Most sleep advice is still a stopwatch. Seven to nine hours, lights out by eleven, aim for a hard target and feel bad if you miss it. It's not wrong, but it quietly skips the question your body actually seems to care about more: whether you keep showing up at the same time.
That's what two separate 2023 analyses of UK Biobank data suggested — not with a sleep survey, but with 10 million-plus hours of wrist accelerometer recordings. The results are hard to shrug off.
The study behind the headline
In Windred and colleagues' Sleep paper, researchers scored around 60,977 UK Biobank participants on something called the Sleep Regularity Index (SRI). It's a day-to-day consistency measure: for every 30-second epoch of the recording, it checks whether you were in the same sleep-or-wake state at that same clock time 24 hours earlier. A score of 100 means perfectly regular. A score of 0 is random.
Participants were followed for a mean of 6.3 years, with mortality tracked up to 7.8 years. Compared to the least regular quintile (SRI below roughly 71.6), the top four quintiles had a 20–48% lower risk of dying from any cause, a 16–39% lower risk of cancer death, and a 22–57% lower risk of cardiometabolic death. When the authors directly pitted SRI against sleep duration, SRI produced a better statistical fit for all-cause mortality, and adding duration on top of SRI didn't meaningfully improve the model.
A second analysis published the same year — Cribb and colleagues in eLife — looked at 88,975 UK Biobank participants and found the same shape of relationship. The least regular sleepers (5th percentile) had a 53% higher mortality risk than the median; the most regular (95th percentile) had about 10% lower. The association was non-linear: the biggest risk cliff sat at the very irregular end.
Both are observational cohorts, so they show association, not cause. The participants skew older and mostly white British. SRI was estimated from a single week of wrist data. But two independent analyses, same database, same direction of effect — that's a stronger signal than one headline-friendly paper.
What counts as "regular" in a real life
A score isn't a bedtime rule, so the practical read is messier than "sleep at 10:30pm." A rough translation of what the researchers' most-regular quintile looked like: your sleep window — the hours you're actually asleep, not just lying there — doesn't drift by more than roughly 30 minutes to an hour from one 24-hour cycle to the next. Weekend sleep-in to catch up on a short week lands squarely on the "irregular" side of the ledger. So does a late Friday followed by an early Saturday workout, then a lazy Sunday.
The circadian explanation the authors offer is a hypothesis, not a tested mechanism in this paper. But it's a reasonable one: your internal clock runs on roughly a 24-hour timer that it recalibrates every day using light, meal timing and activity. Shift those inputs by two hours on Friday and Saturday and your Monday-morning body is, functionally, two time zones away from your calendar.
| What people usually track | What the regularity data points at |
|---|---|
| Total hours of sleep | How consistent the sleep window is |
| Bedtime | Both bedtime and wake time, day after day |
| Weekend "catch-up" sleep | Weekend drift as a risk, not a reward |
| Hitting 8 hours occasionally | Hitting a similar window reliably |
How do I actually make my sleep more regular?
Start with the end, not the start. Wake time is the easier anchor because it's set by an alarm, light exposure, and your first appointment of the day — all more enforceable than bedtime. Pick a wake time you can keep seven days a week, give or take 30 minutes, and let bedtime back-calculate from there.
A few things that make this less painful in practice:
- Treat weekends as weekdays-with-fewer-meetings. If your weekday wake-up is 7am, aim for 7:30–8am on Saturday, not 10am. A one-hour drift is a world less disruptive than three.
- Front-load light in the morning. You don't need to measure lux. Open the blinds, eat breakfast near a window, walk to the kopitiam. Morning light is one of the strongest anchors your clock has.
- Give yourself a fixed wind-down trigger, not a fixed bedtime. A specific cue — brushing your teeth, dimming the lights, the same book — travels with you across time zones and busy nights better than a clock.
- Sacrifice duration for consistency on bad nights. Counterintuitive, but if you got to bed at 2am, waking up at your normal time (even if that means a shorter night) tends to protect the next night's bedtime better than sleeping in and losing the anchor entirely.
What this doesn't say
It doesn't say duration doesn't matter. Chronic short sleep still has its own well-established costs, and nothing in these studies implies six hours on a tight schedule beats seven and a half with a bit of drift. It says, roughly, that for a given duration, how reliably you take it seems to matter in its own right. The authors are careful about that: for cardiometabolic mortality specifically, regularity's edge over duration narrowed once duration was added to the fully adjusted model.
It also doesn't medicalise shift work or new parenthood. If your schedule is genuinely non-negotiable, this isn't a reason to feel worse. It's a reason to protect what regularity you can keep — a consistent wake time on your off days, a stable meal pattern, morning light — rather than piling an irregular bedtime on top of an already irregular week.
For Singapore readers specifically: the equatorial light cycle is one of the most regular on earth, which is a free consistency anchor if you let it in. Daylight at roughly 7am and dusk at roughly 7pm, year-round, is a circadian gift. Spending every daylight hour in an office under the same fluorescent lights, scrolling the same bright phone until 1am Fridays, is the version where that gift gets wasted.
