When Should Night-Shift Workers Eat?

The trial randomized 19 healthy young adults inside a laboratory across four 28-hour days of forced desynchrony designed to simulate night work, in dim light of roughly 3 lux. One group kept every behavior including eating on the 28-hour cycle, so meals drifted across day and night. The other kept everything on the 28-hour cycle except eating, which stayed on a 24-hour daytime cycle. Calories, macronutrients, physical activity and posture were matched between arms, so the tested variable was timing alone (Chellappa et al., 2021).

The 3-hour glucose response to a breakfast test meal, a measure on which a higher value means poorer glucose tolerance, increased by 19.4% in the night-eating arm versus its own baseline (95% CI, 4.7 to 34.2%; 18.4 mg/dl, 95% CI, 4.8 to 31.8 mg/dl; P = 0.002), while the daytime-eating arm showed no significant change from its baseline (95% CI, -13.9 to 10.1%). The paper does not leave that contrast to a comparison of two within-arm p values. It reports a formal between-arm test, and meal timing significantly modified the impact of simulated night work on the breakfast glucose profile (interaction p FDR = 0.003), which is what licenses saying daytime eating prevented the change at that meal. A second and separate outcome, the 28-hour average glucose level, rose 6.4% in the night-eating arm (95% CI, 2.7 to 10%; P = 0.003), did not change significantly with daytime eating, and was also significantly modified by the meal-timing intervention. Those two figures describe different things and should never be swapped for each other.

The result was meal-specific, and insulin moved with it. After the dinner test meal there was no significant interaction and no significant within-arm change in either glucose or early-phase insulin, which the authors attribute to opposing influences of circadian misalignment and circadian phase. At breakfast, meal timing significantly modified the early-phase insulin response (interaction p FDR = 0.008): it fell 52.9% in the night-eating arm (95% CI, -98.6 to -7.1%; -23.5 uU/ml, 95% CI, -42.8 to -4.2 uU/ml; P = 0.01) and was not significantly affected in the daytime-eating arm (95% CI, -39.8 to 3.4%). Late-phase insulin showed no significant interaction at either meal, and average insulin across the 28-hour cycle was not significantly modified, so the insulin finding here is early-phase and confined to breakfast.

In the night-eating arm the endogenous circadian glucose rhythm shifted by 9.81 hours as a mean direction, with a circular variance of 0.11. In the daytime-eating arm it shifted 0.57 hours, and the difference between arms was significant on a Watson-Williams F test (P < 0.001), another direct comparison of the two groups rather than of each group against itself. Those dispersion figures come from circular statistics and are not confidence intervals or standard deviations. The authors describe the night-eating shift as closely matching the 12-hour shift of the imposed sleep and wake cycle, so 12 hours is the behavioral displacement, not the measured rhythm shift.

The central clock tells a different story. Core body temperature phase moved 0.63 hours in the night-eating arm and 0.57 hours in the daytime-eating arm, a difference that was not significant. The central clock was barely displaced in either group over this protocol, so the intervention did not protect it. What daytime eating prevented was the peripheral glucose rhythm pulling away from a central clock that stayed roughly where it was. Rhythm amplitudes did not change significantly for either glucose or core body temperature.

Randomization licenses attributing the trial's own biomarker changes to the assigned meal-timing protocol, but the setting was simulated night work over 14 laboratory days, and the participants were healthy young volunteers with a mean age near 27, body mass index between 18.5 and 29.9, and HbA1c between 4.9 and 5.4%. They were not shift workers. There were no commutes, no daylight, no social schedules and no family meals. The study measured intermediate biomarkers only, with no diabetes incidence, no weight outcome, no clinical endpoint and no long-term follow-up.

The observational literature runs alongside that trial rather than confirming it. Pooling 28 studies, Sun et al. (2018) reported that night shift work was associated with higher odds of obesity or overweight, with an odds ratio of 1.23 (95% CI, 1.17-1.29). An odds ratio of 1.23 means 23% higher odds, not 23% more cases. The estimate was weaker in cohort designs, at a risk ratio of 1.10, than in cross-sectional ones, at an odds ratio of 1.26, a pattern consistent with residual confounding or reverse causation, and the literature search ended in March 2017.

Those pooled figures cannot tell you that meal timing explains the association. Sun et al. (2018) measured obesity rather than glucose tolerance, and reported higher odds among permanent night workers, at an odds ratio of 1.43, than among rotating shift workers, at an odds ratio of 1.14, both associational and both on the odds scale, along with an odds ratio of 1.35 for abdominal obesity. Those subgroup estimates are published as point estimates without confidence intervals, so treat them as weaker than the pooled figure and as association only.

Sample size is the first limit. Ten participants ate at night and nine ate in the daytime, after 24 people entered the laboratory protocols, 4 discontinued with mild adverse events unrelated to the intervention, 20 were randomized and one was excluded for being unable to consume all meals. The breakfast glucose interval runs from 4.7% to 34.2% and the early-phase insulin interval from -98.6% to -7.1%, so both point estimates are imprecise and support no promise about how any individual worker would respond.

Fasting duration also differed between arms before the breakfast test meal during simulated night work, at roughly 16 hours for night eaters versus roughly 12 hours for daytime eaters. The authors argue both sit within American Diabetes Association guidance of 8 to 16 hours, and cite evidence that 12-hour and 36-hour fasts did not differ in glucose area under the curve, but the difference travels with the result. Blinding was partial by necessity, since participants knew they might eat at unusual times but not that separate meal-timing groups existed, and the trial ran at a single center. Sleep structure did not differ between groups and did not significantly modify the meal-timing effects.

The funding and disclosure picture deserves a line. The work was funded by the National Institutes of Health, and several authors report commercial ties: one is an employee of and holds stock in Biogen, stated as unrelated to this work, and others report lecture fees, consulting fees, or institutional grants and gifts from pharmaceutical, device and airline sponsors. None of that invalidates a randomized result, but anyone turning this into workplace nutrition messaging should know it is there.