Higher REM sleep linked to lower risk of 83 diseases in large study

Researchers analyzed a week of wrist-worn accelerometer data from 95,559 UK Biobank participants, then used an algorithm called SleepNet to estimate how much time each person spent in REM sleep and other sleep stages. They followed participants for a median of 8.9 years, tracking who developed any of 1,049 different health conditions. After statistical correction for testing so many outcomes at once, the researchers found 156 significant links between sleep patterns and disease risk.

Higher amounts of REM sleep were associated with lower risk of 83 of those diseases, and more deep sleep was linked to lower risk of 7 more. In the other direction, greater night-to-night irregularity in sleep timing was linked to higher risk of 3 diseases, and more time spent awake after initially falling asleep was linked to higher risk of 6. The study also found that sleep duration relates to disease risk in a curved, not straight-line, way: for 86 disease outcomes, the relationship between total sleep time and risk was nonlinear, and for 69 of those the lowest risk fell within a 6-to-8-hour window. People sleeping less than 5 hours accounted for 37 of the 41 significant risk increases identified when comparing extreme sleep durations to that 6-to-8-hour reference group.

The traditional way to measure REM sleep is an overnight polysomnogram, a test that records brain activity, eye movement, muscle tone, breathing, and heart rhythm simultaneously while a person sleeps in a clinic. It is treated as the gold-standard method, but it is costly, technically demanding, and impractical for studying tens of thousands of people at once. Wrist-worn accelerometers, in contrast, are cheap and can be worn at home for days at a stretch, which is why the UK Biobank could gather sleep data from such a large group.

The tradeoff is that accelerometers only measure movement, not brain waves directly, so translating raw movement data into sleep stages such as REM requires a computational model. The researchers used SleepNet, a machine-learning system that was checked against 1,113 nights of laboratory polysomnography recordings before being applied to the accelerometer data. That validation step is what allowed the researchers to estimate REM, light sleep stages called N1 and N2, and deep sleep (N3) from wrist movement alone, across a group far larger than any polysomnography study could practically enroll.

The Biobank findings sit alongside several polysomnography-based studies that also tie reduced REM sleep to worse outcomes. A Stanford doctoral dissertation combined three separate cohorts — the Osteoporotic Fractures in Men study, the Wisconsin Sleep Cohort, and the Sleep Heart Health Study — and found that every 5 percentage-point drop in the share of sleep spent in REM was associated with roughly 13% higher all-cause mortality, with similar effects for death from cardiovascular disease, cancer, and other causes. A meta-analysis across the three cohorts produced an overall hazard ratio of 1.13, and people with less than 15% REM sleep had consistently higher mortality than those above that threshold. The sex-specific results did not agree across cohorts, however: in the Wisconsin cohort the REM-mortality link was significant in women but not men, while in the Sleep Heart Health Study it was significant in men but weaker in women.

A separate analysis of the RICCADSA trial looked specifically at 356 people with coronary artery disease and obstructive sleep apnea, a condition in which breathing repeatedly stops or is partly blocked during sleep. Patients in the lowest quarter of REM sleep, corresponding to REM making up less than 8.7% of total sleep time, had a mortality rate of 12.8% over a median follow-up of 4.7 years, compared with 4.4% among those with more REM sleep. After adjusting for age, sex, body mass index, and CPAP treatment assignment, reduced REM sleep still predicted more than double the risk of death. A separate deep-learning study that estimated a person's biological age from sleep-study recordings noted that decreased REM sleep amounts have previously been associated with all-cause mortality, consistent with this broader pattern.

All of these studies are observational, meaning they can show that low REM sleep tracks with disease and death but cannot prove that low REM sleep causes either. The Biobank researchers stated plainly that their design

precludes causal inference

and remains open to residual confounding, factors that affect both sleep and disease risk without one causing the other. The Stanford dissertation raised a specific alternative explanation: reduced REM sleep might be a marker of accelerated brain aging or frailty rather than a direct risk factor for death, meaning it could be a signal of an underlying process rather than a cause in its own right.

A separate line of research on REM sleep behavior disorder, a condition where the muscle paralysis that normally accompanies REM sleep fails and people act out their dreams, illustrates how much remains uncertain even where a link is well established. That disorder is described as a potential early warning sign of diseases involving a buildup of a protein called alpha-synuclein, including Parkinson's disease, and can precede those conditions by decades. But researchers reviewing that link note it remains unclear whether REM sleep behavior disorder is connected to these diseases through shared protein buildup or through a shared anatomical pathway in the brain. Across all these studies, the sex differences, the unresolved question of causation, and the open question of mechanism mean the disease associations reported in the Biobank study are described by its authors as a foundation for future hypothesis-testing rather than a finished explanation.