Prediction of response after cardiac resynchronization therapy with machine learning

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Cardiac resynchronization therapy (CRT) is a treatment for heart failure, but nearly one-third of patients do not respond.
• Identifying non-responders before implantation is crucial for optimizing treatment and resource allocation.

Purpose of the Study:

• To develop and evaluate machine learning (ML) predictive models for CRT response.
• To identify easily attainable patient features predictive of CRT success prior to implantation.

Main Methods:

• Retrospective analysis of 752 CRT recipients' baseline characteristics.
• Development and comparison of nine ML models (logistic regression, elastic net, lasso, ridge, neural network, SVM, random forest, XGBoost, k-NN).
• Evaluation of models using sensitivity, specificity, AU-ROC, and other metrics; comparison against current guidelines.

Main Results:

• Six ML models achieved an AU-ROC above 0.75, with logistic regression, elastic net, and ridge regression showing the highest predictive power (AU-ROC 0.77).
• All ML models outperformed existing guidelines in predicting CRT response (P < 0.05).
• Key predictors identified include left bundle branch block, left ventricular end-systolic diameter, and prior percutaneous coronary intervention.

Conclusions:

• Machine learning algorithms can create effective predictive models for CRT response using pre-implantation data.
• These predictive tools have the potential to enhance the selection of CRT candidates and decrease the rate of non-response.