Five Small Studies and 219 People: What the Cannabinoid Sleep Evidence Does and Does Not Show

The source for the original headline was a systematic review and meta-analysis by Bhagavan and colleagues in CNS Drugs (2020;34(12):1217-1228, PMID 33244728). It identified five studies of cannabinoids in insomnia disorder, two randomized and three non-randomized, with 219 participants in total. Only three could be combined, and those three were the non-randomized ones.

The authors’ own conclusion reads: despite some possible signals for efficacy, the heterogeneity of participants, interventions, efficacy outcomes and results, and the high risk of bias across included trials, do not reliably inform evidence-based practice. A title asserting that cannabinoids improve sleep quality and duration is not supportable from that paper, and it has been replaced.

Insomnia is one of the most common reasons people try cannabis without telling a clinician. A headline promising improved sleep quality and duration, published under a physician’s byline, carries weight that the underlying evidence does not.

The more useful message is narrower and more honest. Some cannabinoid preparations have produced short-term improvements in how people rate their own sleep. The objective measurements are weaker, the trials are small and brief, and the one recent trial designed with an objective primary endpoint did not meet it.

The previous version of this page was titled Cannabinoids Improve Sleep Quality and Duration and stated that recent clinical trial data published in Sleep Medicine demonstrated that cannabinoids improve both sleep quality and duration. Two things were wrong with that. The source was not a clinical trial in Sleep Medicine; it was a systematic review and meta-analysis in CNS Drugs. And that review did not demonstrate the claim, it declined to draw it.

The page also referenced survey data released earlier in the year without identifying the survey. That reference has been removed rather than reworded, because an unidentifiable source cannot be checked.

Bhagavan and colleagues searched four databases through December 2019 for studies of cannabis-based products in adults with insomnia disorder, where the disorder was either formally diagnosed or quantified with a validated instrument. Five studies met criteria, totalling 219 participants: two randomized controlled trials and three non-randomized studies. Risk of bias was assessed with the RoB 2 tool for the randomized trials and ROBINS-I for the rest.

The pooled Pittsburgh Sleep Quality Index result, the number most often quoted as evidence that cannabinoids improve sleep, came entirely from the three non-randomized studies. It showed a mean difference of 1.89 points in favor of cannabinoids at four weeks or less (95 percent confidence interval 1.10 to 2.68, n=176) and 2.41 points at eight weeks (95 percent confidence interval 1.46 to 3.36, n=166). Uncontrolled before-and-after data on a self-rated sleep questionnaire is the weakest form of evidence in the review, not the strongest.

The two randomized trials were small and measured different things. A double-blind crossover trial of 32 participants compared nabilone, a synthetic THC analogue, with amitriptyline and found better Insomnia Severity Index scores after two weeks, adjusted difference 3.25 points (95 percent confidence interval 1.24 to 5.26), but no effect on overall sleep quality on the Leeds Sleep Evaluation Questionnaire. A single ascending-dose trial of nine participants found THC shortened sleep onset latency against placebo at 10, 20 and 30 mg, by 43, 62 and 54 minutes respectively, with confidence intervals that came close to zero at two of the three doses.

Every included study was rated poor quality, principally for small sample size, short treatment period, uncertain clinical significance, and high risk of bias.

The most encouraging randomized result in this field came after the meta-analysis closed. Walsh and colleagues published a double-blind placebo-controlled crossover trial in Sleep (2021;44(11):zsab149, PMID 34115851) testing two weeks of nightly sublingual cannabinoid extract in adults with chronic insomnia symptoms. Twenty-three of twenty-four randomized participants completed the protocol.

Against placebo, the extract reduced Insomnia Severity Index scores by 5.07 points (95 percent confidence interval 2.86 to 7.28) and increased self-reported total sleep time by 64.6 minutes. Actigraphy, which is objective, showed a smaller effect in the same direction: total sleep time up 33.4 minutes, wake after sleep onset down 10.2 minutes, sleep efficiency up 2.9 percentage points. There were no serious adverse events, and 36 of the 40 reported mild adverse events occurred during the active arm.

This is a genuinely positive trial and it deserves to be read as one. It is also twenty-three people, for two weeks, with a self-reported effect roughly twice the size of the objectively measured one. That gap between what people report and what devices record recurs throughout this literature and is the single most important thing to understand about it.

Two recent trials tested individual cannabinoids with predefined primary endpoints, which is the design that settles questions.

Wang and colleagues gave 125 people with insomnia a formulation of 300 mg cannabidiol plus eight terpenes, at least four nights a week for four weeks, in a placebo-controlled crossover design with no THC present (Journal of Clinical Sleep Medicine 2025;21(1):69-80, PMID 39167421). The prespecified primary outcome, percentage of the night spent in slow wave plus rapid eye movement sleep on a wrist-worn tracker, increased by a mean of 1.3 percentage points (95 percent confidence interval 0.1 to 2.5). Total sleep time did not change at all. The authors describe the increase as marginal, and that is the right word.

Lavender and colleagues ran a randomized double-blind placebo-controlled three-arm single-night crossover of cannabinol in twenty adults with physician-diagnosed insomnia disorder, using overnight polysomnography (Journal of Sleep Research 2026;35(4):e70284, PMID 41698831). The primary endpoint was wake after sleep onset. Cannabinol did not change it: 6.3 minutes at 300 mg (95 percent confidence interval 18.2 minutes less to 5.5 minutes more, p=0.29) and 4.0 minutes at 30 mg (p=0.50). Secondary outcomes at 300 mg were more favorable, with more stage 2 sleep, better subjective sleep quality, shorter sleep onset latency and fewer electroencephalographic arousals, and there were 247 mild-to-moderate adverse events across the three arms. A trial that misses its primary endpoint has not shown efficacy, whatever the secondary outcomes do.

Cannabinol is sold widely as a sleep aid on the strength of reputation rather than trial data. This is the trial that tested it directly, and the headline result was negative.

Two larger syntheses published since give a fuller picture and complicate the simple version in both directions.

A systematic review and meta-analysis in Sleep Medicine Reviews (2025;84:102156, PMID 40929927) pooled six randomized trials with 1077 patients and found cannabinoids improved self-reported sleep quality against placebo, standardized mean difference 0.53 (95 percent confidence interval 0.03 to 1.02). The confidence interval nearly touches zero and heterogeneity was 88 percent, which is very high. The subgroup analysis is the part worth keeping: preparations containing cannabinoids other than cannabidiol showed an effect, standardized mean difference 0.82, while cannabidiol-only preparations showed none, standardized mean difference 0.13 with a confidence interval straddling zero.

A much larger review in Lancet Psychiatry (2026;13(4):304-315, PMID 41856154) covering 54 randomized trials and 2477 participants across mental health and substance use indications found that cannabinoids increased sleep time in people with insomnia, both by device and by sleep diary, with standardized mean differences near 0.55. It also rated the certainty of evidence as low for most outcomes, found 44 percent of trials at high risk of bias, and reported higher odds of any adverse event on cannabinoids, odds ratio 1.75, corresponding to one additional person experiencing an adverse event for every seven treated. Serious adverse events were not increased. The authors’ summary judgement was that routine use is currently rarely justified.

Taken together: there is a signal, it is small, it rests disproportionately on subjective measures, it appears stronger for THC-containing preparations than for cannabidiol alone, and it comes with a measurable side effect cost.

Nothing here says cannabinoids never help anyone sleep. It says the trials are too small, too short and too heterogeneous to support a general claim, and that the specific products most heavily marketed for sleep, cannabidiol alone and cannabinol, have the weakest trial support of the group.

The step that gets skipped is diagnosis. Insomnia that persists is often a symptom rather than a condition, and sleep apnea, restless legs, thyroid disease, depression, pain and medication effects each change what treatment is appropriate. A sleep aid taken over an undiagnosed apnea is a missed diagnosis, not a treatment.

Cognitive behavioral therapy for insomnia remains the first-line treatment for chronic insomnia disorder, with a far larger and better-controlled evidence base than anything reviewed here. That is not a reason to refuse to discuss cannabinoids with a patient who is going to use them regardless. It is a reason to sequence the conversation properly.

Weak, and weak in a specific way worth naming. The problem is not that the trials came out negative. It is that the positive results sit mostly on self-reported questionnaires in small, short, heterogeneous studies, while the two recent trials with objective prespecified primary endpoints returned a marginal result and a null result respectively.

The pooled effect sizes in the larger syntheses are real but modest, with heterogeneity near 90 percent in one and low certainty of evidence in the other. In practical terms, the evidence supports saying that some people report better sleep on some cannabinoid preparations. It does not support a general statement that cannabinoids improve sleep quality and duration.

Subjective sleep measures are especially vulnerable in cannabinoid trials, because THC-containing products produce noticeable effects that make blinding imperfect. A participant who can tell they received active treatment rates their sleep differently. This is the most likely explanation for the recurring gap between self-reported and device-measured effects.

Sample sizes are the other persistent problem. Twenty-three completers, twenty participants, nine participants in one of the pooled randomized trials. Effects estimated from samples that size are unstable, and the confidence intervals in several of these results run close to zero at one end.

Publication and reporting incentives also favor the secondary outcomes of a trial that missed its primary endpoint, which is exactly the situation with the recent cannabinol study.

None of this evidence establishes long-term efficacy. The longest treatment period in any of the trials discussed here is four weeks, and insomnia disorder is by definition chronic. Nothing here speaks to tolerance, dependence, rebound insomnia on discontinuation, or effects after months of nightly use.

Nor does it identify which patients respond, which cannabinoid ratio to use, what dose, or what time before bed. It does not compare cannabinoids against cognitive behavioral therapy for insomnia, and it does not address sleep in the context of untreated sleep apnea, where a sedating agent carries its own risk.

Cannabinoids for sleep is one of the widest gaps in cannabis medicine between what is marketed and what is measured. Cannabinol in particular is sold on a reputation as the sedating cannabinoid, a reputation that predates any controlled trial in humans with diagnosed insomnia. When such a trial was run, it missed its primary endpoint.

The pattern across the literature is consistent: THC-containing preparations show more signal than cannabidiol alone, short-term self-reported outcomes show more signal than objective ones, and uncontrolled studies show more signal than randomized ones. That ordering is what one would expect if part of the effect is real and part is expectation, which is the most defensible reading of the current evidence.

Patients tell me cannabis helps them sleep, and I believe them about their own experience. What I cannot do is turn that into a claim about what cannabinoids do to sleep, because the trials that would support such a claim have not been run at the size or length required.

The part that concerns me clinically is not the modest effect size. It is sequence. Someone who has been sleeping badly for a year and reaches for a cannabinoid product has skipped the question of why they are sleeping badly. I have found apnea, thyroid disease and untreated depression in people who came in asking only about a sleep product.

And when a post on this site carried a title stronger than its own source, the fix is to change the title and say why, not to keep the headline and add caveats underneath it.

The meta-analysis this page was built on pooled five poor quality studies and 219 participants and concluded that the evidence does not reliably inform practice. Later trials have not changed that much: the strongest positive result comes from 23 people over two weeks with self-reported effects roughly double the objectively measured ones, a 125-person cannabidiol trial produced a marginal change with no increase in total sleep time, and a polysomnography trial of cannabinol missed its primary endpoint. Evaluate the cause of the insomnia first.

Keep three distinctions. Self-reported sleep is not measured sleep. A secondary outcome is not a primary endpoint. And a preparation containing THC is not interchangeable with cannabidiol or cannabinol, because the pooled data separate them. Anyone quoting this literature without those three distinctions is quoting it wrong, which is how the previous title on this page came to exist.