Diagnostic Accuracy of AI–Enabled ECG for Atrial Fibrillation Detection: A Meta-Analysis

Abstract 2312141, presented at Western Atrial Fibrillation Symposium 2026

As AI-enabled ECG tools move rapidly into routine clinical use, whether they reliably detect atrial fibrillation at the moment it occurs has become a central clinical question. Much of the existing literature focuses on AF risk prediction or latent AF signatures during sinus rhythm rather than true rhythm-level detection on contemporaneous ECGs. We aimed to evaluate the diagnostic accuracy of AI-based ECG algorithms for detecting AF present at the time of ECG acquisition over the last decade.We conducted a systematic review and diagnostic test accuracy meta-analysis in accordance with PRISMA-DTA guidelines. Eligible studies evaluated AI-based ECG algorithms for detection of atrial fibrillation present on the ECG at the time of recording, used a physician-adjudicated reference standard, and provided extractable true positive, false positive, false negative, and true negative counts. Studies focused on AF prediction during sinus rhythm or episode-based surveillance were excluded. Pooled sensitivity and specificity were estimated using a random-effects diagnostic framework accounting for between-study heterogeneity.Five diagnostic accuracy studies between 2018 and 2025 were included. They encompassed clinical and wearable single-lead ECG platforms, representing several thousand ECG recordings across inpatient, outpatient, and ambulatory settings. Across studies, sensitivity for AF detection ranged from approximately 85% to 95%, while specificity ranged from 82% to 98%. Pooled sensitivity was approximately 90% (95% CI ~87–93%), and pooled specificity was approximately 93% (95% CI ~90–96%), indicating reliable identification of AF when present and strong discrimination from non-AF rhythms. Diagnostic performance was consistent across device types when ECG recordings were interpretable.Over the past decade, AI-enabled ECG interpretation for atrial fibrillation detection has matured into a reliable diagnostic tool with consistently high accuracy across platforms. These findings support its growing role in improving AF detection and timely initiation of treatment. At the same time, increased identification of clinically silent AF highlights the importance of clinician judgment in determining which diagnoses warrant intervention. These findings suggest AI-ECG tools are best viewed as adjuncts that enhance rhythm detection rather than replacements for clinical decision-making.