Artificial intelligence-enabled electrocardiographic sex discordance and the risk of incident atrial fibrillation: a multi-national cohort study.

Summary

An AI-ECG model generated a continuous “sex discordance” score that was strongly associated with incident AF risk in women across three large test cohorts and improved CHARGE-AF discrimination. No significant association was seen in men; the score correlated with hormone imbalance, adiposity, and atrial remodeling in women.

Key Findings

• External validation AUCs for AI-ECG sex prediction were 0.91 (CODE-15%) and 0.90 (MIMIC-IV).
• Each SD increase in sex discordance score was associated with higher AF risk in females (HR 1.28 and 1.32 in two cohorts), but not in males.
• Adding the score to CHARGE-AF improved C-index in females by 0.026 and 0.020; replicated in UK Biobank (HR 1.23; C-index +0.024).
• In females, the score correlated with sex hormone imbalance, pericardial/visceral adiposity, atrial remodeling, and adverse lifestyle factors.

Clinical Implications

Incorporating AI-ECG sex discordance into risk models may improve AF screening and risk factor modification strategies in women, informing monitoring intensity and lifestyle interventions.

Limitations

• Observational design limits causal inference; model performance and thresholds require clinical implementation work
• Male cohorts showed no significant association, limiting generalizability across sexes

Future Directions

Prospective implementation studies to test screening strategies triggered by sex discordance in women, calibration across devices/populations, and integration with biomarkers and imaging.