and Difficulties Questionnaire-based internalizing and externalizing problems scores at ages 11 and 14. A generalized random forests approach was used to estimate average treatment effects (ATEs) and conditional average treatment effects (CATEs), adjusted for relevant demographic, socioeconomic, body mass index, parental distress, baseline mental health, and co-occurring movement intensities; day-of-week activity-pattern variables were used as exploratory candidate moderators. Results: Estimated ATEs were small. After applying the Benjamini-Hochberg false-discovery-rate (FDR) correction across the 12 primary exposure-outcome tests, only the association between higher SB and lower internalizing problems at age 11 remained statistically significant (ATE = -0.013 SDQ points per additional min/day; 95% CI, -0.019 to -0.007; q < 0.001). No other pathway survived FDR correction, and other nominal or directionally suggestive estimates were interpreted as exploratory. CATE summaries and calibration tests provided limited evidence of reproducible heterogeneity. Conclusions: Accelerometer-derived waking movement behaviors at age 7 were associated with later SDQ problems scores in small and outcome-specific ways among children with overweight or obesity. Since the study is based on observational data and relies on measured-confounder assumptions, the findings should be interpreted cautiously, and as model-based estimates rather than definitive causal evidence.