Jingsong Luo and five colleagues at Peking University and Capital Medical University in Beijing derived sleep stages from those movements. The study was published open access in PLOS Medicine on 17 September 2026. REM sleep is the stage of rapid eye movements and the most vivid dreams. Deep sleep is the slowest stage and the hardest to break.
The wrist cannot see sleep stages. A deep-learning algorithm called SleepNet estimates them. It was trained on more than 1,100 nights recorded with both an accelerometer and polysomnography, the laboratory test that measures sleep with electrodes on the head, eyes and muscles. In telling apart wake, REM and non-REM sleep the algorithm reaches an F1 score of 0.49, on a scale where 1 is perfect agreement. Compared with polysomnography it underestimates REM sleep by 17 minutes and overestimates total sleep by 48.
The authors followed participants for a median of 8.9 years and searched hospital discharge records for 1,049 diseases. With a strict correction for multiple comparisons they found 156 associations. Each 47.6 more minutes of REM sleep went with fewer diagnoses of 83 diseases. For deep sleep, per 47.5 more minutes, the count is 7. The analyses account for age, sex, deprivation, education, weight, smoking, alcohol, physical activity and other factors.

Results are given as hazard ratios, the ratio between the rates at which a diagnosis appears in two groups. For heart failure the hazard ratio linked to REM sleep is 0.74. For dementias it is 0.54. For parkinsonism, the group of movement disorders that resemble Parkinson’s, it is 0.20, with a 95 per cent confidence interval from 0.11 to 0.39. Deep sleep goes with fewer diagnoses of type 2 diabetes, 0.89, major depression, 0.86, and Parkinson’s disease, 0.70.
The PLOS press release, republished by ScienceDaily on 23 September 2026, gives REM sleep a hazard ratio of 0.20 for Parkinson’s disease. In the paper the 0.20 refers to parkinsonism. Parkinson’s disease appears under deep sleep, with a hazard ratio of 0.70.
The main text does not report how many people developed each disease. Diseases with fewer than 20 cases were left out. Without case numbers a hazard ratio cannot say how many diagnoses would change with more REM sleep.

Hazard ratios, with 95 per cent confidence intervals, for selected diseases per 47.6-minute increase in REM sleep (blue) and 47.5-minute increase in deep sleep (orange), estimated with a wrist accelerometer in 95,559 UK Biobank participants. Log scale, the vertical line at 1 marks no difference. Chart produced by ScienceOnline from data in Luo J. et al., PLOS Medicine 2026, doi 10.1371/journal.pmed.1005213 (article licensed CC BY). Original newsroom chart.

The authors list eight limitations. The study is observational and cannot establish cause. Sleep was measured for a single week. Other characteristics of the participants had been collected on average 5.7 years earlier. Diagnoses come from hospital admissions only. UK Biobank volunteers are healthier, wealthier and more often of European ancestry than the British population.
The heaviest limitation concerns the direction of the link. Neurodegenerative diseases begin years before diagnosis, and sleep can change in that phase. When diagnoses in the first two years are excluded, 95 of the 156 associations remain significant, about six in ten. The paper does not say which of the others were lost.
The study was funded by the National Natural Science Foundation of China and Beijing programmes for young researchers. The authors declare no competing interests and share their code in a public repository.
Short sleep went the other way. Under five hours a night, compared with six to eight, accounted for 37 of the 41 associations with higher risk.
The sleep is one week’s worth, read at the wrist between 2013 and 2015. The diagnoses came later, in hospital.

References
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, 23, 9, e1005213, published 17 September 2026, open access under CC BY. doi: 10.1371/journal.pmed.1005213
PLOS, press release, republished by ScienceDaily on 23 September 2026 as «More REM sleep linked to lower risk of 83 diseases» (secondary source).