REM and deep sleep stages tied to lower disease risk across 83 conditions in UK Biobank study

A large cohort study of 95,559 adults drawn from the UK Biobank has found that the structure of a person's sleep, meaning how much time is spent in rapid eye movement (REM) sleep and in deep sleep, is tied to the risk of later developing a wide range of diseases, and not only to how many hours a person sleeps. The peer reviewed paper appeared in PLOS Medicine on September 17, 2026 under the title Accelerometer-derived real-world sleep stages and risk of incident diseases: A UK Biobank cohort study and phenome-wide association analysis.

The authors are Luo J, Liu R, Yin J, Cao W, Sun S and Chen R. The journal assigned the article the citation PLoS Med 23(9): e1005213 and the DOI 10.1371/journal.pmed.1005213, publishing it open access after receiving it on November 26, 2025 and accepting it on July 29, 2026. The paper describes a peer reviewed analysis linking objectively measured sleep stages to the incidence of 1,049 health outcomes.

The UK Biobank is a resource that stores biological, genetic and lifestyle information on 500,000 volunteer participants in the United Kingdom, and the new analysis drew its 95,559 people from that pool. Each participant wore a wrist accelerometer for seven consecutive days and nights. A deep learning algorithm called SleepNet then derived real world sleep metrics from those recordings: the REM, N1, N2 and N3 stages, total sleep duration, night to night sleep irregularity and wakefulness after sleep onset, which researchers abbreviate as WASO.

Participants were followed from the moment they wore the accelerometer until the first recorded disease, death, or April 1, 2024, giving a median follow up of 8.9 years. The team used Cox proportional hazard regression, phenome wide association and restricted cubic spline analyses to map the sleep measures against 1,049 health outcomes. Within that design, 156 sleep and disease associations survived Bonferroni correction, a strict statistical threshold meant to limit false positives. REM sleep is the stage during which most dreaming happens, while deep sleep is often described as the most restorative stage.

Key Facts

PLOS Medicine reported on September 17 that higher REM sleep was associated with lower risks for 83 diseases, while higher deep sleep was associated with lower risks for 7 diseases. Greater sleep irregularity and increased wakefulness after sleep onset moved in the opposite direction, linked to elevated risks for 3 and 6 diseases respectively.

Medical Xpress reported on September 17 that the REM sleep finding was measured per interquartile range of 47.6 minutes. Within that analysis, the hazard ratio for heart failure was 0.74, for dementia 0.54 and for Parkinson's disease 0.20. Greater deep sleep was linked to a lower risk of seven conditions, including type 2 diabetes and major depressive disorder.

The Independent reported on September 18 that the pattern extended across the full list of outcomes. More REM sleep was associated with lower risk of heart failure, dementia and Parkinson's, while higher sleep irregularity and wakefulness after sleep onset tracked with higher risk of conditions such as anxiety and substance use disorders. Participants who regularly slept six to eight hours carried the lowest risk of several diseases, and those who slept more than eight hours were also at risk.

Sleep duration behaved in a nonlinear way across the phenome. The researchers found significant nonlinear relationships between sleep duration and 86 disease phenotypes, and the minimum risk duration for 69 of them sat inside a 6 to 8 hour window. People sleeping less than five hours showed the most widespread vulnerability, accounting for 37 of the 41 significant adverse associations when compared with the 6 to 8 hour reference group.

Medical News Today reported on September 18 that participants had an average age of 56.2 years. Increased REM sleep at baseline was associated with reduced risk of more than 80 health conditions including heart failure, hypertension, atrial fibrillation, ischemic heart disease, parkinsonism, multiple sclerosis and dementias, while increased deep sleep was associated with reduced risk of seven diseases including type 2 diabetes, major depressive disorder, sleep apnea and Parkinson's disease. A higher wakefulness after sleep onset score was associated with increased risk of six conditions including psychoactive substance dependence, alcohol abuse and osteoarthritis, and irregular sleep was linked to abdominal pain, anxiety and major depressive disorder.

Analysis

HealthDay reported on September 18 that a team led by Jingsong Luo of Peking University tracked more than 95,000 adults who wore wrist monitors for seven days and nights. HealthDay also noted that the PLOS Medicine press release attributes the study to corresponding author Shengzhi Sun of Capital Medical University, while HealthDay and The Independent name first author Jingsong Luo of Peking University, reflecting a multi institution author team. That discrepancy matters less than the direction of the results, which is consistent across the coverage.

What this really means is that the sleep conversation, which has generally been organised around counting hours, may need to be reorganised around what happens inside those hours. Two people can both log seven hours and end up with very different amounts of REM and deep sleep, and the study suggests those differences track with meaningfully different disease profiles. The familiar public health message of seven to nine hours is nudged sideways here; Medical News Today noted that the researchers concluded six to eight hours is optimal, slightly narrower than the more commonly quoted range.

The effect sizes deserve to be read carefully. Hazard ratios of 0.74 for heart failure and 0.54 for dementia per interquartile range increase in REM sleep are not trivial, yet they come from an observational design. The authors themselves stress that the observational structure precludes causal inference and that residual confounding may remain. Reverse causation is a live possibility too, because sleep disruption is a common symptom of depression, so at least some of the signal for psychiatric outcomes may run backwards.

The bigger picture here is that sleep stages behave less like a single dial and more like a set of separate readings, each connected to a different part of the disease map. Light sleep pulled in two directions at once, lowering risk for some conditions such as heart failure and cellulitis while raising it for major depressive disorder, which is what a genuine physiological signal tends to look like rather than a measurement artefact. Wakefulness after sleep onset and night to night irregularity each picked out their own small clusters of conditions.

Why It Matters

For middle aged and older adults, the practical takeaway offered by the authors is that six to eight hours of sleep can act as a health safeguard. That is a cheap and widely available intervention compared with most of what medicine offers, and it arrives attached to a large sample and a long follow up rather than a small laboratory experiment.

The size of the exposed population is considerable. Medical Xpress reported on September 17 that the 2026 China Sleep Health White Paper found the average surveyed Chinese person sleeps 6.97 hours per night and that 23 percent sleep less than 6 hours. Apply that lens to any large country and the number of people sitting below the five hour mark, the group with the greatest clinical vulnerability across 37 conditions, becomes substantial.

The study also reframes how clinicians might question patients. Asking how long someone sleeps is easy; asking about the regularity of that sleep and how often someone wakes during the night is closer to what the data suggest actually matters. Medical News Today also sought outside comment, with family physician Scott Nass recommending a consistent wake up time.

Next Up

The authors are straightforward that more work is needed. Because the design is observational, it cannot prove that sleep patterns cause disease, only that they are connected, and HealthDay reported on September 18 that further research is needed to understand why. The team flags reverse causation as a possibility for some conditions and residual confounding as a broader concern.

Two directions look most obvious from the paper itself. One is to test whether interventions that specifically target REM and deep sleep, rather than total time in bed, shift the disease associations observed across the 1,049 outcomes studied. The other is to work out which of the 156 statistically significant associations survive replication in independent cohorts using different accelerometer hardware and different populations.