Machine learning based predictive model for cardiovascular disease mortality risk in patients with stage 4 cardiovascular–kidney–metabolic syndrome

Patients with stage 4 cardiovascular–kidney–metabolic (CKM) syndrome are exposed to an elevated cardiovascular disease (CVD) mortality risk. However, large-sample, multidimensional prediction models for CVD mortality risk that integrate clinical indicators, as well as highly applicable supporting tools, remain lacking. A total of 2893 patients with stage 4 CKM were randomly allocated to training and validation sets at a 7:3 ratio. The prediction model was constructed via variable screening using least absolute shrinkage and selection operator combined with multivariate Cox regression. Discrimination, calibration, and clinical utility of the model were evaluated with receiver operating characteristic curves, calibration plots, and decision curve analysis. A nomogram and web-based calculator were constructed. Six machine learning models were used to validate the model, and predictive factors were identified via SHapley Additive exPlanations analysis. Nine risk factors were ultimately identified for constructing the predictive model: age, gender, mobility difficulty, chronic kidney disease, anemia, neutrophil, blood urea nitrogen, albumin, and C-reactive protein. The model’s discriminate ability on training set (C-index = 0.773) was comparable that on validation set (C-index = 0.779). Additionally, the model exhibited optimal performance in predicting 15-year CVD mortality (area under the curve = 0.871). The calibration curve demonstrated comparable fit, and decision curve analysis revealed that the nomogram score provided higher clinical benefit compared with any single variable. Participants with high nomogram scores exhibited a lower survival rate (log-rank test, P  < .001). Among the 6 machine learning models, Random Forest and XGBoost achieved the best performance. SHapley Additive exPlanations analysis further indicated that age and C-reactive protein made the greatest contributions to predicting CVD mortality risk. The CVD mortality risk prediction model developed for patients with stage 4 CKM demonstrated robust performance. This model may enable clinicians to identify high-risk patients early and to formulate individualized intervention strategies through a reliable decision support tool.