Researchers have developed an artificial intelligence model that uses data collected during routine overnight sleep studies to identify patients at elevated risk of cardiovascular disease, cognitive decline and death.
The model analysed physiological signals that are typically collected during sleep studies but are not fully represented by the summary measures commonly used in clinical practice. These studies record information about brain, heart, lung and muscle activity, but are primarily used to assess conditions such as sleep apnoea.
The research team found that AI could identify hidden patterns within these signals and divide patients into five groups with substantially different long-term health outcomes. Patients placed in the highest-risk group had twice the risk of death over the following five years compared with those in the lowest-risk group.
These differences were not captured by the apnoea-hypopnoea index, the standard measure used to assess sleep apnoea severity.
The model was developed by sleep physicians, data scientists, neuroscientists and AI researchers working through the Cleveland Clinic–IBM Discovery Accelerator, a 10-year research partnership focused on applying AI and quantum computing to life sciences.
Researchers trained the model using data from the Cleveland Clinic Sleep Signals, Testing, and Reports Linked to Patient Traits registry. The system predicted outcomes consistently among men and women, while the apnoea-hypopnoea index has historically shown stronger predictive performance in men.
The findings were independently validated using a nationwide patient cohort.
Reena Mehra, professor of medicine at the University of Washington School of Medicine and senior author of the study, said conventional sleep assessments have historically reduced an overnight study to a limited number of summary measurements.
She said AI offered a way to analyse the broader range of physiological information contained in sleep data and identify clinically meaningful patterns that conventional assessments may overlook.
An estimated one to four million sleep studies are conducted in the United States each year, primarily to assess sleep apnoea. The researchers said the approach could eventually expand the use of these tests from diagnosing sleep disorders to supporting broader health-risk assessment.
The team said prospective clinical studies will be required to validate the model further and assess how its risk classifications could support earlier and more personalised care.
