Artificial intelligence-enabled electrocardiographic sex discordance and the risk of incident atrial fibrillation: A multinational cohort study

Abstract

BACKGROUND Binary classification of sex fails to capture the sex-related continuum of atrial fibrillation (AF) risk. OBJECTIVE This study aimed to develop an artificial intelligence (AI)-enabled electrocardiography (ECG) model for sex prediction and explore its association with AF risk. METHODS An AI-ECG model for sex prediction was developed from the Severance Hospital training set and externally validated using Clinical Outcomes in Digital Electrocardiography 15% (area underthe curve 0.91) and Medical Information Martfor Intensive Care IV (area under the curve 0.90) datasets. A sex discordance score-defined as 1 minus the AI-ECG-predicted probability (continuous) forself-reported sex-was estimated in AF-free individuals on 3 multinational test sets (Severance Hospital [n = 205,769], Yongin Severance Hospital [n = 112,942], and UK Biobank [n = 40,525]). RESULTS In the Severance Hospital and Yongin Severance Hospital test sets, sex discordance score increase was associated with higher AF risk in females, with Cohorts for Heart and Aging Research in Genomic Epidemiology Atrial Fibrillation (CHARGE-AF)-adjusted hazard ratio per standard deviation of 1.28 (95% confidence interval [CI] 1.24-1.33) and 1.32 (95% CI 1.27-1.36), respectively. No significant association was observed in males. Adding sex discordance score to the CHARGE-AF model significantly improved discrimination for AF in females, with a C-index increase of 0.026 (95% CI 0.013-0.037) and 0.020 (95% CI 0.010-0.032) in the respective datasets, but not in males. In the UK Biobank test set, a similar association between sex discordance score and incident AF risk was observed in females (CHARGE-AF-adjusted hazard ratio per standard deviation 1.23 [95% CI 1.13-1.32]; C-index increase 0.024 [95% CI 0.005-0.040]). In females, sex discordance score correlated with sex hormone imbalance, pericardial and visceral adiposity, atrial remodeling, and adverse lifestyle factors. CONCLUSION The AI-ECG sexdiscordance score captures females with disproportionately elevated AF riskwith implications for enhanced risk factor modification and surveillance.