Deep Learning Model Predicts Heart Risk From Sleep ECGs

NIH-supported researchers used single-lead electrocardiograms recorded during sleep studies to forecast 10-year cardiovascular outcomes.

Updated on Nov. 2, 2026 in Heart Disease

Photograph of an empty sleep laboratory bed with a folded blanket and wires hanging from a ceiling bracket in soft morning light. — NIH-supported researchers developed a deep learning model that forecasts 10-year cardiovascular risk using single-lead electrocardiograms recorded during routine sleep studies. AI Illustration. Upload story photo >

Live Poll

Would you want heart-risk screening added to your next overnight sleep study?

An NIH-supported research team developed a deep learning model that predicts cardiovascular outcomes from electrocardiograms recorded during sleep studies. The findings were published in the journal Sleep.

Why it matters

The approach could add cardiovascular screening to tests already performed during sleep studies, which may help identify people at risk earlier. Its predictions improved on established risk factors, suggesting it captures information that conventional measures miss.

The model was trained on 15,809 patients at Massachusetts General Hospital and validated on 9,810 patients at Emory University Hospital and 12,576 patients at Beth Israel Deaconess Medical Center, predicting 10-year risk. Adding its score improved cardiac outcome prediction beyond established risk factors, but it does not yet predict heart attacks or strokes.

The players

National Institutes of Health (NIH)

The NIH is the primary U.S. federal agency for biomedical and health research, and it supported the research team behind this study.

Massachusetts General Hospital

Massachusetts General Hospital is a major teaching hospital in Boston and affiliate of Harvard Medical School, and it supplied the 15,809-patient dataset used to train the model.

Emory University Hospital

Emory University Hospital is an academic medical center in Atlanta that provided one of the validation datasets for this model.

Beth Israel Deaconess Medical Center

Beth Israel Deaconess Medical Center is a Boston teaching hospital affiliated with Harvard Medical School that provided a second validation dataset.

The details

Researchers used single-lead electrocardiograms paired with expert-annotated sleep stage data, rather than the usual 12-lead setup. The model sorted people into groups with different levels of long-term cardiovascular risk and retained its predictive value after adjusting for age, sex, BMI, diabetes, hypertension, sleep apnea severity, arousals, and sleep efficiency.

November 2, 2026: NIH released the announcement of the study findings.

Big Picture

Conventional cardiac screening relies on the 12-lead electrocardiogram, which requires multiple electrode placements. This study instead shows that a single-lead ECG recorded during sleep, combined with sleep stage data, can forecast long-term cardiovascular outcomes.

If sleep-study ECG screening is adopted, patients already undergoing sleep testing could receive an added cardiovascular risk estimate without a separate 12-lead ECG visit. The findings are early, so readers should not change care based on this research alone.

The takeaway

Sleep studies may eventually offer more than a sleep diagnosis, since the same recording could flag heart risk. Ask a clinician about cardiovascular screening if you have a sleep study or known risk factors for heart disease.