Loading...
Loading...
Contributors:
Dataset URI:
10.3389/fpubh.2026.1966094.s001IntroductionPhysical activity is widely associated with better mental health, but observational estimates can be distorted by pre-existing health and socioeconomic differences, and it remains unclear whether potential benefits can be reliably individualized. We examined the association between regular physical activity and subsequent depressive symptoms among Chinese adults and evaluated whether treatment-effect heterogeneity was interpretable and predictively useful.MethodsWe used three waves of the China Family Panel Studies (CFPS; 2018, 2020, and 2022). Depressive symptoms, prior physical activity, and demographic, health, and socioeconomic characteristics measured in 2018 were treated as pre-exposure covariates. Regular physical activity was defined as participation at least once per week in 2020, and the 2022 Center for Epidemiologic Studies Depression Scale (CES-D) score was the outcome. The main analytic sample included 10,159 adults aged 18–64 years. Average effects were estimated using DoubleML-IRM with random-forest learners, household-clustered 5-fold cross-fitting, and 10 repeated sample splits. Causal forests were used to examine conditional treatment-effect heterogeneity and cross-fitted benefit ranking. Sensitivity analyses included alternative exposure definitions, common-support restriction, complete cases, and inverse-probability-of-observation weighting for attrition.ResultsThe revised household-clustered DoubleML estimate was −0.1833 CES-D points (SE = 0.0921, 95% CI: −0.3637 to −0.0028, p = 0.0465), indicating a small negative average association. The complete-case estimate was similar (−0.1693, SE = 0.0986, 95% CI: −0.3625 to 0.0239, p = 0.0859), and the attrition-weighted sensitivity estimate was −0.2088 (95% CI: −0.3783 to −0.0393, p = 0.0158). Causal-forest calibration suggested some treatment-effect heterogeneity, but benefit-oriented AUTOC and QINI were not statistically significant; socioeconomic gradients were unstable, and the larger estimated benefit among unmarried participants remained exploratory.DiscussionRegular physical activity was associated with a small subsequent reduction in depressive symptoms, but the magnitude was modest and the observational design does not support strong causal or clinical claims. Household-aware inference and attrition weighting did not materially change the direction of the estimate. Although heterogeneity signals were present, the available covariates did not support reliable individual-level prioritization.
Keyword(s):
causal forest; China family panel studies; depressive symptoms; double machine learning; heterogeneous treatment effects.
MeSH Term(s):
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)