For years, sleep researchers have leaned on a blunt metric to judge whether someone is sleeping well: how many hours they spend in bed. A study published September 17 in the open-access journal PLOS Medicine argues that number is missing most of the picture. Using real-world wrist accelerometer data from nearly 96,000 people followed for almost a decade, researchers found that the internal architecture of sleep — specifically how much rapid eye movement (REM) and deep, slow-wave sleep a person gets — predicts the future risk of dozens of diseases independent of total sleep time.
Nine Years, 96,000 Wrists, Over 1,000 Diseases
The study, "Accelerometer-derived real-world sleep stages and risk of incident diseases: A UK Biobank cohort study and phenome-wide association analysis," led by Jingsong Luo and colleagues, drew on 95,559 UK Biobank participants who wore wrist-based accelerometers continuously for seven days and nights. Rather than relying on self-reported sleep diaries or a single night in a sleep lab, the researchers used a deep-learning model to classify each participant's movement patterns into likely sleep stages — wake, REM, and non-REM sleep — across a full week of ordinary life.
Participants were then tracked for a median of 8.9 years using linked health records, and the researchers systematically tested for associations between sleep-stage patterns and more than 1,000 distinct disease outcomes, applying strict statistical correction to guard against false positives from testing so many outcomes at once.
More REM and Deep Sleep, Lower Risk Across the Board
The results were striking in their breadth. Greater REM sleep, measured per interquartile range increase of about 47.6 minutes, was associated with a lower risk of 83 diseases. Some of the individual associations were substantial:
- Heart failure: hazard ratio of 0.74, meaning roughly a 26% lower risk
- Dementia: hazard ratio of 0.54, roughly a 46% lower risk
- Parkinson's disease: hazard ratio of 0.20, roughly an 80% lower risk
Deep, slow-wave sleep showed its own pattern of protective associations across a separate set of conditions, reinforcing the idea that REM and deep sleep are doing biologically distinct work rather than simply serving as proxies for "good sleep" in general. On the other end of the spectrum, total sleep time under five hours a night was linked to elevated risk across a wide swath of the more than 1,000 conditions studied, regardless of how that short sleep was distributed across stages.
A Real Strength — and a Real Caveat
Outside experts who reviewed the study, including sleep researcher Dr. Nina Rzechorzek, noted that its scale and design are genuine strengths: nearly 96,000 participants, close to nine years of follow-up, and rigorous statistical correction for testing over 1,000 outcomes simultaneously, alongside multiple sensitivity analyses meant to rule out reverse causation and confounding.
But the study has an important limitation that researchers themselves flagged: sleep stages were not measured with gold-standard polysomnography, the electrode-based test used in sleep labs. Instead, they were inferred by a machine-learning model trained to guess sleep stage from wrist movement alone, and that model's own reported accuracy for distinguishing REM, non-REM, and wake was only moderate. That means some of the specific stage-disease associations could reflect noise in the classification model rather than true biology — a caveat that tempers, without erasing, the study's headline findings.
What This Means for Patients
This study cannot tell any individual how to get more REM or deep sleep, and it does not prove that poor sleep architecture directly causes any of the 83 diseases identified — the associations could partly reflect early, undiagnosed disease processes disrupting sleep rather than the reverse. What it does add is a large-scale signal that clinicians and consumer sleep-tracking companies alike should stop treating total sleep duration as the only number that matters. Patients whose smartwatches or rings report low percentages of deep or REM sleep, even while logging what looks like a full night in bed, may be a more informative flag for future health risk than hours logged alone — a distinction that could eventually shape how primary care doctors decide who needs a closer look at their sleep, and how urgently.
