More REM sleep tracked with lower risk across 83 conditions in 95,559 adults

An extra 47.6 minutes of REM sleep, the spread between the typical low-REM and high-REM sleeper in the UK Biobank, goes with a 26 percent lower hazard of heart failure and a 46 percent lower hazard of dementias. Those figures come from a cohort of 95,559 adults who wore wrist accelerometers for seven days and were then followed through hospital records for a median of 8.9 years.

The association runs across 83 conditions in total, according to a paper published September 17, 2026 in PLOS Medicine. It is an observational result built on a motion sensor rather than a sleep lab, and the people who read it most closely say so.

A 95,559-person cohort and a wrist sensor

Jingsong Luo and Ruiyi Liu of Capital Medical University’s School of Public Health, with senior author Rui Chen, ran the analysis on UK Biobank participants whose accelerometer records allowed sleep staging. The mean age at wear was 56.2 years. A deep-learning algorithm called SleepNet converted seven days of wrist movement into estimates of REM, deep and light sleep, total duration, irregularity and time awake after sleep onset.

The wearable data come from the UK Biobank’s wrist-worn accelerometer program, whose main collection ran between June 2013 and January 2016 for about 100,000 participants. Each person was tracked from the date of wear until the first occurrence of disease, death or April 1, 2024.

The authors mapped hospital outcomes onto 1,049 well-defined disease phenotypes and tested each against each sleep measure, a phenome-wide design that casts a wide net rather than testing one disease at a time. Median REM time in the cohort was 81.7 minutes, with an interquartile range of 59.1 to 106.6 minutes.

The 83 conditions linked to REM

REM was the stage with the broadest reach. Greater REM duration was associated with lower risk for 83 of those phenotypes, measured per interquartile range, which here is 47.6 minutes. The paper’s published examples include heart failure at a hazard ratio of 0.74 (95 percent confidence interval 0.68 to 0.80), dementias at 0.54 (0.47 to 0.62), Alzheimer’s disease at 0.69 (0.57 to 0.81) and Parkinson’s disease at 0.70 (0.62 to 0.80). A hazard ratio below 1 means fewer cases per unit of follow-up in people with more REM, not that any individual was protected.

Deep sleep was a narrower story. It was associated with lower risk for only seven diseases, among them type 2 diabetes at a hazard ratio of 0.89 (0.85 to 0.93). Coverage of the paper by Medical News Today adds hypertension, atrial fibrillation, multiple sclerosis and major depressive disorder to the list of conditions the stages touched.

Short sleep, long sleep and the 6-to-8-hour window

Duration behaved differently from stage. The authors found non-linear relationships across 86 disease phenotypes, with the lowest risk concentrated in a window of six to eight hours. Sleeping under five hours was associated with significant adverse associations in 37 of 41 phenotypes tested, a pattern a press summary of the paper reports as increased risk for 37 conditions. People in the six-to-eight-hour group also had the highest share of REM, 0.222 of the night, so the stage and duration findings overlap.

Irregular sleep and time awake after sleep onset pointed the other way, toward higher risk for conditions such as psychoactive substance dependence, alcohol abuse and osteoarthritis in the Medical News Today account.

Wrist accelerometer staging versus polysomnography

Two sleep scientists quoted in a Science Media Centre reaction roundup aimed at the measurement. Dr. Nina Rzechorzek of the University of Cambridge noted that the sleep stages were inferred by a deep-learning model from wrist movement, rather than measured directly with polysomnography, the gold standard. Dr. Greg Elder of Northumbria University said that movement-based watches cannot yet accurately determine which sleep stage a person is in overnight.

The authors list their own limits: the design is observational, residual confounding and reverse causality remain concerns, only seven days of sleep were recorded, and outcomes relied on inpatient diagnoses. Reverse causality is easiest to see with depression, where disrupted sleep can be a symptom of the illness rather than a cause. New Atlas framed the result as sleep architecture, and not duration alone, mattering for disease risk, which is the claim the data can carry without implying that adding REM minutes would change anyone’s outcome.

The unresolved point is whether the accelerometer’s REM estimate tracks real REM closely enough at the level of a single night. Every one of the paper’s 83 associations rests on that estimate, and the two sleep scientists quoted above put that dependence at the center of their caution. Until REM minutes derived from a wrist can be checked against laboratory staging at the scale of a cohort this size, the hazard ratios describe a relationship between a model’s output and later hospital diagnoses.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

More from Morning Overview