Machine Learning Model Predicts Three Year Mortality in Cardiovascular Kidney Metabolic Syndrome

Cardiovascular-kidney-metabolic (CKM) syndrome is associated with a high risk of mortality, yet accurate individual risk prediction remains limited. We developed and validated machine learning models to predict 3-year all-cause mortality in 219,561 hospitalized patients with CKM stages 2-4 from 29 medical centers. Extreme Gradient Boosting (XGBoost) and least absolute shrinkage and selection operator (LASSO) models were developed in a derivation cohort ( n = 132,404) using 101 variables and evaluated in internal ( n = 56,744) and center-based ( n = 30,413) validation cohorts. An 8-variable XGBoost model consistently outperformed the LASSO model, achieving receiver operating characteristic curve [ROC-AUC] of 0.831, 0.826, and 0.813 in the derivation, internal, and center-based validation cohorts, respectively. Based on the optimal model, patients were stratified into low-, moderate-, and high-risk groups. Compared with the low-risk group, high-risk patients had substantially higher risks of 3-year all-cause mortality (hazard ratio [HR], 9.62 [8.86, 10.45]) and cardiovascular mortality (HR, 12.53 [10.97, 14.31]). A web-based risk calculator was developed to facilitate clinical application. This parsimonious 8-variable XGBoost model provides accurate mortality risk stratification and may support personalized management of patients with CKM syndrome.