A multidisciplinary research team has developed an AI foundation model capable of identifying long-term health risks, including heart disease and cognitive decline, from routine sleep study data. Published in Nature Communications on August 3, 2026, the study reveals that AI can detect hidden physiological patterns that traditional clinical measures often overlook.
For decades, clinicians have relied on a small subset of data to evaluate patients in sleep labs. While an estimated 1 to 4 million polysomnograms are performed annually in the United States, the focus has historically remained on grading sleep apnea severity. This approach, according to researchers, ignores a wealth of information regarding how a patient’s brain, lungs, muscles, and heart behave during sleep.
The Failure of the Apnea-Hypopnea Index
The standard clinical tool for assessing sleep apnea is the apnea-hypopnea index (AHI), which tracks the average number of breathing pauses and shallow breathing events per hour. However, the new AI model suggests that AHI is a limited predictor of long-term outcomes. The research found that standard AHI severity categories demonstrated no predictive value in certain contexts, with high-risk patients scattered across different AHI strata.
The AI model’s ability to stratify risk proved superior to the AHI in several key ways. Most notably, the model predicted outcomes with equal accuracy for both men and women, whereas the AHI has historically performed better in men. By moving beyond simple summaries, the AI identified clinically meaningful
subtypes of patients whose health trajectories differed sharply despite having similar AHI scores.
Five Risk Categories and Mortality Predictions
Using data from the Cleveland Clinic Sleep Signals, Testing, and Reports Linked to Patient Traits (STARLIT) registry, researchers grouped patients into five distinct risk categories. The implications for patient survival are stark: individuals in the highest-risk group faced a 100 percent increase in five-year mortality risk compared to those in the lowest-risk tier.
This risk stratification was not limited to death. The model showed strong associations with incident cardiovascular, neurologic, and psychiatric outcomes. According to data published in Nature, the highest-risk group (RG5) exhibited an elevated incidence of these conditions, a finding that was independently confirmed in a nationwide patient cohort.
The Cleveland Clinic-IBM Discovery Accelerator
The model is the product of a 10-year research partnership between Cleveland Clinic and IBM known as the Discovery Accelerator. This program integrates AI and quantum computing to speed up discoveries in the life sciences.
The research suggests that the “hidden” signals AI can extract from a single night of sleep are effectively biomarkers for chronic disease. Because these signals are invisible to the human eye, they have remained underutilized in standard medical practice.
“For decades we have distilled an overnight sleep study into a handful of summary measures,” said sleep medicine specialist Dr. Reena Mehra, “AI gives us the opportunity to move beyond those summaries and learn from the full richness of sleep physiology.”
Dr. Reena Mehra, professor of medicine at the University of Washington School of Medicine and the study’s senior author
Future Clinical Application and Validation
The goal is to transition the sleep study from a simple diagnostic test for apnea into a comprehensive prognostic tool. With nearly 70 million Americans living with chronic sleep and wakefulness disorders, a more personalized approach to sleep medicine could allow for earlier intervention in cardiovascular and neurological health.
However, the researchers emphasize that this is an early step. The next phase involves validating these findings across more diverse populations and expanding collaborations with industry partners and professional societies. While the AI can identify risk, the clinical response remains focused on foundational health habits.
Dr. Reena Mehra noted that as these methods continue to be validated in prospective studies, they have the potential to transform the sleep study from primarily a diagnostic test into a richer source of information about an individual’s future health and may accelerate discoveries about the relationships between sleep physiology and chronic disease.
The broader implication of the study is that routine medical tests across the board may contain substantially more physiologic information than current clinical practices are equipped to extract, potentially transforming how preventative healthcare is delivered.
Keep reading
