Prediction of Atrial Fibrillation Using Artificial Intelligence-Enhanced Electrocardiography　- Does Left Atrial Size...

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Diagnostics

Background:

• Artificial intelligence-enhanced electrocardiography (AI-ECG) shows promise for detecting atrial fibrillation (AF) from sinus rhythm ECGs.
• Immediate ECG confirmation of AF is not always achieved, necessitating repeat testing, especially in patients with dilated left atrium.
• Investigating the frequency of newly detected AF in patients with high AI-ECG risk and dilated left atrium is crucial for targeted monitoring.

Purpose of the Study:

• To determine the incidence of newly detected atrial fibrillation (AF) in patients stratified by AI-ECG risk and left atrial diameter (LAD).
• To assess the correlation between AI-ECG-AF diagnostic probability (AIECG-AF-DP) and LAD in predicting new-onset AF.
• To identify patient subgroups that may benefit from intensified AF screening protocols.

Main Methods:

• Utilized a convolutional neural network-based AI-ECG algorithm on the Shinken database (n=12,595) of patients without prior AF events (2010-2022).
• Compared the 3-year incidence of newly detected AF across three LAD categories (<35 mm, 35-39 mm, ≥40 mm) stratified by AIECG-AF-DP (high ≥0.8 vs. low <0.8).
• Analyzed receiver operating characteristic (ROC) and precision-recall curves across LAD categories.

Main Results:

• The incidence of newly detected AF significantly increased with LAD categories among patients with high AIECG-AF-DP (3.2%, 6.1%, 11.6%/year for small, middle, large LAD, respectively; P<0.001).
• Incidence rates for low AIECG-AF-DP were substantially lower across all LAD categories (0.5%, 0.6%, 1.5%/year).
• Precision-recall curves showed marked differences across LAD categories, unlike ROC curves, highlighting the predictive value in specific subgroups.

Conclusions:

• Newly detected AF is particularly frequent in patients with both high AI-ECG-AF-DP and large LAD (≥40 mm).
• These findings suggest that repeated AF screening may be particularly warranted in this high-risk population.
• AI-ECG combined with LAD assessment can help identify individuals needing closer AF monitoring.