Machine learning-based predictive model for atrial arrhythmia following transcatheter atrial septal defect closure

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Atrial septal defects (ASDs) are commonly treated with percutaneous closure.
• Atrial arrhythmias can occur despite successful ASD closure.
• Risk factors for post-procedural atrial arrhythmias are not well-defined.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting atrial arrhythmias after transcatheter ASD closure.
• To identify key predictors of atrial arrhythmias in patients undergoing ASD closure.

Main Methods:

• Retrospective analysis of 148 adult patients with secundum-type ASDs undergoing transcatheter closure.
• Utilized a deep neural network with transfer learning to extract features from preprocedural ECGs.
• Ensemble survival models combined ECG-derived features with clinical, demographic, biochemical, and hemodynamic data.

Main Results:

• The ML model demonstrated strong predictive performance (AUC 0.823).
• 28 out of 148 patients (18.9%) developed atrial arrhythmias during follow-up.
• ECG-derived features significantly contributed to the model's predictive accuracy.

Conclusions:

• A novel ML-based risk model was developed to predict atrial arrhythmias post-transcatheter ASD closure.
• The model shows promise for risk stratification in clinical practice.
• External validation is recommended for further refinement and implementation.