Artificial intelligence can now extract far more from a standard overnight sleep test than ever before, potentially identifying patients at sharply higher risk of heart disease, cognitive decline, and early death. A new study published in Nature Communications shows that an AI model, developed by researchers including those at Cleveland Clinic, uncovered hidden patterns in routine sleep study data that traditional measurements completely miss.
The findings center on polysomnograms, in-lab sleep tests performed an estimated 1 to 4 million times each year in the United States. These tests collect rich data on brain activity, breathing, muscle movement, and heart function, yet clinicians have historically boiled that information down to a single summary score, the apnea-hypopnea index, to grade sleep apnea severity. The new AI model, by contrast, learns from the full complexity of a night’s sleep.
Using data from Cleveland Clinic’s STARLIT registry, the research team grouped patients into five distinct risk categories. Those in the highest-risk group faced twice the mortality risk over the next five years compared with the lowest-risk group, a difference that the standard apnea-hypopnea index failed to capture. The model also predicted outcomes equally well for men and women, whereas the traditional measure has historically performed better in men. The results were independently confirmed in a nationwide patient cohort.
“For decades we have distilled an overnight sleep study into a handful of summary measures,” said Reena Mehra, M.D., professor of medicine at the University of Washington and the study’s senior clinical author. “AI gives us the opportunity to move beyond those summaries and learn from the full richness of sleep physiology.” The model was developed through the Discovery Accelerator, a 10-year research partnership between Cleveland Clinic and IBM focused on applying AI and quantum computing to life sciences.
The implications extend beyond sleep medicine. The approach demonstrates that routine medical tests may contain substantially more physiologic information than current practice extracts. By detecting latent features invisible to the human eye, the AI model can stratify risk for cardiovascular and neurologic disease, and survival, opening the door to earlier and more personalized care. Nearly 70 million Americans live with chronic sleep disorders, according to Matheus Lima Diniz Araujo, Ph.D., a sleep researcher at Cleveland Clinic, and this discovery could expand the value of every sleep test performed.
What Happens Next
Researchers stress that validation is the next critical step. Carl Saab, Ph.D., chief scientist of the Discovery Accelerator, said the team plans to confirm these findings in diverse populations and expand collaborations with medical experts, industry partners, and professional societies. “Because everyone sleeps, sleep studies offer a remarkable window into human health that extends far beyond the diagnosis of sleep disorders,” said lead author Erhan Bilal, Ph.D. “Our work shows how foundation models can begin to unlock the richness of these complex signals. And this is only the beginning.”
For patients, the promise is a future where a single night of sleep could reveal much more than whether they stop breathing during the night, potentially flagging hidden risks years before they become clinical problems. The study team hopes that within the next few years, this AI approach will move from research labs into standard clinical practice, making every sleep study a more powerful tool for preventive health.