Explainable machine learning for risk stratification following TAVI

Objectives: Transcatheter Aortic Valve Implantation (TAVI) is a minimally invasive procedure for treating severe aortic stenosis, but accurately predicting post-operative outcomes remains a challenge. This study employs machine learning (ML) techniques to predict four critical outcomes following TAVI: 30-day complications, mortality, Conduction Abnormalities (CA), and Paravalvular Leak (PVL). Methods: Using a dataset of 256 patients from Hamad Medical Corporation Heart Hospital TAVI program, we processed 54 clinical parameters through several preprocessing methods such as multiple imputation for missing values and synthetic data generation using Synthetic Minority Oversampling Technique (SMOTE) to address missing data and class imbalance. Various machine learning models, including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Artificial Neural Networks (ANN), were trained and evaluated. Results: The models demonstrated strong predictive performance across all tasks, with SVM and KNN models achieving the highest ROC-AUC scores. SHapley Additive exPlanations (SHAP) analysis revealed key predictors such as diabetes, pacemaker, dyslipidemia and type of valve, providing insights into the factors contributing to each prediction. Conclusion: Our findings suggest that ML, combined with explainability analysis, can effectively identify high-risk patients and potentially improve clinical decision-making in TAVI procedures.