Comparative analysis of cardiometabolic multimorbidity predictors in China and the USA: A machine learning approach

Background
Cardiometabolic multimorbidity (CMM) – coexisting cardiovascular (CVD) and metabolic diseases (MD) – represents a major public health challenge in the USA and China, but early screening tools remain inadequate and lack cross-national applicability. We aim to build bidirectional model to determine risk factors in different countries.
Methods
We utilised data from CHARLS (China, n = 3,401 CVD/n = 797 MD) and HRS (USA, n = 3,533 CVD/n = 1,507 MD) to develop bidirectional machine learning (ML) prediction models including logistic regression (LR), gaussian naive Bayes (GNB), and extreme gradient boosting (XGBoost), evaluating with AUC and validating across nations. SHAP analysis identified consistent and varied risk predictors.
Results
LR (AUC = 0.70 in both countries) were best in CVD prediction, LR (AUC = 0.71 in China) and GNB (AUC = 0.65 in USA) were best in MD prediction. Model performance decreased during external validation. Core predictors included disease counts and moderate physical activity. Priority predictors differed: pulse, mild physical activity, emotional problems and falling in USA; full-tandem stance and left hand’s grip strength in China.
Conclusions
Our bidirectional, cross-nationally validated ML model enables precision CMM screening. Region-specific strategies are advised: China should prioritize grip strength and balance screening, while the USA focuses on pulse monitoring and psychological interventions.