Artificial‑Intelligence‑Enabled Wearables for Early Detection of Atrial Fibrillation: A Systematic Review

Abstract 2294778, presented at Western Atrial Fibrillation Symposium 2026

Atrial fibrillation (AF) often presents intermittently or without symptoms, which delays diagnosis and increases the risk of stroke and heart failure. Smartwatches and similar devices now incorporate photoplethysmography or single-lead ECG sensors coupled with artificial intelligence (AI) algorithms to detect AF. We evaluated the diagnostic performance and clinical impact of AI-enabled wearables for early AF detection. PubMed, Embase and IEEE Xplore were searched (January 2015-September 2025) for studies reporting sensitivity and specificity of AI-enabled wearables against physician-interpreted ECG. Data on user adherence and downstream clinical outcomes (e.g., initiation of anticoagulation) were extracted. Pooled estimates were calculated using random effects models. Twenty-two studies (~350,000 recordings) met inclusion criteria. The majority used deep learning algorithms to interpret photoplethysmography or single-lead ECG signals. Pooled sensitivity and specificity for AF detection were 95% (95% CI 92–97) and 92% (88–96). False positives were mainly due to motion artefacts and ectopic beats. Only five studies reported clinical outcomes; these showed earlier initiation of anticoagulation and increased patient satisfaction versus standard care. Device adherence declined after six months. AI-enabled wearables provide high diagnostic accuracy and hold promise for timely AF diagnosis, though confirmatory ECG is still required. Large prospective trials are needed to determine whether their use improves stroke prevention and long-term outcomes.