AI-enhanced ECG identifies increased heart disease risk in women
A new artificial intelligence model analyzing electrocardiograms helps identify female patients at higher risk for heart disease, offering earlier detection and more personalized treatment.
AI-enhanced ECG identifies increased heart disease risk in women | Image Credit: © metamorworks - © metamorworks - stock.adobe.com.
Introduction
An artificial intelligence (AI) model is able to evaluate an electrocardiogram (ECG) to identify increased heart disease risk in female patients, according to a recent study published in Lancet Digital Health.1
An ECG is one of the most common medical tests performed globally, capable of recording electrical activity in the heart. Research stated the algorithm will allow for improved treatment by identifying women with increased heart disease risk sooner.1
“In the clinic we use tests like ECGs to provide a snapshot of what’s going on but as a result this may involve grouping patients by sex in a way that doesn’t take into account their individual physiology,” said Arunashis Sau, PhD, Academic Clinical Lecturer at Imperial College London. “The AI enhanced ECGs give us a more nuanced understanding of female heart health.”1
A link has been reported between sex misclassification in AI-ECG models and adverse prognosis, but it is unclear whether the misclassification is a failure of the models themselves.2 Therefore, the retrospective cohort study was conducted to evaluate the significance of AI-ECG sex misclassification in low-risk cohorts.
Study design
Participants included a cohort from the Beth Israel Deaconess Medical Center (BIDMC) and a cohort from the UK Biobank. The BIDMC cohort had more comorbidities than the UK Biobank cohort because of being hospital-based.2
Patients were considered healthy if they did not present with diabetes, hypertension, heart failure, hyperlipidemia, atherosclerotic cardiovascular disease, smoking history, or myocardial infarction. Those who were aged over 40 years, inpatients, or with prevalent cardiovascular disease were considered high-risk.2
Pre-processing of ECGs was performed using a bandpass filter 0·5 to 100 hz, a notch filter at 60 hz, and re-sampling to 400 hz. The BIDMC cohort was considered the derivation dataset when developing the model, with data split by patient identification and stratified based on ECG presence.2
Sex predictions were generated for all ECGs with an AI-ECG sex discordance score calculated by subtracting self-reported sex from the absolute of all AI-ECG sex prediction. All-cause and cardiovascular death were reported as primary outcomes of the analysis, secondary outcomes included myocardial infarction, future heart failure, and non-cardiovascular mortality.2
Study results
There were 1,163,401 ECGs from 189,539 individuals in the BIDMC cohort, with a mean follow-up duration of 3.41 years. Death during follow-up was reported in 18.4%, and the area under the receiver operating characteristic (AUROC) curve of AI-ECG for identifying sex was 0.943, indicating accuracy.2
Higher risk cohorts and ECGs with abnormalities reported reduced performance from the model. However, when validating the model in a cohort of UK Biobank volunteers, accuracy remained high for predicting sex, with an AUROC of 0.971.2
Based on data from both cohorts, the model had significant sex prediction accuracy. The greatest effects were noted at the extremes of the distributions, with an accuracy of 93.1% in BIDMC and 95.5% in UK Biobank when sex predictions were under 0.2 or over 0.8. This accuracy did not differ based on gender.2
Investigators also found a link between sex discordance score and all-cause mortality, with hazard ratios (HRs) of 1.22 and 1.17 for male and female patients, respectively. Additional HRs included 1.20 and 1.25, respectively, for cardiovascular mortality and 1.23 and 1.15, respectively, for non-cardiovascular mortality.2
Conclusion
“Harnessing the potential of this type of research could help better identify those patients at highest risk of future heart problems and reduce the gender gap in heart care outcomes,” said Fu Siong Ng, PhD, consultant cardiologist at Imperial College Healthcare NHS Trust.1
“However, one test alone will not level the playing field,” Ng added. “Ensuring every person gets the right heart care they need when they need it will require change in every part of our health care system.”1
References
- AI model can read ECGs to identify female patients at higher risk of heart disease. Imperial College London. February 25, 2025. Accessed March 7, 2025. https://www.eurekalert.org/news-releases/1074814
- Sau A, Sieliwonczyk E, Patlatzoglou K, et al. Artificial intelligence-enhanced electrocardiography for the identification of a sex-related cardiovascular risk continuum: a retrospective cohort study. The Lancet Digital Health. 2025;7(3):E184-E194. doi:10.1016/j.landig.2024.12.003
Single iron infusion in late pregnancy significantly reduces anemia risk
January 22nd 2025A recent study found that a single dose of intravenous iron in the third trimester is superior to oral iron in reducing anemia rates in pregnant women, ensuring better outcomes for mother and baby.
Read More
