Interpretable machine learning predicts cardiac resynchronization therapy responses from personalized biochemical and...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Cardiac Resynchronization Therapy (CRT) is a key treatment for left ventricle (LV) failure.
• Many patients do not respond to CRT, necessitating methods to identify non-responders pre-treatment.
• Predicting CRT response requires integrating demographic, biomarker, and LV function data.

Purpose of the Study:

• To develop a machine learning algorithm for predicting individual patient response to CRT.
• To integrate diverse patient data types for enhanced predictive accuracy.
• To identify patient subgroups with varying likelihoods of CRT response.

Main Methods:

• An ensemble classification algorithm was developed using data from the SMART-AV CRT clinical trial (n=794).
• A five-fold cross-validation approach was used for model training (n=635) and validation (n=159).
• SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were employed for model interpretability.

Main Results:

• The algorithm achieved 71% accuracy in predicting CRT response using 56 variables (demographics, biomarkers, LV function).
• Stratification identified subgroups with high or low response likelihood, achieving 96% accuracy.
• Feature importance analysis provided insights into key predictive variables.

Conclusions:

• Integrating patient characteristics, comorbidities, therapy history, biomarkers, and LV function data can improve CRT response prediction.
• Machine learning offers a powerful tool for personalizing CRT treatment strategies.
• Accurate prediction of CRT response can optimize patient selection and improve therapeutic outcomes.