A predictive model built from China Health and Retirement Longitudinal Study (CHARLS) data using LASSO regression, logistic modeling, and SHAP-interpreted machine learning distilled 46 candidate variables into six key disability predictors for elderly adults with cardiometabolic multimorbidity (CMM): depression, cognition, stroke history, CMM count, age, and falls. Notably, depression outperformed physical function markers—grip strength and walking speed—as a disability predictor. The CMM-disability relationship plateaued after three concurrent diseases, and cognition followed an inverted U-shaped curve with an inflection point at 12.043. Blood markers flagged by SHAP included HbA1c, creatinine, uric acid, hematocrit, fasting glucose, triglyceride-glucose index (TyG), and CVAI.
This finding carries real clinical weight: it repositions depression screening as a frontline disability-prevention tool in cardiometabolic care, arguably more actionable than grip dynamometry in resource-limited settings. The plateau effect at ≥3 diseases is practically meaningful—clinicians may need to shift focus from disease counting to functional and mental health management beyond that threshold. The mediation finding (grip strength explaining 12.3% of stroke's disability effect) hints at exercise-based intervention windows post-stroke. However, several limitations temper enthusiasm: the observational CHARLS dataset cannot establish causality, cross-sectional design limits temporal inference, and the Chinese elderly cohort restricts generalizability. Cognitive scoring nuances may also introduce measurement noise. Critically, this is a preprint posted to medRxiv and has not yet undergone peer review—the nomogram's clinical utility and reported effect sizes require independent validation before adoption.
