Melatonin: what the sleep trials measured

Melatonin has the most-quoted number in sleep and the least-examined split. Pooled across 1,683 people it moves sleep onset by 7.06 minutes. Analysed by age, the effect survives only in children and adolescents; in adults it is not statistically significant. Both come from meta-analyses in the same literature, and almost no article prints the second.

The pooled number, and what it averages over

The 2013 PLoS One meta-analysis pooled 19 randomised placebo-controlled trials, 1,683 subjects, adults and children with primary sleep disorders. Sleep latency fell by a weighted mean difference of 7.06 minutes (95 % CI 4.37 to 9.75), Z = 5.15, p below 0.001. Total sleep time rose 8.25 minutes (1.74 to 14.75), Z = 2.48, p = 0.013. Sleep quality improved by a standardised mean difference of 0.22 (0.12 to 0.32), Z = 4.52, p below 0.001.

Meta-regression found that longer trials and higher amounts gave greater effects on latency and total sleep time, and that trial duration and amount had no significant effect on sleep quality. The authors call the effects modest, note they do not appear to dissipate with continued use, and state that the absolute benefit is smaller than other pharmacological treatments for insomnia.

That is the honest version of the headline. A 7-minute shift in how long it takes to drop off is the entire pooled signal on the endpoint that drives most of the marketing, and the sleep-quality effect is 0.22 of a standard deviation. Both are small, both significant, and not the same size.

The same literature split by age

The 2022 systematic review in Sleep Medicine Reviews identified 24 randomised controlled trials of chronic insomnia, all against placebo, and ran subgroup analyses by age group because the heterogeneity was too large to pool.

In non-comorbid insomnia, melatonin was significantly effective for sleep onset latency and total sleep time in children and adolescents. In the adults group it was not significantly effective for sleep onset latency, total sleep time or sleep efficiency. In comorbid insomnia it improved latency across all age groups, but that adult subgroup rested on a single study. The authors conclude that melatonin did not appear effective in adults but might be effective in children and adolescents.

This is the same body of trials the first meta-analysis drew on, and the two results are not in conflict. Pooling children with adults produced 7.06 minutes. Stratifying by age shows the aggregate is carried by the younger half. Both are true and only one is usually printed.

How much, and how long before: the amount-response work

The 2024 amount-response meta-analysis in the Journal of Pineal Research took 26 double-blind randomised trials published between 1987 and 2020, 1,689 observations, and modelled how the schedule and the amount change the two endpoints.

Modelled response showed latency falling and total sleep time rising gradually, peaking at 4 mg a day. Meta-regression found insomnia status (beta 0.50, p below 0.001) and the interval between administration and the sleep episode (beta -0.16, p = 0.023) to predict sleep onset latency. Time of day (beta -0.086, p below 0.01) was the only predictor of total sleep time. The authors frame the result as a contrast with what clinical practice usually uses.

Two things are worth noticing. The model puts a ceiling at 4 mg, not a threshold below which nothing happens, and the abstract prints no minimum. And the two endpoints are driven by different variables, which cautions against reading any coefficient as a mechanism.

Where the objective measurements did not move

The most informative trial here used polysomnography rather than questionnaires. Ninety-seven consecutive middle-aged patients with primary insomnia were randomised to 3 mg fast-release melatonin (51) or placebo (46) for four weeks, taken an hour before bedtime.

The results paragraph lists what moved: early wake time decreased by 30.63 minutes (95 % CI -53.92 to -7.34, P = 0.001) and N2 sleep by 7.07 % (-13.47 to -0.68, P = 0.031). Everything else in that paragraph is a null: sleep latency, sleep efficiency, wake during sleep, and the percentages of N1, N3 and REM. The questionnaire endpoints were null too, PSQI 1.53 (95 % CI -0.55 to 3.61, p = 0.504), ISI 0.81 (-2.27 to 3.88, p = 0.165), ESS -0.83 (-3.53 to 1.88, p = 0.147). No serious adverse events were reported.

A discrepancy is worth naming rather than smoothing. The abstract’s conclusion sentence names total sleep time and REM among the improvements while the results paragraph names early wake time and N2 and lists latency as unchanged. This page quotes the results paragraph. A 30-minute shift in early wake time in a middle-aged insomnia population is not the same finding as the 7-minute pooled latency shift.

Two disease-specific trials that disagree with each other

In Parkinson’s disease, two randomised trials of the same compound give opposite pictures, and the difference is about population and endpoint rather than about the molecule.

One randomised 107 screened idiopathic Parkinson’s patients, 86 included, to 3 mg or placebo for eight weeks, 73 completing. Everything moved in favour of melatonin: PSQI 1.87 (95 % CI 1.5 to 2.1, p = 0.001), Epworth 1.25 (0.80 to 1.71, p = 0.001), NMSS 6.11 (5.27 to 6.92, p = 0.001), PDQ-39 8.12 (6.97 to 9.50, p = 0.001), and on polysomnography a latency difference of 8.36 minutes (4.38 to 12.34, p = 0.001) and total sleep time 14.51 minutes (5.00 to 24.41, p = 0.005).

The same abstract then says it is not clear whether such changes would have real-life impact, that the patients were young with short disease duration and high anticholinergic use, and that the authors are doubtful it generalises to the typical Parkinson’s population, who are older. That limit disclosure is printed by the authors and is what keeps this from being read as an efficacy claim.

The other randomised 30 Parkinson’s patients with REM sleep behaviour disorder to 4 mg prolonged-release melatonin or matched placebo, an eight-week intervention bracketed by four weeks of observation on each side, 15 patients per group analysed. No difference at the primary endpoint: 3.4 events per week on melatonin against 3.6 on placebo, difference 0.2 (95 % CI -3.2 to 3.6), P = 0.92.

Two trials, one compound, same disease, opposite direction. The plausible reasons are the endpoint (sleep quantity against dream-enactment events), the preparation, the sample (86 against 30) and the population. None of those reconciliations is printed in either abstract, so this page does not assert one.

Jet lag, where the effect size is genuinely different

The one indication with a large number is jet lag, measured completely differently: a Cochrane review of 10 trials, 9 adequate enough to contribute, in airline passengers, staff and military personnel.

It reports that amounts between 0.5 and 5 mg were similarly effective, except that people fell asleep faster and slept better after 5 mg than after 0.5 mg, and that amounts above 5 mg appeared no more effective. The number needed to treat was 2. The benefit was judged greater the more time zones were crossed and less for westward flights.

The endpoints there are subjective jet-lag ratings, not polysomnographic latency. An NNT of 2 and a 7-minute pooled latency shift are not comparable quantities, and the first belongs to a syndrome rather than to insomnia. The abstract also contains a recommendation sentence; this page reports the trial counts and the NNT and attributes no recommendation to anyone.

Where this sits on the shelf

There is no melatonin product in this catalogue and this page carries no product link, for the reason in the note at the top. The honest neighbour on the same topic is the DSIP page, a different compound with a completely different evidence base.

What the data do not show

They do not show a 7-minute latency shift in adults, because the age-stratified analysis found that endpoint not significant in the adult group, nor an effect on adult sleep efficiency, for the same reason. They do not show that questionnaire measures of insomnia move, because the one polysomnography trial found PSQI, ISI and ESS all null. They do not show a meaningful effect in the typical older Parkinson’s population, because the authors of the positive trial say so themselves. They do not show an effect on REM sleep behaviour events, because that trial was null at P = 0.92. And they do not show anything about any amount other than the two or three trials have actually used, because that is the only evidence base there is.

What we supply

Nothing on this page is a product of ours and this page recommends nothing. It exists so a reader chasing a sleep-latency claim can find the pooled estimate and the age stratification under it.

Research use only. This page summarises published studies for research reference. It is not medical advice, not a protocol, and not a suggestion for human use. Nothing we supply is for human or veterinary use.