A Novel Predictive Model for Perioperative Complications Following Valve Replacement in Severe Valvular Heart Disease

Abstract

: Background: Valve replacement for severe valvular heart disease carries a risk of perioperative major adverse cardiovascular and cerebrovascular events (MACCE), which routine clinical indicators may help predict. Objective: To develop a machine learning model for predicting perioperative MACCE after valve replacement in severe valvular heart disease. Materials and Methods: We retrospectively included 327 patients with severe valvular heart disease who underwent valve replacement from January 2022 to December 2024; 229 were randomly assigned to the training cohort and 98 to the validation cohort at a 7:3 ratio. Demographic, clinical, and laboratory data were collected. MACCE during hospitalization within 30 days was the endpoint. In the training cohort, univariate analysis, least absolute shrinkage and selection operator (LASSO), and multivariate logistic regression identified independent predictors. Random forest (RF), support vector machine (SVM), and K-nearest neighbor (KNN) models were built in Python. The area under the curve (AUC) was the main performance metric, and the DeLong test compared models. Results: Multivariate analysis showed that pulmonary artery systolic pressure (PASP) and cardiopulmonary bypass (CPB) time were independent risk factors, whereas left ventricular ejection fraction (LVEF) and serum albumin were protective factors ( p < 0.05). Higher area under the curves (AUCs) were observed with the RF model (0.774 in training, 0.759 in validation) than with SVM (0.741, 0.723) or KNN (0.726, 0.707). However, pairwise comparisons using the DeLong method revealed no statistically significant differences between RF and the other two models ( p > 0.05). We adopted RF as the preferred model, favored for its sustained discriminative accuracy and stability from ensemble learning. Conclusions: A machine learning model using four routine parameters (LVEF, PASP, CPB time, and albumin) can assess MACCE risk after valve replacement in severe valvular disease. External validation is warranted before clinical adoption.