Machine learning prediction of mitral regurgitation improvement after TAVI

The fate of mitral regurgitation after transcatheter aortic valve implantation is highly heterogeneous, and accurate preoperative prediction of mitral regurgitation improvement remains challenging. This study aimed to develop and validate a machine learning-based predictive model for regurgitation improvement following transcatheter aortic valve implantation. This single-centre retrospective study included 324 patients with moderate-to-severe mitral regurgitation who underwent transcatheter aortic valve implantation between 2019 and 2024 (development cohort) and 120 patients in 2025 (temporal validation cohort). Eleven machine learning algorithms were compared. Feature selection was performed using a dual-screening strategy combining the Boruta algorithm and Least Absolute Shrinkage and Selection Operator. Model performance was evaluated using the area under the receiver operating characteristic curve, calibration curves, and decision curve analysis. SHapley Additive exPlanations analysis was used to interpret predictor contributions to model predictions. Six variables were identified as robust predictors: functional mitral regurgitation, interventricular septal thickness, atrial fibrillation, left ventricular ejection fraction, surgical approach, and left atrial diameter. In temporal validation, logistic regression achieved an AUC of 0.788 (95% CI 0.694–0.883), followed by Least Absolute Shrinkage and Selection Operator regression (AUC, 0.787; 95% CI 0.693–0.882) and CatBoost (AUC, 0.778; 95% CI 0.683–0.872). Interventricular septal thickness showed a significant nonlinear association with mitral regurgitation improvement, with approximately 1.1 cm suggested as a provisional breakpoint in an exploratory analysis. This machine learning model enables individualized prediction of mitral regurgitation improvement after transcatheter aortic valve implantation using readily available preoperative parameters. The identification of interventricular septal thickness as a nonlinear predictor with a potentially clinically relevant threshold suggests promise for preoperative risk stratification. Nevertheless, because both the development and temporal validation cohorts were derived from a single centre, these findings are preliminary and require prospective multicentre validation before any clinical implementation.