AI-Detected Asymptomatic Atrial Fibrillation and Risk of Incident Ischemic Stroke and Cardiovascular Events: A UK Biobank Study

BackgroundAdvances in wearable devices and machine-learning-based ECG analysis enable highly accurate detection of atrial fibrillation (AF) outside traditional clinical settings, leading to increasing identification of asymptomatic AF. However, the prognostic significance of AI-detected asymptomatic AF and its implications for downstream cardiovascular risk remain unclear. In contrast to clinically diagnosed AF, evidence guiding risk stratification and further evaluation in this population is limited. We therefore investigated the association between AI-detected asymptomatic AF and incident cardiovascular outcomes in a large population-based cohort. MethodsWe applied a validated open-source ECG-based deep learning model for atrial fibrillation detection (AI-AF) to 12-lead ECG recordings from participants in the UK Biobank. Participants with AI-detected AF on ECG and no prior clinical AF diagnosis were classified as asymptomatic AF (c). Kaplan-Meier curves and log-rank tests were used to compare the incidence of ischemic stroke and major adverse cardiovascular events (MACE: myocardial infarction, ischemic stroke, or cardiovascular death) across AF subgroups. Cox proportional hazards models were used to evaluate the association between AI-AF risk and incident MACE, adjusting for age, sex, current smoking, systolic blood pressure, total and HDL cholesterol, and prevalent type 2 diabetes. Follow-up was administratively censored at 6 years. ResultsThe study included 96,531 participants with mean [SD] age of 65 [8] years; 52% female; median follow-up [IQR] of 4.7 [1.6-7.2] years. ECG data were available for 64,029 participants and an additional 32,502 participants with clinically diagnosed atrial fibrillation (AF) without ECG recordings were included. Among participants without prior clinical AF and with available ECGs, 2,399 were classified as asympAF based on AI detection, while 58,879 were AF-free. Over 6 years of follow-up, the incidence of ischemic stroke was significantly higher in participants with asympAF compared with AF-free individuals (1.5% vs 0.52%, p = 7x10-7) and significantly lower than in participants with clinically diagnosed AF (1.5% vs 3.4%, p = 2x10-5). Similar patterns were observed for myocardial infarction and cardiovascular death. Using a more liberal AI-AF threshold corresponding to a 15% false-positive rate (asympAF15) yielded consistent findings: participants classified as asympAF15 had a 62% higher risk of incident MACE in adjusted Cox PH models (hazard ratio 1.6, 95% CI 1.2-2.2) over six years. ConclusionAI-detected asymptomatic AF identified individuals at elevated risk of ischemic stroke and major adverse cardiovascular events. As ischemic stroke is a hallmark complication of atrial fibrillation, these findings support the hypothesis that AI-ECG models may capture subclinical AF-related risk not detected by conventional clinical assessment. This approach may help extend the window for preventive interventions in populations without clinically diagnosed AF.