Honest, tested sleep product guides. Real prices, no paid rankings. Affiliate disclosure
Sleep News & Research

More REM Sleep Linked to Lower Disease Risk: What the 2026 UK Biobank Study Actually Found

A 2026 PLOS Medicine study of 95,559 adults linked more REM sleep to lower rates of 83 diseases. The findings, the caveats, and what it means for sleep.

Share
8 min read
By GetSome Rest editorial team · Independent sleep product guides Published · Last updated
Abstract scientific illustration of a sleeping head in profile with glowing blue and purple REM brain-wave patterns over a dark night sky

People with more REM sleep developed 83 diseases at lower rates than people with less — including dementia, heart failure, and Parkinson’s disease — per a 2026 study of 95,559 UK Biobank adults. It is the largest objective look at sleep stages and disease risk yet, and its authors stress it cannot show cause and effect. Here is what the research actually found — and what it did not.

Contents

At a Glance: What the Research Says

A 2026 study asks a sharper question than ‘how many hours’: which stages filled those hours? It is the largest objective test yet of whether REM and deep sleep predict later disease — and REM emerged as the standout.

  • The study: 95,559 UK Biobank adults (ages 37–73) wore a research wrist accelerometer for seven days and nights; a deep-learning model called SleepNet estimated REM, deep, and light sleep. Hospital records were followed for a median of 8.9 years across 1,049 diseases.
  • REM sleep stood out: each additional ~47.6 minutes of REM was linked to lower rates of 83 diseases — dementia 46% lower, heart failure 26% lower, Parkinson’s disease 80% lower.
  • Deep sleep was narrower: more deep sleep lined up with lower rates of seven conditions, including type 2 diabetes, major depression, and sleep apnea.
  • Total duration mattered too: six to eight hours sat in the lowest-risk zone for 69 diseases; under five hours was linked to higher rates of 37 conditions.
  • Steadiness counts: irregular night-to-night sleep was linked to higher rates of anxiety and depression; more time awake after falling asleep was linked to higher rates of six conditions.

This is a research report, not medical advice. We describe what researchers found — not what any individual should do about their health.

How to Read a REM Sleep Study

Before the details, a quick guide to what makes a study like this strong — and where its limits are.

1. Association is not causation. This is a prospective cohort study: researchers watched sleep patterns and later diagnoses without asking anyone to sleep differently, so it cannot show that gaining REM sleep prevents disease. When the authors say “linked to,” that is exactly what they mean.

2. Wrist motion is not brain activity. Lab staging uses brain waves; here stages were estimated from wrist movement. Good enough for population patterns, but each night’s number carries uncertainty.

3. Read hazard ratios as comparisons. A hazard ratio of 0.74 for heart failure means the higher-REM group developed it at about 74% of the rate of the lower-REM group — a 26% relative reduction. It says nothing about any one person’s chances.

4. Testing 1,049 diseases invites flukes. The researchers applied a Bonferroni correction — a strict statistical bar for multiple testing — and only the 156 associations surviving it were reported. That discipline is a strength of this study.

5. Reverse causation is real. The authors flag this themselves: early, undiagnosed disease can degrade sleep years before diagnosis, so the sleep–disease arrow could point either way — or both.

6. One week of sleep, nine years of outcomes. Sleep was measured over a single week: one photograph standing in for a decade-long film.

Person sleeping peacefully in a dark bedroom with soft blue moonlight and faint glowing wave lines suggesting REM sleep above the head
REM sleep — the dream-rich stage that grows longer toward morning — was linked to the broadest pattern of lower disease rates in the study.

The Findings, Side by Side

Five sleep measures were tested against more than a thousand diseases. Here is what survived the strictest statistical cut, with each finding’s caveat.

Sleep measureFindingLimitation
REM sleepPer interquartile-range increase (~47.6 min): lower rates of 83 diseases across 12 categories. Dementia HR 0.54 (0.47–0.62); heart failure HR 0.74 (0.68–0.80); atrial fibrillation HR 0.83 (0.79–0.86); Alzheimer’s disease HR 0.69 (0.57–0.81); multiple sclerosis HR 0.48 (0.36–0.64); Parkinson’s disease HR 0.20 (0.11–0.39)Stages estimated from wrist movement, not brain waves; one week of data; reverse causation possible
Deep sleepLower rates of seven conditions, including type 2 diabetes HR 0.89 (0.85–0.93), major depressive disorder HR 0.86 (0.82–0.91), Parkinson’s disease HR 0.70 (0.62–0.80), and sleep apnea HR 0.78 (0.71–0.86)Narrower signal than REM; same movement-based measurement caveats
Total sleep durationNonlinear link with 86 disease phenotypes; the 6–8 hour window held the lowest risk for 69 diseases, while under 5 hours was linked to higher rates of 37 conditionsU-shaped curve; short sleepers differ in lifestyle and health, so confounding is hard to fully remove
Sleep irregularityMore night-to-night variation linked to higher rates of anxiety and major depressive disorder, with some summaries also flagging substance-use disordersMeasured over one week; may partly reflect shift work or caregiving rather than a fixable habit
Wake after sleep onsetMore time awake in the middle of the night linked to higher rates of six conditionsFragmented nights can be an early sign of other health problems, not necessarily a cause

Our Verdict: The Honest Takeaway

After reading the paper and the independent commentary around it, here is the most defensible reading — what is supported, what is plausible but unproven, and what the headlines should not have claimed.

Supported: REM sleep tracked the broadest pattern of lower disease rates. Across 95,559 people, objective wrist data, and nearly nine years of follow-up, more estimated REM lined up with lower rates of 83 diseases — far more than any other sleep measure. That breadth is the study’s real contribution. As our guide on deep sleep and memory explains, deep sleep plays its own distinct role — but here, REM carried the wider signal.

Plausible but unproven: that changing your sleep changes your risk. No one was asked to sleep differently, so nothing tests whether gaining REM prevents disease. Good sleepers differ from poor sleepers in exercise, diet, stress, and wealth. Strong association; no causal chain established.

Do not overreach: no one should chase REM on a wrist tracker. The REM numbers came from a research algorithm trained on lab data, not a consumer wearable’s sleep score, and experts quoted in the coverage note that algorithm-generated stage estimates carry meaningful uncertainty. This study says nothing about what a gadget’s nightly REM readout means for your health.

Practically: what the data reward is unglamorous — a consistent six-to-eight-hour window, similar bedtimes, and uninterrupted nights. REM concentrates in the last hours of sleep, so early alarms and late-night wake-ups shave it disproportionately. For persistent sleep problems, every credible voice in this literature says the same thing — talk to a clinician, not a chart. We report research; we do not diagnose or treat.

Study by Study: The Details

The headline numbers deserve their full context. Here is what the study did, how it measured sleep, and what its authors say it cannot show.

The UK Biobank study: the largest objective look at sleep stages and disease

Published in PLOS Medicine on September 17, 2026 (Luo J et al., doi:10.1371/journal.pmed.1005213), the study was led by Jingsong Luo of Peking University’s School of Public Health with senior author Shengzhi Sun of Capital Medical University. It drew on 95,559 UK Biobank volunteers aged 37 to 73 (mean 56.2; 43.7% men), each wearing an Axivity AX3 research accelerometer on the wrist for seven days and nights. The raw movement data were run through SleepNet, a deep-learning algorithm estimating nightly REM, light (N1/N2), and deep (N3) sleep, plus total duration, wakefulness after sleep onset, and night-to-night irregularity. Median total sleep was 392.4 minutes. Researchers then checked hospital records over a median 8.9 years against 1,049 disease phenotypes built from 10,515 ICD-10 codes, using Cox models adjusted for demographics, lifestyle, and environment. After Bonferroni correction, 156 associations remained significant — 83 tied to REM, seven to deep sleep. For the complementary stage, see our roundup Deep Sleep and Memory: What Scientists Just Learned (2026).

The measurement question: SleepNet and the wrist-tracker caveat

SleepNet is the engine the study runs on, so it deserves scrutiny. It classifies stages from wrist movement alone — no brain waves, no eye-movement sensors. Sleep scientists quoted in coverage — including Michael Grandner of the University of Arizona — cautioned that algorithm-generated REM and deep estimates carry significant uncertainty. The study’s signal survived despite that noise, which arguably makes the pattern more interesting — but no one should read it as validation of their smartwatch’s nightly REM score. Consumer wearables use different, generally less-validated algorithms.

The gap it filled: decades of self-reported sleep research

Most large sleep cohorts of the past two decades relied on self-reported sleep duration — a questionnaire asking roughly how long people sleep. Self-report is a poor proxy for what happens overnight, and duration says nothing about stage composition. This study is the largest attempt to replace the questionnaire with objective, stage-level measurement and scan the result against the whole human phenome. That is the methodological leap. It also helps explain why weekend catch-up sleep looks different in objective studies than in surveys: how you sleep and when you sleep are distinct variables.

What the authors flagged themselves: reverse causation and the exploratory design

The paper’s discussion does not sell more than it measured. The authors explicitly note that conditions like depression can disrupt sleep rather than stem from it — the causation arrow may run disease-to-sleep, especially for neurological outcomes. The phenome-wide design is exploratory: scanning 1,049 diseases is hypothesis-generating, not hypothesis-testing. Diagnoses came from hospital inpatient records, so milder conditions are underrepresented. The honest summary, in the authors’ own terms, is a map of associations — a starting point, not a verdict on how anyone should sleep.

How We Reported This Story

This article is a research report, so our method is literature coverage rather than product testing. We read the original PLOS Medicine paper (Luo J et al., 2026;23(9):e1005213), cross-checked the headline numbers against independent summaries in Healio and 2 Minute Medicine, and checked expert commentary for measurement caveats. We gave limiting context the same prominence as the striking findings — including the authors’ own cautions about reverse causation and the exploratory design.

We do not prescribe, diagnose, or treat: this is news about research, and persistent sleep problems deserve a clinician’s input, not a headline-driven routine change. This story contains no product recommendations and no affiliate links — there is nothing here for us to earn a commission on.

Frequently asked questions

What did the UK Biobank REM sleep study actually find?

Researchers led by Jingsong Luo (Peking University) and Shengzhi Sun (Capital Medical University) analyzed 95,559 UK Biobank adults who wore wrist accelerometers for seven days and nights; a deep-learning model called SleepNet estimated each night’s REM, deep, and light sleep. Over a median 8.9 years of hospital-record follow-up, each additional ~47.6 minutes of REM was linked to lower rates of 83 diseases across 12 categories — dementia 46% lower, heart failure 26% lower, atrial fibrillation 17% lower, Alzheimer’s 31% lower, and Parkinson’s disease 80% lower. Deep sleep was linked to seven conditions, including type 2 diabetes and major depression. Published in PLOS Medicine on September 17, 2026.

Does more REM sleep cause better health?

No — and the study does not claim that. It is observational: researchers watched sleep patterns and later diagnoses without asking anyone to sleep differently, so it can show associations, not causation. People who sleep well also tend to exercise more, eat differently, and face less stress, and early, undiagnosed disease can itself degrade sleep years before diagnosis — a problem called reverse causation. Whether changing sleep changes risk remains untested.

How was REM sleep measured?

Not with brain waves. Participants wore an Axivity AX3 research accelerometer on the wrist for seven consecutive days and nights, and the raw movement data were passed through SleepNet, a deep-learning algorithm trained to estimate sleep stages from motion patterns. That is a major advance over questionnaires, but movement-based estimates are less precise than sleep-lab polysomnography, and independent experts quoted in coverage noted meaningful uncertainty in algorithm-generated REM and deep numbers. One week of data also stands in for years of habits. Treat the stage estimates as good enough for population patterns, not as personal medical measurements.

What are the study's limits?

Four worth knowing. First, it is observational, so it cannot prove causation. Second, REM and deep sleep were estimated from wrist movement, not measured by brain waves, and consumer wearables use different algorithms. Third, sleep was measured over a single week while outcomes unfolded over nearly nine years. Fourth, diagnoses came from hospital inpatient records, so milder conditions are underrepresented, and the phenome-wide scan was exploratory rather than pre-registered. The authors flag reverse causation themselves — early disease disrupting sleep before diagnosis. Replication will decide how durable the map is.

How does this relate to deep sleep and memory?

The UK Biobank study found REM sleep had the broader association pattern, but deep sleep mattered in its own right — linked to lower rates of seven conditions including type 2 diabetes and major depression — and other research ties deep sleep specifically to memory consolidation and physical repair. Think of the two stages as dividing the night’s work: deep sleep dominates the early hours while REM grows longer toward morning, so cutting the night short at either end costs you different things. For the deep-sleep side of the story, see our research roundup Deep Sleep and Memory: What Scientists Just Learned (2026).

Key Takeaways

  • In the largest objective study of its kind, 95,559 adults with more estimated REM sleep developed 83 diseases at lower rates over nearly nine years — dementia 46% lower, heart failure 26% lower, Parkinson’s disease 80% lower per ~47.6 extra REM minutes.
  • Deep sleep showed a narrower signal: seven conditions including type 2 diabetes and major depression. Six to eight hours sat in the lowest-risk zone; under five hours was linked to higher rates of 37 conditions.
  • The findings are associations, not proof. The authors flag reverse causation: early disease can degrade sleep years before diagnosis, and no one was asked to change their sleep.
  • REM was estimated from wrist movement by a deep-learning model — not measured by brain waves — and each night’s number carries real uncertainty. Consumer wearables use different algorithms entirely.
  • What the data reward is unglamorous: a steady six-to-eight-hour window, similar bedtimes, and uninterrupted nights. REM concentrates late in the night, so the end of sleep matters as much as the start.

Reporting note: this article summarizes a single peer-reviewed study and its coverage through October 2026. It is not medical advice and does not recommend any product, habit change, or treatment.

Filed under: Sleep News & ResearchSources: Luo J, Liu R, Yin J, Cao W, Sun S, Chen R, PLOS Medicine 2026;23(9):e1005213 (doi:10.1371/journal.pmed.1005213); Healio primary-care summary, Sep 2026; 2 Minute Medicine clinical summary
Share this article

You might also like