Go to bed after 12:30 a.m. and your risk of liver cirrhosis runs 2.57 times higher than if you'd turned in between 11:00 and 11:30. Not "sleep less" — sleep later, and less regularly. That single hazard ratio, drawn from 88,461 UK Biobank adults wearing wrist accelerometers over an average 6.8 years, is the tell for what the entire study is arguing: the clock and the consistency matter more than the count.
This is association-grade evidence — an observational cohort, Cox-model hazard ratios, replicated in a second population (NHANES), not causation. Hold onto that. But within those limits, the finding is unusually clean.
Last week in this space we looked at the sleep-duration threshold and the reverse-causation problem — the reason "too much sleep kills" studies keep collapsing under scrutiny. This week the frame moves one step further: away from how long, toward how, and specifically toward rhythm and fragmentation as the dominant disease signal.
The clock and the consistency matter more than the count: sleep rhythm carried three times as many disease links as duration and onset.
Here is the paradox. In a phenome-wide scan where 172 diseases tracked with sleep, the metric every public-health campaign fixates on — hours slept, objectively measured — was a relatively weak predictor. Sleep rhythm carried three times as many disease links as duration and onset, and was tied to nearly half of the 172 diseases.
Eighty-three of the 172 associations — 48.3 percent — were sleep-rhythm specific. They did not show up in the existing subjective-measure literature at all, because that literature was busy counting hours.
The attributable-risk numbers make the stakes concrete. For Parkinson's disease, 37.05 percent of risk was attributable to sleep traits. Acute kidney failure, 21.85 percent. Across all 172 diseases, an average of 23.01 percent of disease burden traced back to sleep.
(A note on scope: Parkinson's appears here strictly as a sleep-associated outcome. Whatever neuroprotective story belongs to it is a different piece.)
Now the interesting part — why duration failed as a metric. It wasn't that duration doesn't matter physiologically. It's that self-reported duration is a corrupted measurement, and the corruption ran in a specific, damaging direction.
Among people who called themselves "long sleepers" — nine hours or more — 21.67 percent actually slept fewer than six hours. Roughly a quarter of the "long sleep" category were short sleepers who spent a long time in bed without sleeping.
That is not noise. It is systematic misclassification, and it manufactured false positives. When the researchers reanalyzed the UK Biobank data with subjective traits, the classic "long sleep is dangerous" associations — stroke, ischemic heart disease, cardiovascular disease, depressive episodes — showed up as artifacts of contaminated categories, not real signals.
The subjective long-sleep-is-harmful literature, in other words, was partly measuring people who couldn't sleep and mislabeling them as people who slept too much. Objective accelerometry found the long-sleep-to-disease link in only one disease.
So the field spent years tracking a mislabeled quantity while the mechanistically live signal sat one column over. And there is a mechanism. A mediation analysis found that inflammatory factors — leukocytes, eosinophils, C-reactive protein — contributed significantly to the newly identified rhythm-disease associations, including the ones replicated in NHANES like COPD and diabetes.
The links that survived objective measurement and cross-population replication are the links that trace through inflammation. That is the difference between an artifact and a candidate causal pathway.
This is why senior author Shengfeng Wang of Peking University frames it as a definitional problem: "It's time we broaden our definition of good sleep beyond just duration." Pair that with the cirrhosis hazard ratio — 2.57-fold for bedtime after 00:30 versus 23:00 to 23:30 — and you have both the thesis and its sharpest single data point in one breath.
The rest of the doubled-risk diseases follow the same logic. Age-related physical debility ran a 3.36 hazard ratio for the lowest versus highest quartile of relative amplitude — a rhythm measure. Gangrene, 2.61, for the lowest quartile of interdaily stability — another rhythm measure. The least efficient sleepers showed a nearly 1.8-fold increase in respiratory failure.
None of those are hours. They are amplitude, stability, efficiency, timing — the dimensions of sleep that duration-based screening cannot see.
Which raises the practical question: if rhythm and fragmentation carry the signal, can we actually measure them, or is this just a more sophisticated way to be vague? The answer is that the tooling already exists, and it's the same accelerometry that generated this study.
Fragmentation is no longer a complaint; it's an index. The standard actigraphy-derived metrics are concrete: wake after sleep onset (WASO), number of awakenings per night, sleep efficiency, and the rhythm measures the study leaned on — relative amplitude and interdaily stability.
Beyond the night-level summaries, high-resolution methods now quantify what fragmentation actually is. Transition density measures how often the system flips between inferred sleep and wake within a window — higher density, more fragmented. Bout-length distributions capture that fragmented sleep is built from shorter, more variable sleep periods, not just "more awakenings."
And the modeling has matured. Hidden Markov Models infer latent sleep states from raw movement, treating sleep-wake as a stochastic process with estimable transition probabilities — turning "I kept waking up" into a number you can put a hazard ratio on.
That is the whole retooling in one sentence: the field is moving from binary sleep-wake classification toward high-resolution temporal modeling, because that's where the disease signal lives. Someone with adequate total sleep and high fragmentation is invisible to a duration screen and, per this cohort, potentially at elevated inflammatory risk.
So what does this do to the advice? It flags the standard message — "catch up on your hours," "aim for seven to nine" — as not just weak but potentially misdirecting.
Telling the 21.67 percent of self-styled long sleepers to spend more time in bed is telling people who already can't maintain sleep to lie awake longer. Telling everyone to optimize a number ignores the metric that showed three times as many disease links. And "catch up on weekends" is, by construction, an instruction to make your rhythm less regular.
The honest limits remain. As sleep physician Matthew Scharf put it, the study "shows many associations but does not necessarily show causation" — and different patients may need different fixes, some longer duration, some more regularity.
The authors say the next step is causality and intervention trials. The clean version of that test, as neurologist Daniel Truong framed it, is almost obvious: does a program that enforces a regular bedtime actually change outcomes in a defined patient group?
Until those trials read out, this is what we have: a large, objectively measured, replicated cohort saying the thing we count is not the thing that matters most. Regularity is now quantifiable. Whether it's fixable — and whether fixing it moves disease risk — is the experiment that hasn't been run yet.