Depression risk in older adults is a complex mental health issue influenced by multiple factors, including individual vulnerability and environmental exposure. Although many methods have been used to predict depression risk in this population, the causal relationships between these factors remain underexplored. This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS) and an environmental pollution dataset, comprising 2,086 older adults. Ten machine learning (ML) models were trained to predict depression risk, and SHapley Additive exPlanations (SHAP) were applied to assess variable importance. A Structural equation modeling (SEM)-based approach was then used to examine the causal pathways and mediating mechanisms linking air pollution variables and individual variables to depression risk. All ten ML models showed modest predictive performance, with area under the curve (AUC) values exceeding 0.6. SHAP analysis suggested pain, insomnia, and instrumental activities of daily living (IADL) as the most important individual variables, while ozone (O3) and sulfur dioxide (SO2) were identified as key air pollution variables. Analysis within the SEM-based approach indicated that O3 indirectly increased depression risk through IADL and pain, while SO2 exhibited both direct and indirect negative associations with depression. This study provides evidence supporting structural mechanisms linking environmental exposure and mental health. The combined ML and SEM-based approach was used to facilitate variable screening and pathway evaluation, and offers a complementary perspective for examining potential causal relationships. This integrative framework may provide a useful reference for future research on mental health and related interventions.
Bi et al. (Mon,) studied this question.