Risk stratification for atrial fibrillation in hypertrophic cardiomyopathy using a deep learning–based ECG model

Background Atrial fibrillation (AF) is highly prevalent in patients with hypertrophic cardiomyopathy (HCM) and confers a substantial risk of ischemic stroke. However, electrocardiography (ECG)-based prediction of AF in this population has not been well established. Purpose We aimed to to evaluate a 12-lead ECG-based artificial intelligence (AI) model to predict incident AF and ischemic stroke in patients with HCM. Methods In this retrospective single-center study, consecutive patients diagnosed with HCM from 1997 to 2021 were enrolled. Baseline 12-lead ECGs obtained at the time of HCM diagnosis were analyzed using a previously developed recurrent neural network–based AI model to estimate future AF risk (SmartECG-AFrisk). Patients with AF at baseline were excluded. The primary endpoint was new-onset AF, and the secondary endpoint was ischemic stroke. Results A total of 585 patients (mean age 58.9 years; 66.2% male) were included. Mean left ventricular ejection fraction and left atrial dimension was 60.4 % and 43.1 mm, respectively. Over a mean follow-up of 6.8 years, 86 patients (14.7%) developed AF and 47 (8.0%) experienced ischemic stroke. Patients classified as AI high risk had a significantly higher incidence of AF compared with those at low risk (27.8 vs. 13.9 per 1,000 person-years; hazard ratio [HR] 2.06, 95% confidence interval [CI] 1.09–3.88; p=0.023). The AI high-risk group also demonstrated a markedly increased risk of ischemic stroke (15.2 vs. 2.4 per 1,000 person-years; HR 6.20, 95% CI 1.50–25.61; p=0.004). Conclusions In patients with HCM, AI-guided ECG risk stratification effectively identified individuals at increased risk of incident AF and ischemic stroke. ECG-based AI assessment may provide incremental value for long-term risk stratification in this high-risk population.