Journal of the American Heart Association · Sep 29, 2026
Artificial Intelligence-Derived Electrocardiographic Age Enhances Stroke Risk Stratification in Patients With Atrial Fibrillation With a CHA₂DS₂-VA Score of 1
By Wu CS, Chen PH, Lin WY et al.
Background To evaluate whether artificial intelligence-derived electrocardiographic-age enhances the predictive utility of the CHA₂DS₂-VA (congestive heart failure/left ventricular dysfunction, hypertension, age ≥75 years [doubled], diabetes, stroke/transient ischemic attack [doubled], vascular disease, age 65-74 years) score in patients with atrial fibrillation representing a borderline indication for anticoagulation with a score of 1. Methods From a multicenter atrial fibrillation and atrial flutter registry, we identified 832 anticoagulation-naïve patients with a CHA₂DS₂-VA score of 1 and at least one 12-lead ECG within 90 days of diagnosis. ECG-age was estimated using a previously validated deep learning (convolutional neural network) model. Chronological age was replaced with ECG-age (1 point for 65-74 years, 2 points for ≥75 years), leaving all other score components unchanged, to form the EA-CHA₂DS₂-VA score. Patients were grouped as EA-CHA₂DS₂-VA ≥2 (n=315) or <2 (n=517). Results Compared with patients with an EA-CHA₂DS₂-VA score <2, those with an EA-CHA₂DS₂-VA score ≥2 had a significantly higher risk of ischemic stroke or transient ischemic attack (hazard ratio [HR], 1.82 [95% CI, 1.06-3.13]; P=0.028). All-cause mortality was also significantly elevated in the EA-CHA₂DS₂-VA ≥2 group (HR, 2.22 [95% CI, 1.43-3.45]; P<0.001). Major bleeding did not differ significantly between groups (P=0.386). Conclusions Substituting chronological age with artificial intelligence-derived ECG-age in the CHA₂DS₂-VA score was associated with improved risk stratification in patients with atrial fibrillation with intermediate stroke risk. This refined risk stratification facilitates timely initiation of anticoagulation therapy, potentially reducing stroke and mortality. Future prospective studies are essential to integrate this artificial intelligence-enhanced strategy into clinical practice. Registration This study was registered on the Open Science Framework (URL: https://doi.org/10.17605/OSF.IO/39BND; unique identifier: 10.17605/OSF.IO/39BND).
doi.org/10.1161/jaha.126.050723Open accessPubMed 42808453Biomarkers