Abstract Aims While cardiovascular–kidney–metabolic (CKM) syndrome has been recognised as a continuum of interconnected metabolic, renal, and cardiovascular dysfunction, practical risk stratification tools for cognitive outcomes in the early, modifiable stages of this syndrome remain lacking. This study aimed to develop an interpretable machine learning model to estimate long‐term risk of incident cognitive impairment among early‐stage CKM syndrome. Materials and Methods This prospective cohort study leveraged data from the China Health and Retirement Longitudinal Study, a nationally representative cohort of adults. A total of 4462 participants with CKM stages 0–2 at baseline (2011) and complete follow‐up through 2020 were included. Incident cognitive impairment was defined as cognitive performance >1 SD below the age‐adjusted mean. Candidate predictors were screened using the Boruta algorithm and LASSO regression, followed by multicollinearity assessment. Nine machine learning models were trained and validated, and their discrimination, calibration, and clinical utility were evaluated. Model interpretability was assessed using Shapley Additive Explanations (SHAP). Results During 10 years of follow‐up, 525 participants (11.8%) developed cognitive impairment. Ten predictors were retained, including education, total cholesterol, age, fasting blood glucose, uric acid, estimated glucose disposal rate, diastolic blood pressure, relative fat mass, CKM stage, and metabolic syndrome. The XGBoost model demonstrated the best performance (validation AUC = 0.726), strong calibration, and favourable decision curve benefit. SHAP analysis identified education, total cholesterol, and age as the top predictors. Conclusions Our interpretable machine learning framework, built upon readily accessible clinical and metabolic parameters, provides a robust and clinically applicable tool for predicting 10‐year risk of cognitive impairment among individuals with early‐stage CKM syndrome.
Duan et al. (2025) studied this question.