Your dinner moved your deep sleep by two and a half minutes

A study went round the internet this month with a headline that sounds like a gift: eat more protein and more fibre, and you will sleep longer and better. It is the rare piece of health news that asks nothing of you except dinner.

So I went and read what is underneath it. There are two papers, not one. The first got its sleep data from a Pokémon game. The second is the most careful study anyone has run on this question — and the protein half of the headline did not survive its statistics.

Neither of those facts makes the finding junk. Something real is in there. It is roughly a tenth the size of what the headline implies, and it is not in the place everyone is looking.

The most careful version of this experiment yet

In August, a group led by Mariya Shkolnik and Hagai Rossman, with Eran Segal's lab at the Weizmann Institute, published an analysis of 4,793 nights from 3,598 adults. Average age fifty-three, roughly half men, mostly healthy. Each person logged meals in real time in an app, and wore a home sleep-test device overnight that reads the pulse in the finger rather than brain waves.

The design is the interesting part. You cannot randomise three and a half thousand people's dinners, so the authors did the next best thing: they used the fact that the same person eats differently on different days, then statistically re-weighted the data so that high-fibre days and low-fibre days resembled each other on everything else that was measured. It is called a target-trial emulation, and it is the best available substitute for a coin toss. It is not a coin toss.

The result: on days when people ate more fibre per calorie, that same night contained more deep sleep and more REM, less light sleep, and an overnight heart rate about one beat per minute lower. All four intervals cleared zero and survived correction for the many things the team tested.

Now do the arithmetic

Here are those numbers in the unit the study reported them in — percentage points of the night — and then in the unit you actually live in.

A percentage point of a seven-hour night is about four minutes. So the headline finding, stated in units a human can feel, is this: a day of eating well moved roughly six minutes of one night out of light sleep and into deep sleep and REM.

That is real. It is also smaller than the difference between your Tuesday and your Wednesday.

Worth saying that an older and much smaller study pointed the same way, with actual controlled food. In 2016 Marie-Pierre St-Onge's group kept twenty-six adults inpatient, fed them a prepared diet for four days, then let them choose their own food on the fifth. The self-chosen day was followed by less slow-wave sleep and a longer time to fall asleep, and within that day more fibre predicted more slow-wave sleep while more saturated fat predicted less. Twenty-six people and a single night of free eating is not much. But a tiny controlled study and a large emulated one agreeing is worth more than either alone.

The protein half of the headline is the weak half

Now the part that did not make the coverage. In the same Nature Health analysis, macronutrient distribution — protein, fat and carbohydrate as shares of energy — produced no robust associations once the authors corrected for how many things they had tested. The plant-forward features of the diet survived. The macros did not.

The protein claim in the headlines comes from a different paper: Seol and colleagues, published in the Journal of Medical Internet Research at the start of 2025. They matched 4,825 people who logged their food in a diet app with their sleep records from Pokémon Sleep, a consumer app built around the game. People in the top quartile for protein intake averaged about eleven minutes more sleep than those in the bottom quartile.

This is not a knock on the researchers, who were careful and clear about what they had. It is a note about what the data can carry. The study is cross-sectional — everything measured once — so it cannot tell you whether eating more protein lengthened anybody's sleep, or whether people who sleep longer simply eat differently. And the sample is people who voluntarily log every meal and track their nights through a Pokémon game, adjusted only for age, sex and body mass index.

Eleven minutes, measured that way, is a hypothesis. The coverage promoted it to advice.

The number worth watching isn't a sleep stage

The most interesting result in the whole business is the one nobody headlined, because heart rate is boring.

Three different plant-forward measures — fibre per calorie, whole-plant food intake, and the sheer variety of plants eaten — each moved overnight heart rate in the same direction, by roughly one beat per minute. Three overlapping but non-identical exposures agreeing is a different and sturdier kind of evidence than one number crossing a significance threshold.

It also happens to be the finding you can actually check on yourself, for a reason that has nothing to do with biology. A wrist device estimates your sleep stages, inferring them from movement and pulse; it measures your heart rate directly. That distinction is the whole game, and it is the difference between what your wearable knows and what it guesses.

So if you want to test any of this on yourself, do not go hunting for deep-sleep minutes. That is the noisiest number on your device chasing the smallest effect in the study — a guaranteed way to find nothing, or to find something that isn't there. Watch your overnight heart rate instead, over a fortnight of ordinary dinners. If you track that trend with the Agen Band, you are looking for the shape of two weeks, not a verdict on last night. That is the general rule for reading sleep data without letting it read you.

The dinner trade-off nobody printed

Two more findings from the same study, which are far more useful sitting next to each other than apart.

A heavier evening meal was followed by about eight minutes more total sleep — and an overnight heart rate about 0.7 beats per minute higher. Eating dinner earlier ran the other way: roughly twelve minutes less sleep, and an overnight heart rate about 0.7 beats per minute lower.

So a large late dinner buys you minutes in bed and charges you a slightly busier night. An early, lighter one does the reverse. There is no item on the menu where both numbers improve — which is a small sign the measurements are honest, because free lunches in observational data are usually artefacts.

Which trade you want depends on which thing you are short of, and that is genuinely a personal question rather than a hidden right answer. It is the same tension that runs through everything we know about when you eat versus what you eat.

The arrow everyone drew points the other way

Now step back, because there is a much larger fact standing just outside the frame of all this coverage.

The diet-to-sleep evidence is observational and small. The sleep-to-diet evidence is randomised and large.

In 2022, Esra Tasali's group at Chicago took eighty adults who habitually slept under six and a half hours and randomised them either to a single session of personalised sleep counselling aimed at extending their time in bed, or to carry on as they were. No diet was prescribed to anyone. Sleep duration rose by about 1.2 hours a night in the extension group. Their energy intake fell by 270 kilocalories a day, with an interval running from −393 to −147.

And that intake figure was not self-reported. It was derived from doubly labelled water and measured changes in the body's energy stores — a method that does not care what anyone writes in an app.

Set the two literatures side by side and the asymmetry is almost comic. On one side, a day of virtuous eating buys a couple of minutes of deep sleep, observed. On the other, an extra hour of sleep changes what you eat the following day by the size of a small meal, in a randomised trial measured with isotopes.

We have been holding the arrow backwards. Dinner is the lever people reach for because dinner is a decision, and sleep feels like weather. But the direction with the strong causal evidence runs the other way — and it is the direction nobody turns into a headline, because "go to bed earlier" does not sell anything. See also: what you can and cannot repay at the weekend, and why regularity tends to beat duration.

What to do with this in the next fortnight

The bottom line

Does what you eat change how you sleep? Yes, slightly, and the fibre part of the story is the part to believe. A more plant-forward day is followed by a marginally deeper, marginally calmer night — on the order of minutes of sleep architecture and about one beat per minute of overnight heart rate. The protein part of the headline rests on a cross-sectional study of app users and vanished in the better-designed analysis.

The useful correction is not that the finding is too small to matter. It is that the interesting number was the heart rate, not the sleep stage — the whole-body signal, not the one the sleep-tracking industry taught everybody to watch. Deep sleep got famous because it is the number a device can put on a screen. Overnight heart rate is less photogenic and more honest, and it is the one that moved the same way three times.

And the biggest lever in the room was never on the plate. It is the hour you did not go to bed. Dinner is worth about six minutes of better sleep. Sleep is worth about a small meal of better eating. Everyone keeps writing about the first one.

If you are assembling the wider picture, this sits alongside the caffeine you drank at two, the drink you had at nine, and the rest of the longevity protocol.