A study published today inPLOS Medicine is the largest look yet at sleep stages and disease. Researchers in Beijing took seven days of wrist-accelerometer data from 95,559 UK Biobank participants, estimated REM, deep sleep, wakefulness and irregularity with a machine-learning model, and followed them a median of nearly nine years across more than a thousand health outcomes. More REM sleep was associated with lower risk of 83 diseases; more deep sleep with seven; more irregularity and more waking after falling asleep with higher risk of a handful, including depression. The sizes are not small: for each additional interquartile range of REM — about 48 minutes — the hazard of heart failure was 26% lower and of dementia 46% lower. Sleeping under five hours accounted for most of the adverse associations with duration, and six to eight hours was the low-risk range for most outcomes.
The three caveats the authors wrote, and the one for your ring. It is observational; people in early illness sleep differently, and when the researchers discarded the first two years of follow-up, only about 60% of the associations survived, which they say suggests some reverse causation. It is one week of measurement, with covariates recorded years earlier. And the sleep stages were estimated by an algorithm the authors say differs systematically from the laboratory gold standard, underestimating REM. That last point is the one to carry to your nightstand. If a research-grade accelerometer with a published algorithm cannot match the sleep lab, the REM number on a consumer ring — a different algorithm, on heart rate and motion, with no published validation against this paper's — is not the quantity this study measured. The finding is real. The number on your phone is a different thing wearing its name.
Loree’s takeI will not tell patients to get more REM, because nobody can aim at a sleep stage. What you can aim at is what produces it: enough hours, regular hours, less alcohol in the evening, which suppresses REM, and treated apnea, which fragments it. Those are the levers, they are free, and they move the stage this study says matters. The ring can tell you whether you were in bed. It cannot tell you what this paper's instrument saw.
Ben’s takeExpect this paper on every sleep-tech product page by the weekend, with the word "REM" doing the work. The paper used a research device, an academic algorithm and a 95,000-person cohort, and it still flagged its own stage estimates as off from the gold standard. The ring companies have never published that comparison. When a product cites a study, ask whether it measured what the study measured. Here it did not.
Why it matters to a recovering optimizer. Sleep is the fundamental the industry has most successfully turned into a number, and this is the strongest evidence yet that the architecture of sleep matters — along with the clearest statement that the consumer version of that architecture is an estimate of an estimate. Keep the hours regular and sufficient, and let the stages take care of themselves. They will, and no device needs to watch.
You can't aim at REM. Aim at regular hours, less evening alcohol and treated snoring — and ignore the stage chart.