Abstract Body (Do not enter title and authors here): Introduction: Using sex as a binary variable may oversimplify the spectrum of inter-individual variability in sex-related cardiovascular (CV) risk. Artificial intelligence–enhanced electrocardiography (AI-ECG) models can accurately predict sex in populations from Europe and the US, in whom sex misclassification is associated with adverse CV outcomes in women, but not in men. It is unknown whether AI-ECG accurately identifies sex or predicts CV risk in diverse populations.
Objective: To validate the AI-ECG sex-discordance score in the diverse ELSA-Brasil cohort and assess whether an increased sex-discordance score is associated with 5-year CV events.
Methods: In the community-based ELSA-Brasil study, we validated the AI-ECG model that predicts sex as a continuous variable. The outcome was mortality or hospitalizations due to myocardial infarction, stroke, heart failure, or revascularization. The AI-ECG sex-discordance score (absolute difference between the AI-predicted sex and self-reported sex, encoded as 0 for men and 1 for women) was analyzed as a standardized continuous variable and quartiles. Association between the sex-discordance score and outcome was assessed using sex-specific multivariable Fine and Gray models accounting for the competing risk of death, adjusted for age, race, education, smoking, physical activity, excessive alcohol consumption, body mass index, hypertension, diabetes, dyslipidemia, and prevalent CV disease. We also evaluated the association of sex-discordance scores with CV risk factors using robust linear regression M-estimator and 95% efficiency.
Results: In 13,730 participants from the ELSA-Brasil study (mean age=52±9, 54% women, 45% Black), AI-ECG accurately identified sex (AUC:0.963, 95%CI:0.960-0.965). In women, each 1-SD increase in sex-discordance score was borderline associated with higher risk of CV outcomes (HR 1.19; 95%CI: 1.00–1.42; p=0.057), but no association was seen in men (HR 1.07; 95%CI: 0.80–1.41; p=0.649). Women in the highest quartile of sex-discordance had significantly higher risk compared to the lowest quartile (HR 1.42; 95%CI: 1.02–1.98; p=0.036), but not men. In women, overweight/obesity, smoking, hypertension, and diabetes were associated with higher sex-discordance scores.
Conclusions: AI-ECG accurately identifies sex in a diverse population, in whom a high sex-discordance score identifies women with higher CV risk, who might benefit from targeted CV prevention.