Potential modification by climate factors of the association between air pollution and dynapenic abdominal obesity across CKM stages 0–3: a machine learning-based cohort study

Whether long-term ambient air pollution is associated with incident dynapenic abdominal obesity (DAO), and whether climate conditions modify these associations across cardiovascular-kidney-metabolic (CKM) stages, remains unclear. This cohort study included 7,919 adults aged ≥ 45 years from the China Health and Retirement Longitudinal Study. Missing covariates were handled using 20 chained-equation-imputed datasets. Separate single-pollutant logistic models were fitted under minimal, primary confounder, and full prognostic adjustment, with the primary model governing inference. Principal component analysis assessed pollution-mixture effects, and 36 pollutant-climate interactions were tested with Benjamini-Hochberg correction. Nine ML algorithms were compared using repeated nested stratified cross-validation, and SHAP was used to interpret the final model overall and across CKM stages 0–3. During follow-up, 184 participants developed DAO. In the primary model, SO 2 was associated with incident DAO (OR per interquartile-range increase, 1.61; 95% CI, 1.20–2.15; q = 0.016) and was the only single-pollutant association retained after correction across 12 tests. The first pollution-mixture principal component explained 71.0% of standardized exposure variance and was positively associated with DAO (OR per SD increase, 1.10; 95% CI, 1.01–1.19). SO 4 2− -dryness and NH 4 + -dryness interactions remained significant after correction across 36 tests (both q = 0.049). XGBoost achieved the highest mean cross-validated PR-AUC (0.131), although sensitivity and PPV were limited. SHAP consistently identified O 3 , SO 2 , and CO as the leading pollutant contributors across CKM stages. SO 2 showed the most robust association with incident DAO. Pollution-mixture and dryness-related findings suggest additional combined environmental effects, while SHAP indicated a largely conserved pollutant-attribution hierarchy across CKM stages. These findings require validation in larger prospective cohorts.