The results showed a clear gradient: people with higher model scores carried higher long-term cardiovascular risk. The model's predictive value held up even after the researchers adjusted for common cardiovascular risk factors — age, sex, BMI, diabetes and hypertension — as well as sleep-related characteristics such as sleep apnea severity, arousals and sleep efficiency.

Not every outcome was predictable with equal confidence. The scientists report that the model successfully predicted atrial fibrillation, heart failure and death from any cause, but additional optimization is needed before it can predict myocardial infarction and stroke.

Why does one ECG lead matter?

The method relies on just one ECG lead — a single electrical signal — rather than the usual 12-lead hospital setup. That difference matters for practicality. A single lead can be captured with simple, low-cost adhesive patches during one night of sleep, making the approach highly accessible and easy to add to routine testing.

"Importantly, we found that adding the score from our deep learning model improved prediction of cardiac outcomes beyond established risk factors," Clifford said. "Adding these scores into sleep studies has great potential for improving early detection of cardiovascular disease, determining who is most at risk and being more proactive in patient care."

What could this change in the clinic?

NIH officials see a screening role for the technology. "This new approach has the potential to identify individuals at risk for cardiovascular disease years before clinical symptoms arise," said David Goff, M.D., Ph.D., acting director of NIH's National Heart, Lung, and Blood Institute (NHLBI).

Goff cautioned that validation work remains. "Evaluating whether this information improves traditional risk prediction is an important next step," he said. "In the future, proven analytical approaches like this could be integrated into existing diagnostic tests for patients undergoing sleep studies as a screening tool for multiple adverse cardiovascular outcomes."

What are the limitations?

The findings are promising but preliminary in several respects. The model still needs optimization for two of the five outcomes studied — heart attack and stroke. And although the researchers tested it on thousands of patients across three hospitals in two cities, all data sets came from U.S. academic medical centers, and outcomes were drawn from electronic health records rather than dedicated long-term follow-up.

The study did not test whether acting on the model's scores actually prevents cardiac events. That would require prospective clinical trials, which the researchers and NIH identify as the logical next step before the tool enters routine care.

Still, the core conclusion is encouraging: a signal that hospitals already record during millions of sleep studies, and largely discard, carries information about a patient's cardiovascular future. If the approach withstands further testing, it could turn a one-night diagnostic test into an early-warning system for heart disease — at almost no added cost.

The paper, by Zuzana Koscova and colleagues, appears in SLEEP (2026), DOI: 10.1093/sleep/zsag229.