Explainable machine learning for risk stratification following TAVI

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Transcatheter Aortic Valve Implantation (TAVI) is a crucial minimally invasive treatment for severe aortic stenosis.
• Predicting post-operative outcomes in TAVI patients remains a significant clinical challenge.
• Accurate outcome prediction can optimize patient selection and management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting critical post-TAVI outcomes.
• To identify key clinical predictors influencing these outcomes.
• To enhance clinical decision-making in TAVI procedures through predictive analytics.

Main Methods:

• Utilized a dataset of 256 TAVI patients, processing 54 clinical parameters.
• Employed data preprocessing techniques including multiple imputation and Synthetic Minority Oversampling Technique (SMOTE).
• Trained and evaluated various ML models: Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Artificial Neural Networks (ANN).

Main Results:

• ML models demonstrated strong predictive performance for 30-day complications, mortality, Conduction Abnormalities (CA), and Paravalvular Leak (PVL).
• SVM and KNN models achieved the highest ROC-AUC scores, indicating robust predictive capabilities.
• SHapley Additive exPlanations (SHAP) identified diabetes, pacemaker use, dyslipidemia, and valve type as significant predictors.

Conclusions:

• Machine learning models, enhanced by explainability analysis, effectively predict TAVI outcomes.
• The findings suggest ML can aid in identifying high-risk TAVI patients.
• This approach has the potential to improve clinical decision-making and patient care in TAVI.