Predicting Recurrence and Outcomes After Stressor-Associated Atrial Fibrillation Using ECG-Based Deep Learning

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Stressor-associated atrial fibrillation (AF) is new-onset AF triggered by acute stressors.
• Identifying patients at high risk for AF recurrence is crucial for management.
• Current clinical factors have limited predictive value for AF recurrence.

Purpose of the Study:

• To evaluate the utility of artificial intelligence (AI)-enabled 12-lead ECG models in estimating the risk of AF recurrence.
• To develop and validate a predictive model for AF recurrence incorporating clinical factors, stressor type, and AI-derived risk estimates.

Main Methods:

• Retrospective analysis of 3371 patients with stressor-associated AF.
• Quantified cumulative incidence of AF recurrence, accounting for death as a competing risk.
• Developed a penalized regression model using clinical data, stressor type, and AI-based ECG risk estimates to predict recurrence.

Main Results:

• The 10-year cumulative incidence of AF recurrence was 41%.
• AF recurrence was associated with a 2.24-fold increased risk of AF-related adverse events.
• The combined clinical-AI model demonstrated superior discrimination for AF recurrence (AUC 0.768) compared to clinical factors alone (AUC 0.707).

Conclusions:

• High rates of AF recurrence following stressor-associated AF pose a significant risk for adverse cardiovascular events.
• AI-enhanced ECG risk estimates improve the prediction of AF recurrence.
• These models can help identify high-risk individuals for targeted monitoring and preventive interventions.