Predicting cardiovascular-kidney-metabolic multimorbidity in Chinese adults with overweight and obesity using machine learning: an internal evaluation

Xin Li, Zhitong Li, Yexin Ji, Yuanyuan Yan, Xiaoqin Chen, Linlin Gao, Baoling Wei, Chenxia Zhang, Ruixue Duan, Shiwei Liu

Aim: To develop and internally evaluate machine learning (ML) models for predicting incident cardiovascular-kidney-metabolic (CKM) multimorbidity in Chinese adults with overweight or obesity, and to identify key predictors. Methods: We included 4,244 participants from the China Health and Retirement Longitudinal Study (CHARLS) with overweight/obesity [body mass index (BMI) ≥ 24 kg/m<sup>2</sup>] and with zero or one CKM disease group at baseline (2015). CKM multimorbidity at follow-up (2018) was defined as coexistence of ≥ 2 disease groups (cardiovascular, kidney, metabolic). A two-stage feature selection [least absolute shrinkage and selection operator (LASSO) with bootstrap stability analysis] identified predictors from 56 sociodemographic, lifestyle, psychological, clinical, and environmental variables. Seven ML algorithms were compared; performance was assessed by area under the curve (AUC), calibration, Brier score, decision curve analysis, and SHapley Additive exPlanations (SHAP) interpretation for internal model evaluation. Results: During 3-year follow-up, 648 (15.3%) participants developed incident CKM multimorbidity. Nine predictors were selected: age, depression, pain, health expectation, weight change, dyslipidemia, hypertension, heart disease, and kidney disease. Artificial neural network (ANN) and logistic regression showed the best discrimination (AUC: 0.758 and 0.757) and acceptable calibration in internal testing (Brier score: 0.113 and 0.114). SHAP analysis identified hypertension, dyslipidemia, and depression as top contributors. A nomogram was developed for preliminary risk stratification, but external validation is required before any clinical application. Conclusion: In this internal evaluation, ANN and logistic regression showed moderate discrimination and acceptable calibration for predicting CKM multimorbidity in Chinese adults with overweight/obesity. Logistic regression performed comparably to complex algorithms yet is simpler; however, external validation is still required before use.