AI-ECG for early detection of atrial fibrillation: First-year results from a stroke prevention study in Shimizu, Japan

Abstract

Background

An artificial intelligence algorithm-guided electrocardiogram (AI-ECG) has been developed to detect atrial fibrillation (AF) in patients with sinus rhythm (SR). However, its utility for population-based screening remains unclear in Japan.

Method and Results

In this prospective cohort study, “SPAFS” (Stroke Prevention by Early Detection of AF in Shimizu), participants who underwent health examinations at the Shimizu Medical Association Examination Center from January 2022 to July 2023 were enrolled, with known AF excluded. ECGs were categorized by AI as low-, moderate-, or high risk: non-SR were labeled as non-applicable (NA). All participants underwent 7-day single-lead ECG monitoring. Among 362 participants (61.1 ± 10.5 years, 38% male, CHADS2 score 0.49 ± 0.70), AF was newly detected in 3.0% (n = 11), with increasing prevalence across AI risk categories. The non-low-risk group (moderate, high, and NA) had a significantly higher AF detection rate than the low-risk group (OR 9.36, 95% CI 1.99–44.01). Subgroup analysis in those aged ≥65 years showed a similar trend (OR 8.09 [95%CI 1.63–39.7]). When the NA group (not eligible for AI) was excluded, similar trends were observed, although statistical significance was attenuated (OR 4.89 [95% CI 0.88–27.1] in the total, 5.09 [95% CI 0.89–29.0] in those aged ≥65 years). In the total cohort, AI-ECG showed higher discriminative ability than the CHADS₂ score ≥1 in both the total cohort (AUC 0.75 vs. 0.68) and participants aged ≥65 years (AUC 0.73 vs. 0.61).

Conclusions

AI-ECG risk determination correlated with AF detection in a Japanese healthy cohort, especially in the aged population, supporting its utility as a population-based screening tool.