Sleep study heart records link to ten-year cardiovascular risk
Across two validation cohorts of over 22,000 patients, an AI score from overnight heart traces yielded hazard ratios of 2.03 and 2.72 for atrial fibrillation.
In an observational study of hospital patients who underwent overnight sleep studies, researchers evaluated an artificial-intelligence model to predict cardiovascular events. The team finetuned a deep neural network on 15,809 patients from one hospital and tested it in two external groups of 9,810 and 12,576 patients. The model analysed single-lead electrocardiograms, which record the electrical activity of the heart, combined with sleep-stage data. Higher model scores were associated with greater ten-year risks of atrial fibrillation (an irregular heartbeat), stroke, heart failure, and death from any cause. These links remained after adjusting for age, sex, smoking, high blood pressure, diabetes, and sleep measures. For heart attacks, the score was linked to higher risk in one test cohort, but the range of likely values included no difference in the other.
Why it matters
Cardiovascular disorders and abnormal heart rhythms become much more frequent as people age, often alongside disrupted sleep. Using existing, routinely gathered sleep recordings to identify long-term cardiac risk could help clinicians spot vulnerable individuals earlier without extra testing.
Caveats
The study was observational and relied on diagnostic codes in electronic health records to identify cardiovascular outcomes. In addition, the findings come from hospital patients who underwent polysomnography, so they may not apply to the wider public.
The paper
Prediction of cardiovascular outcomes using electrocardiography from overnight polysomnography
Show 12 more authors
Neal K Bhatia, Ali Bahrami Rad, Reza Sameni, Haoqi Sun, Niels Turley, Umakanth Katwa, Dennis Hwang, Katie L Stone, Lynn M Trotti, Robert J Thomas, Emmanuel Mignot, Brandon M Westover,Emory University
Sleep, 9 Oct 2026