Artificial intelligence-enhanced electrocardiography for the identification of a sex-related cardiovascular risk continuum: a retrospective cohort study.

Summary

Using over 1.2 million ECGs for derivation and external validation in UK Biobank, an AI-ECG model produced a continuous sex discordance score that identified women at higher risk of cardiovascular death and incident HF/MI, while no association was seen in men. The score also tracked “male-like” cardiac and body composition phenotypes in at-risk women.

Key Findings

• AI-ECG sex classification achieved AUC 0.943 (BIDMC) and 0.971 (UK Biobank).
• Higher sex discordance score predicted cardiovascular death in women (HR 1.78 BIDMC; 1.33 UKB) but not in men.
• Women with higher discordance had greater future HF/MI risk and exhibited male-like cardiac (higher LV mass/volumes) and non-cardiac phenotypes (more muscle, less fat).

Clinical Implications

Clinicians could use AI-ECG sex discordance to flag women with normal ECGs but elevated latent risk for intensified risk factor control, surveillance, or referral for advanced imaging.

Limitations

• Retrospective design with potential residual confounding and selection biases
• Generalizability beyond studied health systems and ancestries requires prospective validation

Future Directions

Prospective implementation trials to test risk-guided prevention in women, mechanistic studies linking ECG features to sex-specific cardiac remodeling, and fairness audits across ancestries.