Value of machine learning in predicting TAVI outcomes.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Transcatheter aortic valve implantation (TAVI) is a common procedure for high-risk aortic stenosis.
• Predicting the benefits of TAVI for individual patients remains challenging.
• Machine learning (ML) offers potential for improving outcome prediction in TAVI.

Purpose of the Study:

• To evaluate the accuracy of traditional ML algorithms for predicting TAVI outcomes.
• To assess the predictive performance of ML models using different data sources.
• To identify the most effective ML techniques for TAVI outcome prediction.

Main Methods:

• Utilized clinical and laboratory data from 1,478 TAVI patients.
• Applied five well-established ML techniques: Random Forest, Logistic Regression, etc.
• Evaluated model performance using Area Under the Curve (AUC) for mortality and dyspnea improvement.

Main Results:

• Random Forest achieved the highest AUC (0.70) for predicting mortality.
• Logistic Regression showed the highest AUC (0.56) for predicting dyspnea improvement.
• Tree-based models demonstrated slightly superior mortality prediction accuracy.

Conclusions:

• ML models, particularly tree-based ones, show moderate predictive accuracy for TAVI mortality.
• ML models performed poorly in predicting improvement in dyspnea post-TAVI.
• Further research is needed to enhance ML model performance for predicting TAVI symptom relief.