Does Melatonin Help Night Shift Workers Sleep?

Nine of the 15 trials evaluated melatonin, at doses of 1 to 10 mg taken after the night shift. Pooled across seven trials with 263 participants, daytime sleep length showed a mean difference of 24 minutes versus placebo (95% CI 9.8 to 38.9). Three trials with 234 participants found a mean difference of 17 minutes for night-time sleep length (95% CI 3.71 to 30.22). Both intervals exclude zero, so the direction is consistent (Liira et al., 2014).

Consistency is not the same as certainty. The review authors graded both findings as low quality evidence under GRADE, and their own conclusion opens by describing "low quality evidence that melatonin improves sleep length" after a night shift. They point to small trials, incompletely described randomization and allocation concealment, and limited outcome reporting. Copy that reports the 24 minutes without the grade is reporting half the finding.

Sleep onset latency did not move. Across five trials with 74 participants, the mean difference was 0.37 minutes with a 95% CI of -1.55 to 2.29, an interval that spans zero and is measured in seconds either way (Liira et al., 2014). That is the opposite of the everyday framing, in which melatonin is described as something that helps a person get to sleep. In this synthesis it did not, and no other sleep quality parameter improved either.

The review also found no dose-response effect anywhere across the 1 to 10 mg range it examined. The intuition that a larger dose does more work has no support here, and a schedule that leaves too few daytime hours for sleep is not repaired by raising a dose. The finding that survives is narrow: slightly longer sleep, at low certainty, and nothing else.

The same review covered two trials of hypnotics, one of modafinil, two of armodafinil, and one of caffeine plus naps. Zopiclone produced a daytime sleep length mean difference of 44.0 minutes, but the 95% CI ran from -22.67 to 110.67 in a single trial of 28 participants, so that estimate is fully compatible with no effect. Lormetazepam data could not be used in the meta-analysis at all (Liira et al., 2014).

The wake-promoting agents look stronger on paper and are the easiest numbers to misuse. Armodafinil before the night shift produced a mean difference of -0.99 points on the Karolinska Sleepiness Scale, on which a negative change means less sleepiness (95% CI -1.32 to -0.67; two trials, 572 participants, moderate quality evidence). Modafinil produced -0.90 points (95% CI -1.45 to -0.35; one trial, 183 participants, moderate quality evidence), and caffeine combined with pre-shift naps produced -0.63 points (95% CI -1.09 to -0.17; one trial). Those two wake-promoting comparisons on their own enroll more people than the 718 the review reports as its total across all 15 trials, before the nine melatonin trials, the two hypnotic trials and the caffeine trial are counted, so the review's own total and its per-comparison counts do not reconcile. Both are quoted on this page exactly as the review published them, because nothing in the source closes the gap.

Those modafinil and armodafinil trials enrolled only patients with diagnosed shift work sleep disorder, and the review identified no trial in shift workers without that diagnosis. The numbers therefore describe a clinical population rather than a workforce. Reported adverse effects, including headache, nausea, blood pressure rise, and post-marketing severe skin reactions, attach to those agents and not to melatonin. The authors explicitly called for systematic reviews of adverse effects instead of treating the current safety picture as resolved.

The pooled trials were small and drawn mostly from hospital staff such as nurses and physicians, plus police and oil rig workers, across the USA, UK, Norway, Korea and Iran. Participants were aged roughly 24 to 42, and the female share ranged from about 5% to 100% depending on the occupation studied. That is a narrow base to generalize from, and it is part of why the certainty rating stayed low (Liira et al., 2014).

Currency matters here too. The literature search closed on 20 September 2013 and the review was published on 12 August 2014. It remains the current version of this Cochrane review, with no later update, so it is the reference point on the question. It is also more than a decade past its search date, which anyone citing it as present-day practice should say out loud.

Finally, set the 24 minutes against the size of the problem. A night worker attempting daytime sleep is typically short by hours rather than by minutes, and this review measured no outcome on errors, injuries, absence, or long-term health. A modest, low-certainty gain in sleep length is worth knowing about. It is not a reason to shorten recovery time between shifts, and it does not carry the weight that schedule design carries.