Background: Artificial intelligence-enhanced electrocardiography (AI-ECG) for detecting atrial fibrillation (AF) using sinus rhythm ECGs has shown promise. However, even when AI-ECG results are positive, ECG confirmation of AF may not always be obtained immediately. Repeat testing may be required, particularly in patients with a dilated left atrium, who may warrant more intensive monitoring. We aimed to investigate whether newly detected AF was frequent in patients with both high risk on AI-ECG evaluation and dilated left atrium.
Methods and Results: We used a convolutional neural network-based ECG algorithm to predict AF using data (2010–2022) from the Shinken database (n=12,595 patients without a prior AF event). The 3-year incidence of newly detected AF was compared across 3 left atrial diameter (LAD) categories: <35, 35–39, and ≥40 mm (small, middle, and large, respectively). Patients were stratified by the AI-ECG-generated diagnostic probability of AF (AIECG-AF-DP). The incidence of newly detected AF increased according to LAD category among patients with high (≥0.8) vs. low (<0.8) AIECG-AF-DP: 3.2 vs. 0.5%/year for small; 6.1 vs. 0.6%/year for middle, and 11.6 vs. 1.5%/year for large, respectively (all P<0.001). Although the area under the receiver operating characteristic curve was similar across LAD categories of <35, 35–39, and ≥40 mm (0.770, 0.753, and 0.784, respectively), the area under the precision-recall curve differed markedly (0.083, 0.114, and 0.236, respectively).
Conclusions: Newly detected AF was particularly frequent in patients with both high AIECG-AF-DP and large LAD, suggesting repeated AF screening may be warranted in this population.
View full abstract