Can a Sleep Study ECG Predict Heart Disease? What the NIH Study Found
At a glance: On October 9, 2026, the U.S. National Institutes of Health highlighted a study that paired the ECG recorded during an overnight sleep study with sleep-stage information. The model added useful discrimination for future atrial fibrillation and heart failure in the studied patients. It does not diagnose heart disease or establish a new reason for everyone to book a sleep study.
What did the researchers test?
The U.S. team used a single-lead ECG already recorded during polysomnography for people referred for sleep evaluation. They fine-tuned the model with records from 15,809 Massachusetts General Hospital patients and tested it in two other hospital cohorts: 9,810 patients at Emory University Hospital and 12,576 at Beth Israel Deaconess Medical Center. Outcomes came from electronic health records, chiefly diagnosis codes.
Where did the model add information?
When its ECG-derived score was added to established clinical and sleep measures, discrimination improved for atrial fibrillation, heart failure and death from any cause. The paper’s decision analyses did not show an added clinical benefit for stroke or myocardial infarction. “Cardiovascular outcomes” in the headline should therefore not be read as one equally strong prediction for every condition.
Does this change care today?
This was a retrospective study among U.S. patients who had already undergone sleep testing. It did not test whether acting on an AI score prevents an event, and it did not establish performance in the general population or in South Korea. It is a promising use of an existing recording, not a new screening recommendation, a diagnosis, or a change in Korean coverage or clinical rules.
What are the limits?
Most events were identified from health-record codes rather than direct clinical confirmation. The mortality outcome included deaths from all causes. Undiagnosed atrial fibrillation already present during the sleep study could sometimes resemble a prediction of future disease; the authors used exclusions and targeted ECG review to reduce this concern. Further prospective testing and assessment of decisions based on the score are needed.
What can a reader take away?
If an existing sleep-study report flags an ECG finding, discuss that finding in the context of symptoms and medical history with the care team. This study alone is not a reason to order a test or change medication. Standard evaluation of blood pressure, diabetes, smoking and other established risks still matters.
Sources and review
NIH, Deep learning model using ECGs during sleep studies can predict cardiovascular outcomes (October 9, 2026); Koscova Z et al., Prediction of cardiovascular outcomes using electrocardiography from overnight polysomnography, Sleep (October 9, 2026). Written by Life Choice Note as independent health-information commentary. Reviewed October 11, 2026.
This research explainer does not provide an individual diagnosis or treatment instruction.