Air quality-stratified interpretable machine learning for depression risk in middle-aged and older adults

Environmental Modeling & Assessment · Springer Nature · 2026-11-10 · DOI: 10.1007/s11869-026-02157-4

Аннотация

This study investigates depression-risk prediction among middle-aged and older adults across different air-quality contexts using data from the China Health and Retirement Longitudinal Study (CHARLS). Air quality, represented by annual average AQI, was stratified at the policy-relevant threshold of 100 defined by the Chinese AQI framework into AQI ≤ 100 and AQI > 100 groups. Machine learning models, including XGBoost, GBDT, GaussianNB, and Random Forest (RF), were developed to predict depression risk in each air quality group. The models’ performance was evaluated using AUC values, and SHAP analysis was used to characterize feature-attribution patterns within each stratum, and post-hoc interaction analyses were conducted to formally examine whether predictor–depression associations differed across air-quality strata. The models demonstrated good discriminatory power, with XGBoost and GBDT achieving the highest AUC values. SHAP analyses identified both shared and distinct feature-attribution patterns across the stratified models. Life satisfaction and memory ability were prominent predictors in both strata, whereas self-rated health and body pain received relatively greater attribution in the unhealthy air-quality model. However, the omnibus predictor-by-air-quality interaction was not statistically significant (P = 0.220), and no predictor-specific interaction remained significant after multiple-testing correction. Air-quality-stratified machine-learning models showed good predictive performance and revealed context-specific differences in model-derived feature attribution. Although formal interaction analyses did not identify robust statistical effect modification, the stratified SHAP findings provide complementary information on how depression-risk models may rely on different combinations of individual characteristics across environmental contexts.

Читать у издателя →

Похожие статьи