Sleep is essential to health, yet disrupted sleep is common across many diseases. Evidence increasingly suggests that changes during sleep can appear before a disease receives a diagnosis.
Rahul Thapa helped lead a Stanford University group that developed an artificial intelligence model to estimate the risk of several diseases from recordings taken during one night’s sleep. The system, SleepFM, examines polysomnography, a sleep examination that records brain, cardiac, muscular and respiratory activity during rest.
The Stanford team trained SleepFM on over 585,000 hours of data from approximately 65,000 participants. It achieved a concordance index of at least 0.75 across 130 clinical conditions, including 0.85 for dementia and 0.81 for heart attack.
A separate team led by Erhan Bilal took a similar approach using recordings from Cleveland Clinic. Its analysis used over 10,000 sleep examinations, while electronic medical records tracked health information over more than one decade.
Bilal’s model placed patients in five risk groups with differing patterns of mortality and cardiovascular and neurological disease. The group at highest risk had a risk of death more than double that of the lowest-risk group.
The researchers also verified Bilal’s system using independent data from the Sleep Heart Health Study.
The two studies found that artificial intelligence can draw on signals from a single night’s sleep that conventional sleep measures miss, particularly when it brings together brain, heart and respiratory activity.
Rachel K. Rowe, an author addressing this sleep research, is affiliated with the Department of Integrative Physiology at the University of Colorado Boulder in Boulder, Colorado 80301, USA.
The findings suggest that a night of sleep may hold early clues about later disease and mortality risk. Whether these predictions change clinical care remains to be established.
Comments (0)