The Sleep Number I Never Tracked Was the Clock

I can tell you my fasting window to the quarter hour. I can tell you how many minutes I lift in a week, because this summer I finally counted. Friday is one meal, late in the day, after training, and the whole day is built around that clock. I lead platform engineering teams, so I also live inside a calendar where standup is at the same time every morning and nobody thinks that is strange.

Sleep is the one thing I do track. I wear an Ultrahuman ring and a Withings ScanWatch Nova, and every morning the score tells me how hard to push the day. My bedtime is typically 10 p.m. and I am up at 6.

Typically. That one word is the whole subject of this piece.

A score is a verdict on last night. "Typically" is a claim about the pattern, and I have never checked how often it is true. I have never put a number on how regular my sleep is: whether the time I fall asleep and the time I wake up land in roughly the same place from one night to the next, as opposed to how well I slept on any one of them.

That gap stayed invisible until I read two large studies, published about a year apart, built on the same question: does the consistency of your sleep predict how you do over the following years, separately from how many hours you get?

Both say it does. The second one says something sharper: that for people with irregular sleep, getting the recommended number of hours did not cancel the extra risk.

I have opinions about what that means and what it does not, and I will get to both. But I want to start with how these studies measured sleep, because it is not how most of us think about it, and the measurement is where most of the confusion starts.

What "Regular" Actually Means in These Studies

Both studies drew on the UK Biobank, a long-running research cohort of adults in Britain, and specifically on the subset who wore a wrist accelerometer for one week. The device records movement in short slices of time. Software then classifies each slice as asleep or awake.

From that record the researchers calculate the Sleep Regularity Index, or SRI. The idea is simple. Take any moment in the recording and compare it to the same moment 24 hours later. Were you in the same state, asleep or awake, both times? Average that across the whole week. An SRI of 100 means perfectly regular: asleep at the same times every day. An SRI of zero means the pattern is no better than random.

Two things about that matter. First, it is a measure of the pattern, not the amount. Someone who sleeps six hours every night from midnight to 6 a.m. scores well. Someone who sleeps eight hours but shifts between 10 p.m. and 3 a.m. across the week scores poorly. Second, because it compares the state at each moment, it captures naps and broken sleep too, not just bedtime. It is a blunter instrument than "I go to bed at 10:30."

In the larger of the two datasets, the median SRI was 81, with the middle half of people falling between about 74 and 86. So most people in the study were neither military-regular nor chaotic. They were somewhere in the ordinary middle, which is also where most of us live.

The Mortality Study

The first paper, by Daniel Windred and colleagues, was published in the journal Sleep. It followed 60,977 UK Biobank participants, average age about 63, for an average of 6.3 years after their week with the accelerometer. In that time, 1,859 of them died.

The researchers split everyone into five equal groups by regularity and compared each group to the least regular fifth. After adjusting for age, sex, ethnicity, socioeconomic factors, lifestyle, and health, the pattern ran in one direction. Relative to the least regular fifth, the all-cause mortality hazard ratios for the next four groups were 0.80, 0.75, 0.72, and 0.70. In plain terms: the most regular fifth had about 30% lower mortality risk over the follow-up than the least regular fifth. For cancer death the figure was a hazard ratio of 0.76. For cardiometabolic death, 0.62.

Notice the shape. Most of the gap opens between the bottom group and the second one. After that the improvement is gradual. The biggest jump is from "irregular" to "somewhat less irregular," not from "good" to "perfect."

The authors also asked the question that made this study travel: is regularity a better predictor than duration? They compared models, and regularity predicted all-cause mortality better than sleep duration did. Adding duration to the regularity models did not significantly improve them.

That does not mean duration is irrelevant. The same paper found that very short and very long sleep were associated with mortality in the simpler model. It means that in this dataset, with this measurement, the timing pattern carried more information than the hour count.

The biggest gap is between irregular and slightly less irregular. Nobody in these data needed to be perfect to look better.

The Heart Study, and the Part About Hours

The second paper appeared in the Journal of Epidemiology and Community Health in November 2024. It looked at 72,269 UK Biobank adults aged 40 to 79, none of whom had a prior major cardiovascular event, and followed them for eight years. The outcome was MACE, major adverse cardiovascular events: heart attack, heart failure, and stroke, taken from hospital and death records. Events in the first year were excluded, a standard attempt to avoid counting people who were already becoming ill when the week of recording began.

The authors divided people into three groups by SRI: irregular (below about 72), moderately irregular (about 72 to 87), and regular (above about 87). Compared with regular sleepers, irregular sleepers had a hazard ratio of 1.26 for MACE. That is roughly 26% higher risk. Moderately irregular sleepers had a hazard ratio of 1.08. The relationship looked close to linear across the range, with the steepest improvement at the higher end of regularity.

Then comes the finding I keep returning to. The researchers asked whether getting enough sleep, measured against age-specific recommendations, would offset the effect of being irregular. For the moderately irregular group it did: the hazard ratio fell to 1.07, with a confidence interval spanning 1. For the irregular group it did not: the hazard ratio was still 1.19, with an interval that stayed above 1.

Hours helped the middle group. They did not rescue the bottom group. Which is the reverse of how most of us budget sleep, where the entire plan is to protect the number of hours and let the timing float.

Two Studies Is Not Two Confirmations

Here is something the headlines skipped. Both papers use the same biobank and the same accelerometer sub-study. They are different analyses, with different outcomes and different follow-up, but they are not independent samples. If something peculiar about people who agreed to wear a wrist device for a week and sit in a research cohort pushes the results in one direction, it pushes both.

So I went looking for the wider picture. A 2025 systematic review in Sleep Medicine Reviews, by Kalkanis and colleagues, pulled together 59 primary studies that looked at sleep regularity and health in adults. Its summary: consistent, moderate-certainty evidence linking greater irregularity to higher depressive and anxiety symptoms, higher body mass index, insulin resistance, hypertension, and incident cardiovascular events. Among five of the lower-bias cohorts, the least regular sleepers had 20% to 88% higher all-cause mortality, independent of sleep duration and quality.

That is a wider net, and it points the same way. Moderate-certainty, in the grading language those reviews use, is a real phrase. It is not high certainty, and the spread of 20% to 88% tells you the size of the effect is not pinned down.

Then there is the last line of that review, which matters more than any of its numbers. The authors describe regularity as modifiable and trackable on a wearable, and recommend it as a priority target for randomized intervention trials. Read that carefully. It is the polite way of saying the trials that would tell us whether changing regularity changes outcomes have not been done at a scale that settles it.

What This Evidence Cannot Tell You

This is the section I would want to read first, so I am putting it before the part where I tell you what I am doing about it.

It cannot tell you that irregular sleep caused anyone's heart attack or death. Every study I have described is observational. Nobody assigned 60,000 people to regular or irregular sleep and waited. The researchers adjusted for a long list of factors, and the heart study excluded first-year events, but adjustment is not randomization. Residual confounding is the permanent tax on this kind of work.

It cannot rule out that illness comes first. Someone with early heart disease, depression, chronic pain, or a sleep disorder can sleep irregularly because they are unwell. In that story the irregular sleep is a symptom, not a cause. Excluding the first year of follow-up helps with some of that. It does not erase it, because many conditions are brewing for years before a hospital record exists.

It cannot tell you why a person's sleep is irregular. Irregular sleep in a 65-year-old who is retired and napping is not the same exposure as irregular sleep in a 45-year-old with a newborn, or someone working rotating shifts, and an SRI score flattens all of them into one number. The score is also a single week. A bad week, travel, or a sick child can move it.

It cannot tell you the device is right. A wrist accelerometer infers sleep from movement. It cannot easily tell the difference between lying awake and lying asleep, and the heart study's own design relies on that inference. The score is a good proxy. It is still a proxy.

It cannot tell you this applies to everyone. The mortality study notes that the cohort was 97.2% white, and a research biobank may not reflect the wider population. How these findings hold for other groups, and for people whose work forces an irregular schedule, is something these data do not answer.

It cannot give you a threshold. The cut points in the heart study, around 72 and 87, are labels for slices of one cohort's distribution. I did not find anything in the abstract suggesting they are clinical thresholds, and I would not treat them as a target to hit. And the mortality study's best-supported statement is not "be regular" but "the least regular fifth fared worst."

And it cannot tell you the mechanism. The usual explanation is circadian: that a consistent schedule keeps the body's internal clocks aligned with each other and with daylight, and that repeatedly shifting the schedule is a bit like flying across time zones every few days. That is a reasonable idea. These studies did not test it. They counted bedtimes and counted deaths.

The honest summary is not "regular sleep protects you." It is "the people with the least regular sleep did worse, we don't fully know why, and nobody has yet shown that fixing it fixes the outcome."

What I Got Wrong About Measuring

Here is the part that is uncomfortable. I have spent years treating the health side of my life as a set of quantities. Grams of protein. Minutes under the bar. Hours in the fast. When I thought about sleep, I thought about the same kind of number: a score and an hour count. Seven hours, eight if I could get it.

Hours are a budget. You spend them, you check the balance, you try to hit the target. A schedule is a different kind of thing. It is a rhythm, and you cannot see a rhythm by looking at one night.

I should have known this, because I manage rhythms for a living. A platform team that runs on a predictable cadence gets something that no amount of heroics can buy: people can plan around each other. The same is true of a body. When I eat my one Friday meal at roughly the same hour, the day holds together. The fast is easier because the clock is doing some of the work that willpower would otherwise do.

I had applied that logic to food and to training and never once to the thing that sits underneath both. A ten o'clock bedtime and a six o'clock alarm are my intention. The ring and the watch know what actually happened, and I have been reading them one night at a time. That is the honest reason this study stuck with me. It did not teach me something about sleep. It showed me I had a blind spot exactly where I thought I was thorough.

I also want to be careful about what I do with it. The temptation, when a study lands on a variable you have only seen one night at a time, is to start tracking it obsessively and to feel bad about the first number. That is how a useful observation becomes another source of anxiety at 2 a.m. The studies did not show that a perfect score is the goal. They showed that the least regular fifth fared worst.

A rhythm is not a quantity. You cannot see it by looking at one night, and you cannot fix it by working harder on any one night either.

What a Regular Clock Costs

Regularity is not free, and the studies are silent on the price.

It costs the late dinner that runs long. It costs the weekend morning that would have been a lie-in. It costs, for a lot of people in leadership, the occasional flight, the on-call night, the launch week. Nobody in these data is telling you to refuse those. The question the evidence raises is narrower: across an ordinary month, does the drift go one way, and is the cause something you chose?

There is also a real trade-off with the thing most of us were taught first, which is to protect duration. If you cannot be regular, you should not conclude that sleeping less is fine. The heart study's own finding runs the other way for moderately irregular sleepers: getting enough hours appeared to help. The more careful reading is that both matter, and that regularity may have been under-weighted.

Here is where I have landed, as a working position rather than a rule.

What I Already Do, and What I'd Add

What I already do is read the score each morning and let it steer the day. A poor night means I ease off. A good one means I push. That is a reasonable use of a nightly number, and nothing in these studies says to stop.

The studies are about something a nightly score can hide. I wear a watch and a ring every night, and their data came from a wrist accelerometer. What I would add costs nothing: open the week view and look at the spread of bedtimes and wake times across seven days, not the score for one. If my ten o'clock is really 9:30 on some nights and midnight on others, I want to know that before I decide whether it matters.

If the spread is wide, I would fix the end I control most easily, which for me is the wake time, and let the bedtime follow. That is a judgment call, not a finding. The studies measured the pattern, not which end of it to anchor.

I would leave the Friday fast and the lifting schedule alone. The point is to learn something, not to rebuild a week that works.