A study of 95,000 people wearing sleep sensors found that REM sleep protects against 83 diseases and deep sleep protects against 7 more. Outside specific sleep windows, those protections disappear.
Public health messaging about sleep has settled on a simple message: get seven to nine hours. The advice is backed by substantial evidence. It is also incomplete.
A new study published in PLOS Medicine tracked 95,559 UK Biobank participants using wrist-worn accelerometers worn continuously for seven days and nights. Rather than asking people how much they slept, the study measured it directly. Researchers then followed participants for nearly nine years, tracking disease incidence across more than a thousand conditions.
They examined six sleep dimensions independently: total duration, REM proportion, deep sleep proportion, sleep timing, regularity, and time spent awake after initially falling asleep. The results showed 156 significant disease associations across all six measures.
REM sleep alone accounted for 83 of them, and every single one was protective. People getting more REM sleep showed lower risk of heart failure, Alzheimer’s disease, atrial fibrillation, hypertension, and dozens of other conditions. Deep sleep was independently associated with lower risk of seven diseases including type 2 diabetes and Parkinson’s.
On the other end, sleeping fewer than five hours raised risk for 37 diseases. Irregular sleep timing raised risk for three conditions. Waking frequently after falling asleep was linked to six more.
Stage matters as much as duration
The critical finding was that REM sleep’s associations held even after controlling for total sleep duration. More REM sleep was not simply a byproduct of sleeping longer. Even among people sleeping similar total hours, those with higher REM proportions showed meaningfully lower disease risk.
This matters practically. Two people sleeping seven hours may have very different REM proportions depending on alcohol use, sleep timing, consistency, and stress levels. The person who sleeps seven hours but wakes frequently, drinks before bed, or has irregular timing may be getting far less REM than someone with the same total hours but more consolidated, consistent sleep.
The six to eight hour window emerged as the sweet spot most consistently associated with favorable sleep architecture. Within that range, REM and deep sleep proportions were highest and disease associations most protective. Below five hours, 37 disease risks rose. Above eight hours, the protective associations weakened.
What REM and deep sleep do differently
REM sleep is most concentrated in the later hours of a night’s sleep. It is associated with memory consolidation, emotional processing, and clearance of metabolic waste from the brain through the glymphatic system. The proteins that accumulate in Alzheimer’s disease, amyloid-beta and tau, are cleared most efficiently during sleep, and REM disruption may impair this process over years.
Deep sleep, or slow-wave sleep, is most concentrated in the early hours of a night. Growth hormone is released, immune function is supported, and glucose regulation is partially restored. Its associations with type 2 diabetes and Parkinson’s risk are biologically plausible through these same pathways.
The study cannot establish causation. People who are already developing disease may sleep worse, meaning worse sleep and higher disease risk are partly caused by a third underlying factor rather than the sleep loss itself. But the use of objective measurement, the nine-year follow-up, the sample size of nearly 100,000 people, and the specific finding that sleep stage associations hold independent of sleep duration all strengthen the case that what you do during sleep matters, not just how long you do it.
“These findings highlight that both sleep architecture and sleep duration are associated with disease risk,” the researchers wrote. “Maintaining six to eight hours of sleep is associated with a more favorable sleep architecture and lower disease risk.”
The study, “Accelerometry-derived sleep patterns and risk of chronic diseases: A prospective study of 95,559 UK Biobank participants”, was authored by Luo et al. at Capital Medical University, Beijing, and published September 2026 in PLOS Medicine.
Source: Capital Medical University. DOI: 10.1371/journal.pmed.1005213
