Editorial NexoScience
A plastic band, about the size of a watch, strapped to the wrist for one week. Between 2013 and 2015, nearly a hundred thousand British adults wore one, went about their lives and mailed it back. Almost nine years later, those seven nights of wrist motion line up with a surprising map of who got sick, and with what.
The map comes from a study published Sept. 17 in the journal PLOS Medicine by a team led from Capital Medical University in Beijing, including Shengzhi Sun and Rui Chen. It’s one of the widest looks yet at REM sleep and disease risk. The researchers took accelerometer data from 95,559 people in the UK Biobank, a long-running health study of middle-aged and older adults, and checked it against 1,049 diseases that later showed up in hospital records. The headline link: people who spent more time in REM sleep, the stage when most vivid dreaming happens, had lower rates of 83 different diseases. Heart failure, dementia and Parkinson’s were among them.
The tracker never read a single brainwave. In a sleep lab, stages are scored from electrodes on the scalp, a test called polysomnography. That’s accurate, but it’s expensive and nobody does it on a hundred thousand people. Here the team used SleepNet, a deep-learning algorithm trained on more than 1,100 nights where people wore a wrist sensor and lab electrodes at the same time. It learned to guess sleep stages from movement alone. Think of it as working out what’s happening inside a house from the creak of the floorboards. You can tell a lot. You can’t tell everything. From those guesses the team pulled six measures: total sleep, REM, deep sleep, light sleep, time awake after first falling asleep, and how much nightly sleep swung from one day to the next.
REM sleep came out on top. The typical participant got about 82 minutes of REM a night. For every extra 47.6 minutes, roughly the gap between a short-REM sleeper and a long one in this group, the risk of dementia was 46% lower and the risk of heart failure 26% lower, after adjusting for age, sex, weight, smoking, drinking, exercise, screen time, income, air pollution and even nighttime noise. Parkinsonism showed the most dramatic number, an 80% lower risk, though it rested on few cases and the uncertainty around it was wide. Across the board, REM-linked diseases spanned 12 body systems, from the heart to the brain to metabolism.
The REM signal wasn’t just a stand-in for sleeping longer. That was the obvious objection, and the team tested it. When they looked at REM as a share of each person’s total sleep, instead of raw minutes, a bigger REM slice still went with lower risk of 54 diseases. A bigger slice of light sleep went the other way, with higher risk of 28.
Deep sleep did quieter work. More deep sleep was tied to lower risk of just 7 conditions, a far shorter list than REM’s. But the list is telling: type 2 diabetes, major depression, sleep apnea and Parkinson’s disease, which was 30% less common per extra 47.5 minutes of deep sleep.
Broken nights told a different story. Being awake for long stretches after first nodding off was linked to higher risk of 6 conditions, including alcohol misuse and drug dependence. Irregular sleep, where one night is short and the next is long, was linked to anxiety, depression and abdominal pain. Here’s the thing. Two people can log the same total hours and end up on opposite sides of these lists, depending on how choppy those hours are.
Six to eight hours kept turning up. Total sleep didn’t follow a straight line. For 86 diseases, risk dipped and then rose again, a U-shaped curve. And for 69 of those, the low point fell between six and eight hours a night. That’s a bit shorter than the seven to nine hours often recommended, something the authors pointed out themselves. People in the six-to-eight-hour band also had the highest share of REM of any group.
Under five hours, the warnings stacked up. When the team split sleepers into five bands, they found 41 diseases with clearly higher risk than in the six-to-eight-hour group. Thirty-seven of those 41 were in people sleeping less than five hours. Dementia risk in that group was about two and a half times higher. Long sleep was a different case. Sleeping more than eight hours was tied mainly to depression, and the authors noted that oversleeping is a well-known early symptom of depression, so the arrow may point backward.
Why sleep would matter this much isn’t settled. For short sleep, the authors point to known pathways: a nervous system stuck in fight-or-flight mode, stress hormones running out of rhythm, and low-level inflammation spread through the body. For REM itself, the study doesn’t offer a mechanism. It wasn’t built to. It’s a map, not an explanation.
The caveats are big ones. This is an observational study, and the authors said plainly that it can’t show sleep causes or prevents any disease. Honestly, the most striking caveat is the one they tested directly. Illness can wreck sleep long before a diagnosis, so the team re-ran the analysis after dropping any disease that appeared in the first two years. Only 95 of the original 156 links survived, about 60%. Some of what looks like poor sleep raising disease risk may simply be disease, quietly getting started, ruining sleep. The outcomes also came only from hospital records, so milder conditions treated by a family doctor were missed. And the UK Biobank skews healthy, white and well-off: 96.9% of this group identified as white.
A wrist is not a sleep lab. SleepNet’s agreement with lab-scored stages was modest. Compared with polysomnography, it underestimated REM by about 17 minutes and overestimated total sleep by about 48. The researchers argue these errors are spread evenly across people, so rankings still hold, but they were careful to call the results “movement-based estimates” rather than true sleep stages. Sleep was also measured for only one week, and other details about participants were recorded on average 5.7 years before the tracker went on.
What the map is really for. The authors describe the work as hypothesis-generating, a list of leads. The next step is to test those leads one by one, with studies built to find out whether boosting REM or steadying a ragged sleep schedule changes anything at all. For now, the strangest thread on the map is also the simplest: a week of wrist wiggles, recorded years ago, still seems to know something about the brain’s future. For more on that side of the story, see our brain health evidence hub.
Source: Luo, J., Liu, R., Yin, J., Cao, W., Sun, S., Chen, R. “Accelerometer-derived real-world sleep stages and risk of incident diseases: A UK Biobank cohort study and phenome-wide association analysis.” PLoS Medicine (2026). DOI: 10.1371/journal.pmed.1005213
This article summarizes published scientific research for general information. It is not medical advice and does not replace consultation with a qualified health professional.
