Researchers using an artificial intelligence model have pulled a warning signal out of routine overnight sleep tests that the standard clinical score does not see. The study, published August 3, 2026, in Nature Communications, came out of the Cleveland Clinic-IBM Discovery Accelerator and was led by clinicians at Cleveland Clinic and the University of Washington School of Medicine.
The AI sorted patients into five groups based on physiological patterns during sleep. People in the highest-risk group were twice as likely to die within five years as those in the lowest-risk group. That gap was invisible in the apnea-hypopnea index, the number most patients are handed after a sleep test.
What the standard score misses
The apnea-hypopnea index, or AHI, counts how many times per hour a person stops breathing or breathes too shallowly during sleep. According to the Cleveland Clinic's patient information, fewer than five events per hour is considered normal, and 30 or more is classed as severe sleep apnea. It is the yardstick used to decide who gets treated and how aggressively.
The new analysis suggests that single number leaves a lot on the table. The AI looked at fuller sleep physiology, including breathing rhythms, oxygen levels, heart signals and brain activity, and found groupings of patients that a raw event count could not capture.
Also read: One in three women don't know if they're in perimenopause – here are the signs
"AI gives us the opportunity to move beyond those summaries and learn from the full richness of sleep physiology," said Dr. Reena Mehra of the University of Washington School of Medicine, the study's senior clinical author.
Signals of heart disease and cognitive decline
The five patient subtypes the model identified were linked not only to mortality but also to heart disease and cognitive decline over the following years. The model was trained on Cleveland Clinic's STARLIT registry of sleep-study data and validated in a separate nationwide cohort.
The team also reported that the AI predicted outcomes about equally well in men and women. That is a notable contrast to the AHI, which has historically been better at flagging risk in men than in women.
Jeffrey Rogers, PhD, the study's corresponding author, said modern AI can "recover much more of the information contained in a night's worth of sleep physiology, revealing clinically meaningful patient groups with very different long-term health risks."
Also read: This is how many calories you can burn in a year by walking every day
What it means for people who have had a sleep test
Between one and four million polysomnograms are performed each year in the United States, mainly to check for sleep apnea. The researchers argue that the raw data from those tests already holds information about long-term health that is being routinely thrown away.
This is one study, and the tool is not yet used in ordinary clinics. Nothing in the paper changes what current AHI results mean for treatment decisions today, and patients should not adjust CPAP therapy or other prescribed care based on it. Anyone worried about sleep, breathing at night or symptoms such as loud snoring, morning headaches or daytime exhaustion should raise it with their doctor.
For now, the finding is a marker of where sleep medicine is heading. As lead author Erhan Bilal, PhD, put it, "because everyone sleeps, sleep studies offer a remarkable window into human health that extends far beyond the diagnosis of sleep disorders."
Also read: Apple Watch flagged heart failure decline about a week before hospital care
Also read: Early warning signs in your mouth that could signal cancer