Comparative impact of cardiometabolic multimorbidity clusters and distinct obesity indicators on cardiovascular disease in chinese middle-aged and older adult

Background The escalating burden of cardiovascular disease (CVD) in China's aging population needs refined risk stratification[1, 2]. While individual cardiometabolic risk factors are established predictors, their clustering into cardiometabolic multimorbidity and the comparative prognostic value of abdominal versus general obesity within risk clusters remain inadequately characterized in this demographic. Purpose This study investigates how cardiometabolic multimorbidity is linked to incident CVD in a representative aging Chinese cohort to guide more precise primary prevention. Methods This prospective cohort study included 2,254 participants aged 45 years and older from the China Health and Retirement Longitudinal Study (CHARLS). We defined the exposures as the full spectrum of cardiometabolic risk factors (hypertension, dyslipidemia, diabetes, obesity), ranging from individual conditions to all possible binary, ternary, and the quaternary multimorbidity clusters. The association between each of these exposures and incident CVD was examined via multivariable logistic regression. Furthermore, sensitivity analyses were conducted to compare obesity defined by body mass index (BMI) versus waist circumference (WC). Subgroup analyses were performed to test the consistency of the observed associations across demographic and lifestyle strata. We compared 11 machine learning models to comprehensively evaluate predictive utility. Their discriminatory performance and clinical utility were assessed using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). Results A strong graded dose-response relationship was observed between the number of cardiometabolic risk factors and CVD risk. Specific clusters involving diabetes and obesity exhibited particularly high odds ratios (ORs). Sensitivity analysis revealed that the association for the diabetes-plus-obesity cluster was attenuated when using WC instead of BMI (OR=2.63, 95% CI: 1.62–4.18), whereas the hypertension-plus-obesity association was slightly strengthened (OR=2.49, 95% CI: 1.88–3.29), underscoring the distinct risk information provided by central adiposity. Subgroup analyses confirmed the robustness of these associations across most demographic strata. Analysis of 11 machine learning algorithms revealed that Random Forest showed superior generalization with the highest AUC (0.713), closely followed by Boosting method (AUC=0.690). DCA confirmed the clinical utility of Random Forest, Lasso, and Logistic model offering the highest net benefit across risk thresholds. Conclusion This study confirms a graded increase in CVD risk with cardiometabolic multimorbidity, particularly in clusters involving diabetes and obesity, and highlights the necessity of measuring central adiposity for precise risk assessment. The predictive power of machine learning models such as Random Forest can support targeted prevention of CVD in China's aging population.