Machine Learning Algorithm to Predict Atrial Fibrillation Using Serial 12-Lead ECGs Based on Left Atrial Remodeling

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Subtle cardiac remodeling preceding atrial fibrillation (AF) occurrence may be detectable through serial electrocardiogram (ECG) analysis.
• Machine learning (ML) algorithms offer potential for analyzing complex ECG data to predict AF.
• Comparing single versus serial ECG analysis for AF prediction using ML is crucial for improving early detection.

Purpose of the Study:

• To compare the predictive performance of two ML algorithms for new-onset AF: one analyzing single ECGs and another analyzing serial ECGs.
• To investigate if serial ECG analysis can detect subtle cardiac remodeling indicative of future AF.
• To evaluate the efficacy of ML in identifying individuals at high risk for developing AF.

Main Methods:

• Development of two ML models (single ECG and serial ECG) using a light gradient boosting algorithm.
• Training ML models on a large dataset of 415,964 ECGs from 176,090 patients.
• External validation of model performance using metrics including sensitivity, specificity, accuracy, F1 score, and area under the receiver operating characteristic curve.

Main Results:

• The serial ECG-based ML model demonstrated significantly superior performance in predicting new-onset AF compared to the single ECG model.
• Serial-ML model achieved higher sensitivity (0.810 vs. 0.744), specificity (0.822 vs. 0.742), accuracy (0.816 vs. 0.743), and AUC (0.880 vs. 0.812).
• Shapley Additive Explanations analysis identified P-wave duration and amplitude as key predictive ECG parameters.

Conclusions:

• ML models utilizing serial ECGs possess a greater capacity for predicting new-onset AF than those based on single ECGs.
• Serial ECG analysis effectively captures evolving cardiac changes associated with future AF development.
• P-wave morphology characteristics are significant indicators for future AF prediction.