Comparing Phenotypes for Acute and Long-Term Response to Atrial Fibrillation Ablation Using Machine Learning

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Predicting patient response to atrial fibrillation (AF) ablation is difficult.
• AF ablation outcomes can be inconsistent, with some patients achieving long-term freedom despite initial procedural failure.

Purpose of the Study:

• To use machine learning to identify distinct phenotypes for acute and long-term AF ablation success.
• To differentiate the physiological factors influencing immediate versus sustained outcomes.

Main Methods:

• Analysis of 72 features from 561 AF patients, including electrograms, ECG, cardiac structure, lifestyle, and clinical data.
• Comparison of six machine learning models to predict acute and long-term ablation endpoints.
• Shapley explainability analysis to define patient phenotypes and external validation in 77 AF patients.

Main Results:

• The 1-year success rate was 69.5%, with a poor correlation between acute termination (49.6%) and long-term success.
• Machine learning models showed higher predictive power for acute termination (AUC=0.86) than long-term outcomes (AUC=0.67).
• Long-term success phenotypes were linked to clinical/lifestyle factors, while acute termination phenotypes reflected electrical features.

Conclusions:

• Acute and long-term responses to AF ablation are driven by distinct clinical and electrical physiology.
• The dissociation of phenotypes suggests potential roles for factors like attenuated AF progression in long-term success.
• Developing reliable procedural predictors for long-term AF ablation success remains a critical unmet need.