An interpretable prediction model for stage progression in early-stage cardiovascular-kidney-metabolic syndrome: a longitudinal cohort study

Cardiovascular-kidney-metabolic (CKM) syndrome spans metabolic, renal, and cardiovascular dysfunction, but tools for estimating progression in its early stages remain limited. We aimed to develop and internally evaluate an interpretable prediction model for 4-year progression in a CHARLS-adapted CKM classification and to examine stage-specific performance. This secondary analysis used prospectively collected data from the China Health and Retirement Longitudinal Study. We included 3773 participants aged 45 years or older with baseline CHARLS-adapted CKM Stages 0–2 and follow-up through 2015. Progression was defined as an increase of at least one CHARLS-adapted CKM stage. Six algorithms were tuned using stratified 10-fold cross-validation in the development set. Out-of-fold predictions were used to select and lock the final algorithm before evaluation in the held-out test set. SHAP values were calculated on the predicted-probability scale. During 4 years of follow-up, 1560 participants (41.3%) progressed. LASSO retained age, BMI, systolic blood pressure, HbA1c, eGFR, baseline CKM stage, sex, and education level. Logistic regression was selected using development-set out-of-fold discrimination and calibration criteria. In the held-out test set, the model achieved an AUC of 0.756, a Brier score of 0.191, a calibration intercept of 0.047, a calibration slope of 0.961, and an ICI of 0.019. Probability-scale SHAP analysis identified baseline CKM stage, HbA1c, BMI, and sex as the leading contributors to model predictions. Stage-specific AUCs were 0.631, 0.691, and 0.591 for CHARLS-adapted CKM Stages 0, 1, and 2, respectively, while calibration varied across stages. The internally evaluated logistic regression model showed moderate discrimination and reasonably close overall calibration. Because the outcome of any stage increase was structurally dependent on baseline stage, the overall performance estimate should not be assumed to apply uniformly across CHARLS-adapted CKM Stages 0–2. External validation is required before clinical application.