C-reactive protein-triglyceride-glucose index and incident mild cognitive impairment among patients with cardiovascular-kidney-metabolic syndrome: a dual-cohort prediction and double machine learning analysis.

PURPOSE: Develop an interpretable machine learning model and evaluate the predictive and potential causal effects of the C-reactive protein-triglyceride-glucose index (CTI) in medium- to long-term incident mild cognitive impairment (MCI) among patients with cardiovascular-kidney-metabolic (CKM) syndrome. METHODS: Models were developed and internally validated using China Health and Retirement Longitudinal Study (CHARLS) data and externally validated in the Health and Retirement Study (HRS). Participants with baseline CKM syndrome stages 1-3 were included. Following feature selection via Shapley additive explanations (SHAP) and recursive feature elimination with cross-validation, six machine learning algorithms were compared and assessed in terms of discrimination, calibration, and clinical utility, with SHAP further used for model interpretation. Double machine learning estimated the average treatment effect (ATE) of CTI and subgroup heterogeneity. RESULTS: 2658 CHARLS and 2796 h participants were included. The gradient boosting classifier, using 10 features including CTI, achieved the best performance (AUC 0.820; external 0.785). SHAP analysis identified CTI as a major predictor, second only to baseline cognitive function and educational attainment. Elevated CTI was associated with adverse estimated effects on incident MCI (ATE: 0.029 in CHARLS; 0.040 in HRS), markedly attenuated in individuals with greater cognitive reserve. CONCLUSIONS: CTI reliably predicts incident MCI in patients with CKM syndrome. The gradient boosting framework provides a robust risk stratification, and higher cognitive reserve may attenuate CTI-associated cognitive risk.