What short sleep does to a diet

In short

Same diet, different sleep, same scale reading: 2.9 kg lost on 8.5 hours in bed, 3.0 kg on 5.5 hours.
Composition diverged. Fat was 1.4 kg of that loss on adequate sleep and 0.6 kg on short sleep; fat-free mass 1.5 kg against 2.4 kg. The share lost as fat fell from 56% to 25% (−31 percentage points, 95% CI −49 to −12).
Ten people did that study, and the standard deviations on those percentages were 35 and 24 points. Read the direction, not the decimal.
Short sleep raises measured intake: pooled +385 kcal/day (95% CI 252 to 517). Extending it cut intake by 270 kcal/day.

An experiment that held the diet constant

Most of what is written about sleep and body weight rests on epidemiology: people who sleep less tend to weigh more. The association is consistent and largely uninformative, because short sleep travels with shift work, stress, pain, poverty and illness. The useful evidence is a much smaller experimental literature.

The most careful example is a randomised crossover study from the University of Chicago, in which each person served as their own control.¹ Ten overweight but otherwise healthy adults, mean age 41 years (SD 5) and mean body mass index 27.4 kg/m² (SD 2.0), spent two fourteen-day periods living in a laboratory, in random order and at least three months apart, with nightly time in bed set at 8.5 hours in one and 5.5 in the other. Everything else was matched: the same individualised diet in both, restricted to 90% of measured resting metabolic rate, food weighed before and after every meal. Intake came out at 1,447 kcal/day (SD 227) under long sleep and 1,450 (SD 236) under short; polysomnography put actual sleep at 7 h 25 min against 5 h 14 min.

Same weight, different weight

Total weight loss came out effectively identical: anyone reading only the scale would have concluded that sleep made no difference. Absorptiometry disagreed.

Table 1. The same ten adults, the same fourteen-day diet, two sleep doses. Means (SD); differences are model coefficients.
Outcome	8.5 h in bed	5.5 h in bed	Difference (95% CI)	P
Weight lost, kg	2.9 (1.4)	3.0 (1.0)	0.2 (−0.2 to 0.7)	0.24
Fat lost, kg	1.4 (0.9)	0.6 (0.6)	−0.7 (−1.4 to −0.03)	0.043
Fat-free mass lost, kg	1.5 (1.3)	2.4 (1.4)	1.0 (0.4 to 1.5)	0.002
Weight lost as fat, %	56 (35)	25 (24)	−31 (−49 to −12)	0.004
Resting metabolic rate, kcal/day	1,505 (262)	1,391 (180)	−147 (−253 to −41)	0.010

Figure 1. Blue is the 8.5-hour condition, grey the 5.5-hour; bar length is kilograms lost, on one scale. The top pair is what a bathroom scale would show; the pairs below are what it was made of.

What the widely quoted figure actually counts

The study is usually cited through one sentence of its abstract: sleep curtailment "decreased the proportion of weight lost as fat by 55%". It pays to know which quantity that refers to, because the retellings drift. It is a relative reduction in the fat share of weight loss — 25% is 55% lower than 56%. The parenthetical that follows, 1.4 versus 0.6 kg, is the absolute mass of fat lost, and it falls by a similar proportion only because total weight loss was nearly identical. Neither figure means anyone lost 55% less weight.

Then the qualifications. Ten people completed the protocol, of twelve enrolled, and the standard deviations around 56% and 25% were 35 and 24 percentage points, so individual responses overlapped heavily. The confidence interval on the difference in fat lost ran from −1.4 to −0.03 kg — it brushes zero, and the P value of 0.043 sits where you would expect. Baseline body fat also differed between conditions (26.4 versus 25.0 kg), which is why the analysis used mixed models with baseline composition as a covariate. The authors' limitation section is one sentence: "The nature of the study limited its duration and sample size."

The sturdier result is the one about lean tissue: loss of fat-free mass was 60% greater under sleep restriction, with a confidence interval well clear of zero (1.0 kg, 95% CI 0.4 to 1.5). Two other measurements agree — resting metabolic rate finished 147 kcal/day lower, and the fasting respiratory quotient rose from 0.80 to 0.83 (P = 0.042), consistent with less fat being oxidised.

The replication is directionally similar and much messier

A later randomised trial took the question out of the metabolic ward.² Thirty-six adults completed eight weeks of caloric restriction alone (n = 15) or with sleep restriction (n = 21), the second group losing a net 169 minutes (SD 75) of sleep a week. Weight, fat mass and lean mass all changed by indistinguishable amounts — −3.3 kg (SD 3.2) against −3.2 (SD 2.5) for weight, with p ≥ 0.85 on all three comparisons. Only the ratio moved. The proportion of lost mass that was fat was 80.7% (SD 43.3) with unrestricted sleep and 16.9% (SD 104.5) with restricted sleep (p = 0.016). Those standard deviations are not typographical errors: over changes this small, absorptiometry throws individual values above 100% and below zero, which is why the authors also gave medians, 82.7% and 58.4%. The direction agrees with the laboratory study; the size of the gap depends on which summary you take.

Why people eat more when they sleep less

The appetite side has an older and cleaner literature. In twelve healthy young men, two days of curtailed sleep against two days of extended sleep — caloric intake and physical activity held constant — lowered leptin, raised ghrelin, and raised reported hunger and appetite by roughly a quarter, the largest increases (33% to 45%) being for calorie-dense high-carbohydrate foods.³ All reached significance at P ≤ 0.04. Because the food was fixed, they measure wanting rather than eating.

Allowed to eat freely, people do eat more. A meta-analysis of partial sleep deprivation found intake higher by 385 kcal/day, with no significant change in total energy expenditure or resting metabolic rate, and the surplus skewed towards fat and away from protein.⁴ Seventeen studies were reviewed (n = 496); only eleven, 172 participants, could be pooled, most of them brief laboratory protocols.

Table 2. Experimental evidence on sleep and energy intake; ranges are 95% CI. The first two rows shorten sleep, the third lengthens it.
Study	Design	Effect on intake or appetite
Spiegel et al., 2004	Crossover, 12 men, intake fixed	Hunger +24%, appetite +23%; leptin −18%, ghrelin +28%
Al Khatib et al., 2017	Meta-analysis, 11 studies, n = 172	Intake +385 kcal/day (252 to 517)
Tasali et al., 2022	Randomised trial, n = 80, free-living	Intake −270 kcal/day (−393 to −147)

The strongest design runs the manipulation the other way.⁵ Eighty adults with overweight, all habitually sleeping under 6.5 hours, were randomised after a two-week baseline either to a single individualised sleep-hygiene counselling session aimed at 8.5 hours in bed, or to carry on unchanged. No diet or exercise was prescribed and everyone stayed at home; intake was measured as doubly labelled water expenditure plus the change in body energy stores. Sleep rose about 1.2 hours a night (95% CI 1.0 to 1.4), intake fell against control, expenditure did not shift, and weight differed by −0.87 kg (95% CI −1.39 to −0.35).

Worth noting what it did not show. Its composition split ran opposite to the laboratory study: of that 0.87 kg, 0.62 kg was fat-free mass and 0.25 kg fat. Over a fortnight and under a kilogram of change, absorptiometry measures hydration as much as tissue, and composition was a secondary outcome the trial was not powered for.

Where the evidence stops

A systematic review of randomised trials manipulating sleep duration found only eighteen qualifying studies in the entire literature — eight with a body-weight outcome, four on food intake, four on total energy expenditure, three on respiratory quotient, four on leptin or ghrelin.⁶ Its conclusion: "sleep restriction increases food intake and total energy expenditure with inconsistent effects on integrated energy balance as operationalized by weight change".

That review and the meta-analysis disagree about expenditure, which is instructive rather than embarrassing: they cover different sets of trials searched years apart, one counted studies by direction while the other pooled effect sizes, and expenditure is measured with an error band comparable to the effect sought. The crossover study makes the point against itself. Its fat-loss difference implies a gap of roughly 400 kcal/day between conditions, and its own doubly labelled water measurement could not resolve one: the within-person standard deviation came out at 340 kcal/day against the 140 assumed in the power calculation.

So the defensible reading is narrow. Sleep modifies what a calorie deficit is made of, and how much people eat when eating is left to them. It is not itself a weight-loss intervention: shortening sleep did not produce more weight, and lengthening it produced under a kilogram in two weeks.

The bottom line

If you are dieting, the case for protecting sleep is not that it burns anything. It is that the same deficit appears to cost more lean tissue when you are short of sleep, and that short sleep pushes measured intake up by a couple of hundred calories a day — enough to cancel a modest deficit unnoticed. Both come from small, brief studies, and the composition finding rests on ten people. Treat sleep as a condition under which a diet works better, not a substitute for one.

The measurable half of this is the intake half, and it is the half people estimate worst. Kettle logs food macros — protein, carbohydrates, fat and energy — alongside your training and water on the iPhone, with the data staying on the device.