Regularity and Mortality: What the Cohort Study Can Say

Windred and colleagues paired a large accelerometer-derived sleep-regularity dataset with later mortality records. Their results are important cohort evidence, not a prescription: people in the more-regular quintiles had lower observed mortality than the least-regular quintile, but those comparisons do not estimate what a person could achieve by changing sleep.

The Cohort and Its Exposure

Windred and colleagues analyzed data from 60,977 UK Biobank participants with usable wrist-accelerometer sleep–wake information. They computed the Sleep Regularity Index from more than 10 million recorded hours. The exposure was a week of activity-based estimates, with multiple valid sleep/wake days required; it was not a lifetime sleep history, clinical interview, or direct measurement of brain activity.

The outcome was mortality recorded after the accelerometer assessment. There were 1,859 deaths during follow-up, at a rate of 4.84 per 1,000 person-years. Mean follow-up was 6.30 years, with mortality reported up to 7.8 years after measurement. Researchers examined all-cause mortality and cause-specific outcomes, adjusting for age, sex, ethnicity, sociodemographic characteristics, lifestyle, and health factors.

These details are why the result is informative: a measured exposure preceded an important outcome, the sample was large, and the paper compared regularity with sleep duration. They are also why careful boundaries matter. One observation week can misclassify people whose schedules vary by season, roster, caregiving, or major life events.

What the Quintile Finding Means

The least-regular quintile was the reference group. Across the top four quintiles of SRI, adjusted all-cause mortality risk was reported as 20%–48% lower than in that reference. Those figures summarize relative between-group associations in this cohort. They do not mean that a person who starts in the least-regular group can lower their own mortality risk by 20%–48% by making sleep more regular.

Quintiles are a ranking device: they divide the observed sample into fifths. They do not identify a threshold where biology changes, and the scores representing each fifth depend on this cohort and algorithm. A high quintile is not a target that all people can or should reach. The analysis compares people who already differed in their schedules, health, and life circumstances when they wore the device.

Relative risk also differs from absolute risk. A 20% relative reduction can correspond to very different absolute differences for people with different baseline risks. The 1,859 recorded deaths provide context for the cohort; they cannot be used to calculate an individual’s expected survival benefit from changing bedtime. The paper measured associations, not the effects of a sleep intervention.

Why Sleep Regularity Could Predict More Than Duration

The study reported that SRI was a stronger predictor of all-cause mortality than sleep duration in its model comparisons. That is a result about prediction within the chosen cohort, outcome, covariates, measurement window, and analytic specification. It does not show that regularity is more important than duration for every person or every outcome, and it does not establish a causal mechanism.

Predictive usefulness is not the same as a treatment target. A variable can help sort people by observed outcome because it reflects stable routines, health status, job conditions, or access to resources. An intervention that changes the variable may not change the outcome, especially if it does not change the underlying causes that produced the association.

The same paper also shows overlap rather than a clean separation: SRI has a relationship to duration variability, but the measures are not identical. Duration asks how long someone sleeps. SRI asks about repeated sleep/wake states at matching points a day apart. Existing social-jetlag coverage handles the weekday-to-free-day midsleep gap; this mortality page stays with device-based regularity and how to read a cohort endpoint.

Residual Confounding and Selection

Adjustment for many measured variables strengthens the analysis, but it cannot account for every relevant difference. Sleep regularity may track shift work, income, chronic illness, depression, household demands, neighborhood noise, physical activity, and healthcare access. Some of these factors may be measured imperfectly or not at all. As a result, an adjusted association may still combine effects of sleep pattern with effects of the conditions surrounding it.

Participation and device wear are also selective. UK Biobank volunteers are not a random sample of every population, and the accelerometer subset is narrower still. Participants had to wear the device and meet data-quality criteria. Older and healthier volunteers, or those more able to participate, may be overrepresented. Statistical weights and adjustment cannot guarantee transportability to other populations.

Cause-specific mortality adds further uncertainty because fewer deaths fall in each category than in the all-cause endpoint. The overall association should not be stretched into a claim about cancer, cardiovascular death, or a specific disease unless that separate estimate and its confidence interval support the claim. A cohort can be large while subgroup estimates remain imprecise.

What a Wearable Captures—and Misses

Actigraphy infers sleep and wake from movement. It is useful for observing many nights in ordinary settings, and the SRI can incorporate naps and multiple bouts. Yet quiet wake may be scored as sleep, movement can be scored as wake, and device algorithms can differ. A wrist device does not directly measure sleep architecture, circadian phase, breathing events, or glucose metabolism.

One week may include both work and free days, but it may miss a two-week roster, a monthly rotation, travel, seasonal change, or an unusually stressful period. The same participant may also vary greatly from one month to another. A single baseline measure is a snapshot; repeat measurements would help distinguish a persistent habit from temporary disruption.

The paper’s data are powerful for epidemiology because thousands of people can be monitored at scale. They should not be mistaken for a personal diagnosis. If tracker results are paired with persistent insomnia, dangerous sleepiness, or breathing symptoms, those symptoms—not a mortality association in a volunteer cohort—are the reason to ask a clinician for evaluation.

The Responsible Takeaway

Windred et al. provide evidence that, in one large cohort, more regular device-derived sleep patterns predicted lower observed all-cause mortality over follow-up, even in models considering duration. That makes regularity worth measuring in future studies and may make it useful as one part of a broader sleep-health profile. It does not establish how much a person should change their routine, whether every schedule is modifiable, or whether the change would extend their life.

For personal reflection, note what the device defines as “regular,” whether the week was typical, and how much sleep was obtained. Do not chase the upper quintile or infer a treatment effect from its contrast with the bottom fifth. If schedule changes are not feasible because of shift work or caregiving, the study does not imply a failure of effort.

For the measurement details, read how SRI is calculated. The neighboring cardiovascular and diabetes pages examine other prospective endpoints; each has its own cohort and cannot be combined into one causal claim. The broader sleep-timing page remains the handoff for bedtime guidance.

It is tempting to treat the 20%–48% range as a gradient and assign the 20% value to one quintile, the 48% value to another, then sketch a smooth treatment curve. The published summary gives an interval of associations across the top four quintiles; those are not verified individual treatment steps. Each quintile represents a fifth of the sample, but the contrast depends on the underlying distribution, covariate adjustment, and the least-regular group used as reference. The reported range cannot identify which single change—earlier bedtime, steadier wake time, fewer naps, or something else—would account for it.

Death is a high-stakes endpoint, but all-cause mortality also combines many causes and social pathways. Even after adjustment, people with regular schedules may differ in health, work stability, income, or access to healthcare. These differences can persist when researchers adjust for variables they were able to measure. Cause-specific estimates can add detail, but fewer events make them less precise; they should not be casually substituted for the all-cause result.

Finally, follow-up beginning after a single measurement week assumes that this short record is informative about the exposure during years of observation. That may work reasonably well for stable routines and poorly for people with rotating shifts or life changes. Repeated actigraphy would improve exposure characterization; an intervention would still be needed to test whether changing the pattern itself changes mortality.

Related Topics