Daily rhythm and metabolic syndrome: a study of 272 adults
Study of 272 adults with a wrist device: metabolic syndrome was linked more to heart rate rhythms than to sleep. An association, not a cause.
Blog · Sleep
Sleep and disease risk: Stanford’s SleepFM model
Model trained on 585,000 hours of polysomnography: from one recorded night it estimates the risk of 130 conditions. A group finding.
In 2026, researchers at Stanford University published a study in Nature Medicine on SleepFM, an artificial intelligence model that reads sleep recordings.
The model was trained on more than 585,000 hours of polysomnography, the full sleep recording made in a laboratory, from about 65,000 participants. From a single night of recording it estimates the future risk of 130 clinical conditions, with an accuracy (C-Index) of at least 0.75.
What the research found
SleepFM analyses the physiological and temporal structure of sleep: the hours slept, the sequence of phases, their relative duration and the regularity of the cycles. From these patterns the model builds representations that are associated with future disease risk.
Among the 130 conditions, the abstract lists for example all-cause mortality (C-Index 0.84), dementia (0.85), myocardial infarction (0.81), heart failure (0.80), chronic kidney disease (0.79), stroke (0.78) and atrial fibrillation (0.78).
585,000 hours of recordings
That is more recordings than any person could read: the model draws out patterns that are not visible at first sight. It does not replace the doctor.
What remains
Everyone sleeps, and many people already record their sleep with a device. SleepFM shows that a full recording of one night holds information on the risk of many diseases. It is a research finding about a group of people: it is not a forecast about you.
The Digital Twin and this research
A ring or a watch that measures sleep does not perform polysomnography: it records fewer signals. SleepFM was built on laboratory recordings, not on that data.
The Digital Twin lines up the data your devices write to your phone, such as steps, heart rate and sleep, and shows you how they change over time. It does not use SleepFM, it does not calculate risk profiles and it does not diagnose: this research describes groups of people, not you. These are data to read together with your doctor.
The source is at the bottom. Open it.
The source
This summary starts from research by others. It is not medical advice for you: talk it over with your doctor.
Their study. Your data.
Related reading
Early Parkinson’s and REM disorder: a study of 413 people
Study of 413 people with early Parkinson's: probable REM sleep behaviour disorder was linked to faster cognitive decline. An association, not a cause.
Irregular sleep and the heart: a study of 1,992 adults
Study of 1,992 adults: highly variable sleep duration and timing were linked to about double the cardiovascular risk. An association, not a cause.
More studies, same rule: read the source.