Sleep and glucose metabolism: what the trials actually measured

Two inpatient sleep-restriction studies and one six-week randomized crossover in women report the same direction on insulin sensitivity, with effect sizes you can quote. The cohort literature agrees on association and refuses to say cause.

goodWell supported, with the usual caveats about effect size.Revised 16 Oct 2026

A small bedside clock with a completely blank face on a folded pale blue cotton sheet beside a plain glass of water

Sleep is the one variable in a longevity protocol you cannot supplement around. Every trial below cut it on purpose and measured the consequence with a clamp or a glucose tolerance test.

Sleep is the input most people in this field try to optimise last and complain about first. The glucose metabolism literature is unusually good news on that front, because it is one of the few corners of longevity science where a variable was removed in a controlled setting and the consequence measured with an instrument rather than inferred from a self-report.

Two inpatient studies that cut sleep on purpose

Buxton and colleagues, in Diabetes in 2010, ran a twelve-day inpatient study on twenty healthy men aged 20 to 35 with a BMI between 20 and 30 kg/m2. Subjects spent ten hours in bed per night for at least eight nights, including three inpatient nights, and then five hours in bed for seven nights. Diet and activity were controlled. On the last two days of each condition they ran an intravenous glucose tolerance test and a euglycemic-hyperinsulinemic clamp.

The clamp is the important one. Insulin sensitivity derived from the IVGTT fell by 20 ± 24 percent after sleep restriction (P = 0.001), and sensitivity by clamp fell by 11 ± 5.5 percent (P < 0.04). Glucose tolerance and the disposition index both fell. What did not change: the insulin secretory response, and the results were not affected by modafinil, the alerting drug given during restriction. Salivary cortisol rose by 51 ± 8 percent with restriction (P < 0.02), but changes in insulin sensitivity did not correlate with cortisol changes, nor with urinary catecholamines, nor with slow wave sleep.

That last line is the detail most write-ups skip. The mechanism people assume was not the mechanism the data pointed at.

Six weeks, in women, at home

The 2024 Diabetes Care crossover from Columbia is the more realistic design and the one that travels better to a normal week. Thirty-eight women aged 20 to 75, none with cardiometabolic disease, all with actigraphy-confirmed habitual sleep of seven to nine hours, went through two six-week phases: adequate sleep, and restriction of 1.5 hours per night. Eleven were postmenopausal.

Total sleep time fell by 1.34 ± 0.04 hours per night relative to the adequate phase. Fasting insulin rose by 6.8 ± 2.8 pmol/L (P = 0.016) and HOMA-IR rose by 0.30 ± 0.12 (P = 0.016). The HOMA-IR effect was more pronounced in postmenopausal women than in premenopausal ones (β = 0.45 ± 0.25 versus 0.27 ± 0.13; P for interaction = 0.042).

The authors’ conclusion is the sentence worth keeping: curtailing sleep to 6.2 hours per night — which they note is the median sleep duration of US adults with short sleep — for six weeks impairs insulin sensitivity, independent of adiposity. Change in adiposity did not mediate the effect and did not change the results when included as a covariate.

What the cohorts add, and what they cannot

Observational data on the long-sleep side comes from Jike and colleagues, in Sleep Medicine Reviews 2018: 137 prospective cohort studies, 5,134,036 participants. Long sleep was significantly associated with mortality (RR 1.39, 95% CI 1.31 to 1.47), incident diabetes mellitus (1.26, 1.11 to 1.43), cardiovascular disease (1.25, 1.14 to 1.37), stroke (1.46, 1.26 to 1.69), coronary heart disease (1.24, 1.13 to 1.37) and obesity (1.08, 1.02 to 1.15). It was not significantly related to incident hypertension (1.01, 0.95 to 1.07), and there were insufficient data for depression and dyslipidaemia.

Read the confidence intervals rather than the point estimates. Obesity at 1.08 and hypertension at 1.01 are the same finding told in two different voices: the signal is thin. Mortality at 1.39 is not thin. And the authors say plainly in their conclusion that future studies should address whether the relationship is causal and modifiable — which means they do not claim it is either.

The one study in this set that is about timing, not duration

Allison and colleagues’ 2021 Current Biology crossover is often filed under sleep and belongs here for a reason. Twelve healthy adults, mean age 26.3 ± 3.4 years, BMI 21.9 ± 1.7 kg/m2, five of them female, ate three meals and two snacks of comparable energy and macronutrient content under two counterbalanced eight-week conditions, with sleep-wake cycles and exercise held constant. The only thing that moved was the clock: intake limited to 0800–1900 h, or to 1200–2300 h.

On the daytime schedule, body weight, HOMA-IR, trunk-to-leg fat ratio, resting energy expenditure, respiratory quotient, fasting glucose, fasting insulin, total and HDL cholesterol, and adiponectin all decreased; triglycerides increased on the delayed schedule. Effect sizes ran d = 0.397 to 1.019.

The result that should give a careful person pause is in the same abstract: circadian phase and amplitude of melatonin, cortisol, ghrelin, leptin and glucose were not differentially altered by the two eating schedules. The metabolic measurements moved and the clock hormones did not.

A boundary case worth naming

Yuan and Larsson’s 2023 epidemiology review of sarcopenia lists extreme sleep duration among the factors associated with increased risk — alongside physical inactivity, malnutrition, smoking and diabetes. Sarcopenia is estimated to affect 10 to 16 percent of the elderly worldwide, and the prevalence ranges from 18 percent in diabetic patients to 66 percent in patients with unresectable oesophageal cancer.

The same review is careful in a way most summaries are not: those risk-factor associations rest mainly on non-cohort observational studies and need confirmation. High-quality cohort, omics and Mendelian randomisation studies are needed. So when someone cites this review to say sleep destroys muscle in the elderly, they are citing a hypothesis the authors flagged as one.

Sleep as a retail category runs the same exercise in a different direction, and what the DSIP sleep-peptide trials measured separates trial endpoints from the language the compound is sold in.

What the data do not show

They do not show a threshold. No study above identifies the point at which sleep loss becomes metabolically harmless, and the Buxton design — ten hours in bed, then five — is a laboratory contrast, not a description of an ordinary week. They do not show that restoring sleep reverses the effect; Buxton tested modafinil during restriction, not recovery afterwards. They do not show that anyone who sleeps six hours a night will develop diabetes: the randomized trials measured insulin sensitivity in weeks, and the cohorts measured incidence over follow-up years without establishing direction of cause. And the daytime-versus-delayed eating study, with twelve participants, was not designed to detect anything about how a real person should arrange their evening.

This article summarises published trials and cohort meta-analyses. It is not medical advice and it is not a recommendation for any individual.