Development and internal validation of a model predicting four year cardiovascular kidney metabolic syndrome stage progression

Abstract To develop a model for predicting cardiovascular-kidney-metabolic (CKM) syndrome stage progression in middle-aged and older Chinese adults and to evaluate the incremental predictive value of routinely available clinical variables beyond baseline CKM stage. This study used data from the 2011 baseline and 2015 follow-up waves of the China Health and Retirement Longitudinal Study. Participants were eligible if they were aged 45 years or older, had baseline CKM Stage 0–3, and had ascertainable CKM stage at baseline and follow-up. The primary outcome was any upward CKM stage transition over approximately 4 years. Three logistic regression models were developed: Model A included baseline CKM stage, age, and sex; Model B included clinical predictors without baseline CKM stage; and Model C added penalization-selected clinical predictors to baseline CKM stage, age, and sex. Missing candidate predictors were handled using multiple imputation by chained equations. Model A was fitted as an unweighted logistic regression model in each imputed dataset and coefficients were pooled using Rubin’s rules. Models B and C were developed using observation-completeness-weighted stacked elastic-net logistic regression across the imputed datasets. Models A and C underwent subject-level bootstrap internal validation. The final cohort included 5,346 participants (median age, 59 years; 52.1% women), of whom 1,362 (25.5%) experienced an upward CKM stage transition. Model C retained baseline CKM stage, age, sex, body mass index, hypertension, dyslipidemia, and lipid-lowering treatment. The apparent areas under the receiver operating characteristic curve (AUCs) were 0.748, 0.679, and 0.770 for Models A, B, and C, respectively; the corresponding Brier scores were 0.154, 0.176, and 0.151. After bootstrap optimism correction, the AUCs were 0.746 for Model A and 0.765 for Model C. The optimism-corrected incremental AUC and Brier score improvement for Model C versus Model A were 0.0190 and 0.00241, respectively. Findings were consistent across sensitivity analyses. Within-stage discrimination of Model C varied substantially, with AUCs ranging from 0.540 to 0.691. Among Chinese adults aged 45 years or older, CKM stage changed substantially over approximately 4 years. Baseline CKM stage provided most of the information for predicting overall stage progression, while routinely available clinical variables, including body mass index, hypertension, dyslipidemia, and lipid-lowering treatment, provided modest incremental predictive value.