Short Sleep and Long-Term Obesity Risk

The odds ratios are large. The kilograms are not.#

Habitual short sleep does predict becoming heavier, in big prospective cohorts, after adjustment, replicated across populations. It is one of the more durable findings in nutrition epidemiology. What almost never travels with the headline is the size of it, and the size changes what you should do about it.

The cleanest long-run measurement comes from the Nurses' Health Study, where 68,183 women reported their usual sleep duration in 1986 and were followed for sixteen years. Adjusted for age and starting BMI, women sleeping five hours or less gained 1.14 kg more than seven-hour sleepers over those sixteen years (95% CI 0.49–1.79); six-hour sleepers gained 0.71 kg more (95% CI 0.41–1.00). Their relative risk of a 15-kg weight gain was 1.32 (95% CI 1.19–1.47), and of becoming obese, 1.15 (95% CI 1.04–1.26). The associations survived adjustment for physical activity and dietary intake, and the authors describe the result in their own conclusion as "a modest increase in future weight gain"¹.

That is the finding to hold in your head: a sixth of a kilogram per decade of extra gain, alongside a 15% relative increase in the odds of crossing the obesity threshold. Both numbers are true, and they feel like they belong to different stories. The pooled odds ratio everyone quotes — short sleepers roughly 25–45% more likely to become obese — is the pillar's territory. This article is about what that ratio is made of.

Sixteen years of short sleep, priced in calories a day#

Run Patel's headline number through an energy ledger and it becomes almost comically small. If those 1.14 kg were entirely fat, at roughly 7,700 kcal per kilogram of fat tissue, that is about 8,800 kcal accumulated across 5,844 days — around 1.5 calories a day. That arithmetic is ours, not the paper's, and it leans on two simplifications: real weight gain is a mix of fat and lean tissue, and the 7,700-kcal-per-kilogram figure is itself an approximation that overstates long-run requirements. Even generously wrong, the answer stays in single digits.

Now set that beside the laboratory result. Controlled sleep-restriction experiments produce a surplus of a few hundred calories a day — the pooled figure is 385 kcal, traced link by link in how sleep loss drives weight gain. Two hundred times the cohort effect. Both literatures are competent, and the gap between them is the most interesting thing on this page. Three readings explain it, and they are not mutually exclusive:

Energy balance is self-correcting. A surplus does not accumulate forever. As body mass rises, so does the energy cost of carrying and maintaining it, so a persistent few hundred calories a day buys a new, slightly heavier steady state rather than unbounded gain. A small permanent difference in equilibrium weight is exactly what a cohort would measure.
The two exposures are different. Four hours in bed for five nights with food freely available is not the same thing as answering "about five" on a questionnaire in 1986. One is an acute provocation; the other is a life.
The measurement is noisy in a way that shrinks estimates. This is the part worth its own section.

The exposure is a memory, not a measurement#

Every number in Patel's study rests on women recalling how long they usually sleep. That variable has been validated against actigraphy in the same research program, and it does not hold up well. Among 669 middle-aged adults, self-reported and wrist-measured sleep duration correlated at just r = 0.47. Measured sleep averaged 6.0 hours; reported sleep averaged 6.8. Reports rose by only 34 minutes for each additional hour of measured sleep — and the overstatement was worst at the short end, where people sleeping five measured hours over-reported by roughly 1.2 hours against about 0.4 hours for seven-hour sleepers².

The authors draw the conclusion themselves: "the true associations between sleep duration and health may differ from previously reported associations between self-reported sleep and health."

This is not the usual hand-wave about imperfect questionnaires. Random misclassification blurs an exposure and biases estimates toward zero — which would mean the true effect of short sleep is larger than Patel measured. But the error here is not random: it scales with how short the sleep is, so the self-reported five-hour group is systematically populated by people sleeping something else. That does not tell you which way the estimate is wrong. It tells you the group labels are not what they say, and the honest posture is a wider error bar than the published confidence interval implies.

The one cohort that measured sleep instead of asking about it#

So what happens when you run the prospective analysis with actigraphy? It has been done, in 612 middle-aged adults from the CARDIA cohort followed between 2000 and 2006, with sleep measured across several nights rather than reported.

Cross-sectionally, the familiar association appeared: shorter sleep and greater fragmentation both tracked higher BMI, and after adjustment, BMI fell by 0.78 kg/m² per increasing sleep category (95% CI −1.6 to −0.002). Read that interval — its upper bound is two thousandths away from zero, which is a statistically significant result that is barely one. Longitudinally, the study reports plainly that there were no longitudinal associations between sleep measurements and change in BMI across five years³.

One subgroup finding in that paper is the most informative line in this whole literature. The cross-sectional sleep-BMI association was strong among snorers and weak among non-snorers. Snoring is the everyday signature of sleep-disordered breathing, and sleep-disordered breathing is caused by excess weight at least as reliably as it causes it. The clearest slice of the association sits precisely where the arrow most plausibly runs backwards.

Study	How sleep was measured	Design	What it found
Patel, 2006 (n = 68,183)	Self-report, single question, 1986	16-year prospective	≤5 h: +1.14 kg; RR of obesity 1.15
Lauderdale, 2008 (n = 669)	Self-report vs actigraphy	Validation	r = 0.47; reported 6.8 h vs measured 6.0 h
Lauderdale, 2009 (n = 612)	Actigraphy, several nights	5-year prospective	Cross-sectional link yes; change in BMI, none

Those three rows are not three attempts at the same experiment. They are a large, long, weakly-measured study; a check on the measurement; and a small, shorter, well-measured study. The first has the statistical power and the wrong instrument. The third has the right instrument and may simply lack the years — five is not sixteen, and 612 is not 68,183. Neither can settle it alone, and pretending otherwise in either direction is where this topic usually goes wrong.

Which way the arrow points#

Beyond measurement, two structural problems sit under every estimate here.

Reverse causation. Carrying more weight promotes sleep-disordered breathing, reflux, joint pain and nocturia, all of which shorten and fragment sleep. Genetic analyses in the pillar find no causal effect of sleep duration on adult BMI while finding insomnia symptoms raising it — and a return leg running from higher BMI to more daytime sleepiness. The traffic on this road moves both ways.

Confounding that adjustment cannot fully remove. People who habitually sleep five hours differ from people who sleep seven in shift schedules, income, caregiving load, depression, chronic pain, and alcohol use. Patel's analysis controlled for diet and activity and the association held, which is a point in its favor. It is not the same as controlling for the life that produced the short sleep — and one of those lives, rotating and night shift work, carries its own metabolic exposure independent of hours slept.

What a modest, real, badly measured risk is worth#

The practical reading is not "sleep doesn't matter." It is that chronic short sleep belongs in the same tier as other durable lifestyle risk factors: real over a lifetime, worth fixing, and far too small to be the explanation for anyone's individual weight.

That also means the reason to protect sleep is the near-term one rather than the actuarial one. The mechanism you can act on this week is the intake side — a few hundred surplus calories arriving after 10pm, and a plan that gets harder to follow — not a sixteen-year risk ratio. Aiming for the seven-hour floor is worth doing because next Tuesday goes better, and the epidemiology is a footnote confirming the direction rather than the argument for it.

And it argues against treating hours in bed as insurance. The prospective study that measured sleep properly found no relationship with subsequent BMI change at all, which is a thin foundation for lying awake trying to bank a number. Sleep is a habit worth keeping for the same unglamorous reasons every other durable habit is: it makes the next day easier to run well. That is a smaller claim than the headlines make, and it is the one the evidence actually supports.

FAQ#

Does short sleep cause obesity, or do heavier people just sleep worse?#

Both appear to be happening, and the strongest clue sits in a subgroup. In the one prospective cohort using actigraphy rather than recall, the cross-sectional link between short sleep and higher BMI was strong among snorers and weak among non-snorers³ — and snoring marks sleep-disordered breathing, which excess weight causes. That same study found no association between measured sleep and subsequent BMI change.

How much weight does chronic short sleep actually add over years?#

About a kilogram over a decade and a half, in the largest study to measure it. Women sleeping five hours or less gained 1.14 kg more than seven-hour sleepers across 16 years (95% CI 0.49–1.79), and six-hour sleepers 0.71 kg more¹. The relative risk of becoming obese was 1.15. The odds ratio sounds alarming; the kilograms are the same finding measured in a unit you can picture.

Why do lab studies show hundreds of extra calories a day but cohorts show almost no weight gain?#

Because a surplus does not compound indefinitely — a heavier body costs more energy to maintain, so a persistent excess settles at a slightly higher weight instead of climbing forever. On top of that, five nights at four hours in a lab is a different exposure from a lifetime of answering "five" on a questionnaire, and self-reported sleep correlates with measured sleep at only r = 0.47². All three effects push the long-run estimate down.

Sources#

This article was researched and drafted with AI assistance and reviewed for accuracy by the BurnWeek team. It is general information, not medical advice. How we research and correct our articles →