Published on 17 September 2026
Cannabidiol, THC, and the combination: what do the 54 available randomised trials in mental health and addiction actually show?
The Lancet Psychiatry · 2026 ; 13(4) : 304–315 · Wilson et al.
DOI 10.1016/S2215-0366(26)00015-5
PMID 41856154
Scientific 74
Editorial 90
The essentials
This systematic review pools 54 randomised trials totalling 2,477 participants, testing a cannabinoid as the primary treatment for a mental disorder or a substance use disorder. The median trial has 31.5 participants (interquartile range 16.5 to 57.8), and 24 of the 54 trials, 44%, are at high risk of bias. Four clinical situations concentrate six favourable estimates: in cannabis use disorder, withdrawal symptoms and quantity used; in Tourette syndrome, tic severity; in insomnia, device-measured and diary-measured sleep time; in autism spectrum disorder, autistic traits. GRADE certainty is tiered, not uniform: very low for cannabis withdrawal, for tics, for autistic traits and for cocaine craving; low for cannabis use and for diary-reported sleep; moderate for device-measured sleep time alone. No benefit could be demonstrated for anxiety, anorexia nervosa, psychotic disorders, post-traumatic stress disorder or opioid use disorder, which is not a demonstration of an absence of effect. Data were too scarce to pool for attention-deficit/hyperactivity disorder, bipolar disorder, obsessive-compulsive disorder and tobacco use disorder. For depression, no randomised trial meeting the inclusion criteria was identified across the period from January 1980 to May 2025. A signal in the opposite direction is reported: cocaine craving increases in people with cocaine use disorder. On safety, all-cause adverse events are more frequent, with an odds ratio of 1.75 (95% CI 1.25 to 2.46) and a number needed to harm of 7 (95% CI 5.4 to 10.8) for one additional adverse event, with no detected excess of serious events or of premature discontinuations. The authors conclude that routine use of cannabinoids for these indications is currently rarely justified.
Context
The question no longer arises in a seminar room, it arises in consultation. One patient walks in with an oil bought over the counter, another with a newspaper article on medical cannabis, a third has already stopped their treatment in favour of a product they describe as natural. The clinician then has to answer quickly, and answer correctly.
A paper published in the same journal in 2019 by Black and colleagues already concluded that evidence was insufficient to recommend cannabinoids for mental disorders. The 2026 authors do not present their review as an update of that work: they cite it, as reference nine, among prior syntheses, and only two authors are shared between the two publications. Seven years on, the number of trials has grown, the scope this time also covers substance use disorders, and the conclusion has shifted little. That small shift is itself informative.
A scope clarification matters here, because it conditions the whole reading. This review covers cannabinoids used as the primary treatment for a mental disorder or a substance use disorder. The scope extends beyond first-line psychiatry: it includes Tourette syndrome, insomnia and autism spectrum disorder, and that is precisely where the positive results lie. It does not cover the effects of recreationally used cannabis, nor the link between cannabis use and the onset of mental disorders. That second subject is the object of a separate 2026 publication in the same journal, by Hall and colleagues, which finds credible evidence for a causal contribution of daily cannabis use to psychosis. Treatment and risk factor are two distinct questions, and conflating them is the most common error in public discussion of this topic.
Four products, four statuses
What the word cannabinoid actually covers
The word denotes a family, not a medicine. Within that family, the molecules share neither the same pharmacology, nor the same adverse-effect profile, nor the same regulatory status. The review itself distinguishes between interventions, and its results do not transfer from one product to another.
| Product | Composition | What the review says about it |
|---|---|---|
| Cannabidiol alone | The non-intoxicating molecule in cannabis | The most represented molecule: 24 of 54 trials. No favourable result is attributed to it in isolation, the cannabidiol-alone subgroups reaching significance neither for cannabis withdrawal nor for tics. It is, however, the only molecule tested in all three cocaine use disorder trials, where craving increases, and the subgroup is associated with an excess of adverse events (odds ratio 1.66, 95% CI 1.01 to 2.72, across 12 trials) |
| Delta-9-tetrahydrocannabinol alone | The main psychoactive compound | 18 trials. No favourable result is attributed to it in isolation. On adverse events, this is the only subgroup that does not reach significance (1.41, 95% CI 0.96 to 2.07, across 10 trials) |
| Cannabidiol and delta-9-tetrahydrocannabinol combined | Both molecules together; nabiximols is its pharmaceutical form | 12 trials. The full text attributes to it the two favourable results that are actually resolved by molecule: the reduction in cannabis withdrawal symptoms, as nabiximols, and the reduction in tic severity. After the published correction, the two autism trials also fall under this category. It is also the subgroup with the most marked excess of adverse events (2.60, 95% CI 1.11 to 6.08, across 8 trials) |
| Cannabinoid, unspecified type | An analytic grouping. Inclusion criteria cover both plant-derived and pharmaceutical cannabinoids, the latter category including synthetic cannabinoids | The formulation used for sleep time in insomnia. On this endpoint, the review does not distinguish between molecules: the differences observed between cannabidiol alone and the combination each rest on a single trial per subgroup and are described by the authors themselves as highly uncertain |
It is worth adding what the review does not test. Medical cannabis in the sense of the French regulatory scheme, that is oils or dried flower dispensed under controlled conditions, is not an intervention category in this meta-analysis. And cannabidiol-containing products sold outside the medicines pathway are not medicines: their content, purity and stability are not subject to the same requirements. Transposing a trial result to a bottle bought over the counter has no basis.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| People treated for a mental disorder or a substance use disorder. 54 trials, 2,477 participants: 1,713 men (69%), 764 women (31%), median age 33.3 years (interquartile range 28.1 to 38.05). Ethnicity data were not available | |
| Intervention | |
| Plant-derived or pharmaceutical cannabinoids, administered as the primary treatment for the condition studied. The published abstract names only one precise intervention, the combination of cannabidiol and delta-9-tetrahydrocannabinol, and otherwise uses the categories “any type of cannabinoid” and “cannabinoids”. The full text details the three analytic subgroups: cannabidiol alone (24 trials), delta-9-tetrahydrocannabinol alone (18 trials), the combination of the two (12 trials). Treatment lasts an average of 4.98 weeks (standard deviation 4.41) and the cannabinoid is given as first-line treatment in 46 of 54 trials | |
| Comparator | |
| Depending on the trial: placebo, active medication, waitlist, or another intervention. Placebo is named as the comparator for only part of the reported comparisons; the others are expressed relative to a control group | |
| Endpoints | |
| Primary: remission of the disorder or reduction of its symptoms. Safety: all-cause adverse events and serious adverse events, with calculation of the number needed to harm for one additional adverse event | |
| Design | |
| Systematic review with meta-analysis of randomised trials. Search of Ovid MEDLINE, PsycINFO, the Cochrane Central Register of Controlled Clinical Trials, the Cochrane Database of Systematic Reviews and Embase, for peer-reviewed articles published between 1 January 1980 and 13 May 2025. Selection and extraction by two independent reviewers. Random-effects model in Review Manager 5.4, odds ratios for binary variables, standardised mean differences for continuous variables. Risk of bias with the RoB 2 tool, certainty with the GRADE framework. Protocol registered on PROSPERO, CRD42023392718 · CEBM 1a |
Quality control
| Criterion | Status |
|---|---|
| Protocol registration | Solid |
| Finding PROSPERO CRD42023392718, cited in the publication and backed by a published protocol | |
| Selection and extraction | Solid |
| Finding Two independent reviewers for screening and for data extraction | |
| Assessment tools | Solid |
| Finding RoB 2 for risk of bias, GRADE for the certainty of the primary endpoints | |
| Risk of bias in the included trials | Severe |
| Finding 24 of 54 trials, 44%, at high risk of bias, 20 (37%) raising some concerns, 10 (18%) at low risk. A separate assessment of conflicts of interest in the included trials rates 11 of 54, 20%, as high risk, mainly because of authors’ industry ties and an ill-defined sponsor role. This is the main limitation of this work, and it concerns the raw material, not the synthesis | |
| Certainty of the evidence | Caveat |
| Finding The full text’s GRADE grading is tiered, not uniform. Very low: cannabis withdrawal symptoms, tic severity, autistic traits, cocaine craving. Low: cannabis use, diary-reported sleep time. Moderate: device-measured sleep time, the only endpoint in the whole review to reach that level. The abstract’s wording, certainty “low for most outcomes”, therefore covers a more contrasted reality, in both directions | |
| Trial size | Caveat |
| Finding 2,477 participants across 54 trials. The median trial has 31.5 participants (interquartile range 16.5 to 57.8), with wide disparities by indication: a median of 105 participants per trial in autism, 78 in cocaine use disorder, 32 in cannabis use disorder, 22 in tics, 17.5 in anorexia nervosa. At this scale, an absence of significant result is as consistent with an absence of effect as with a lack of power | |
| Representativeness | Caveat |
| Finding 69% men, median age 33.3 years, ethnicity data not available. Transposition to women, adolescents and older adults is not supported by this population | |
| Independence | Caveat |
| Finding Declared funding: the Australian National Health and Medical Research Council, with no role in design, analysis or writing. The full text’s acknowledgements detail individual support: an NHMRC investigator grant for the last-named author (GNT2017346) and for three other authors, an NHMRC doctoral scholarship and a Monash scholarship for another, a UK Research and Innovation grant for the UK-based author (MR/Y017560/1). The conflict-of-interest statement is not silent: two authors declare consulting fees from the World Health Organization; one further declares payment for expert testimony on cannabis risks, another declares membership of the Australian health department’s medicinal cannabis working group and funding from the Therapeutic Goods Administration for independent reviews on medicinal cannabis. The remaining authors declare no conflicts of interest. No link to a cannabis industry sponsor appears in this statement | |
| Published correction | Caveat |
| Finding A correction appeared in the same journal, 2026; 13(8): e11, applied to the online version on 16 June 2026. It corrects the x-axis spacing of figure 2 and, in table 1, the classification by cannabinoid type: for anxiety disorders, 1 trial of delta-9-tetrahydrocannabinol alone and 5 of cannabidiol alone; for autism spectrum disorder, 0 trials of cannabidiol alone and 2 of the combination, both with follow-up beyond one month. Useful consequence for clinical reading: the pooled autism trials belong to the combination, not to cannabidiol alone. No pooled estimate is changed. One caveat remains: the “combination” cell in the autism column still reads 1 in the online version consulted on 10 August 2026, where the correction states 2 | |
Results
| Indication | Published result |
|---|---|
| Cannabis use disorder, withdrawal symptoms | Pooled result, 12 trials: standardised mean difference -0.70 (95% CI -1.32 to -0.09), very low GRADE certainty. In subgroup analysis, only the pharmaceutical combination of cannabidiol and delta-9-tetrahydrocannabinol reaches significance: -0.29 (95% CI -0.57 to -0.02) versus placebo. Neither cannabidiol alone nor delta-9-tetrahydrocannabinol alone does |
| PEB readingDoes not survive sensitivity analysis removing the high-risk-of-bias trials, the pooled result drops to -0.84 (95% CI -1.75 to 0.06) and stops being significant. The figure quoted everywhere, -0.29, is a subgroup estimate whose upper bound brushes zero | |
| Cannabis use disorder, quantity used | Pooled result on consumption, most often self-reported: -0.74 (95% CI -1.18 to -0.30), low GRADE certainty. The often-cited -1.00 (95% CI -1.69 to -0.30) corresponds to the subgroup of trials measuring weekly grams; the frequency-of-use subgroup is not significant (-0.48, 95% CI -1.07 to 0.11) |
| PEB readingMeasurement subgroup, not a molecule the abstract attributes this figure to the combination of the two molecules, the full text makes it a subgroup defined by the unit of measurement. And pooling is done in standardised mean differences: PEB does not convert -1.00 into grams per week | |
| Tics and Tourette syndrome | Pooled result, 5 trials, 168 participants: tic severity -0.62 (95% CI -0.92 to -0.32), very low GRADE certainty. It is the full text, not the abstract, whose wording does not establish it, that attributes the significant result to the combination: -0.68 (95% CI -1.03 to -0.34), against -0.24 (95% CI -1.55 to 1.07) for cannabidiol alone and -0.47 (95% CI -1.17 to 0.22) for delta-9-tetrahydrocannabinol alone. No effect on premonitory sensations (-0.20, 95% CI -0.70 to 0.31) |
| PEB readingThe largest effect size in the review an interval entirely on the favourable side, but very low certainty, only five trials, and an adverse-event odds ratio of 4.93 (95% CI 1.80 to 13.48). An indication outside first-line psychiatry | |
| Insomnia, sleep time | 4 trials, 137 participants. No effect on insomnia symptoms (-0.44, 95% CI -1.28 to 0.41). Any type of cannabinoid: +0.54 (95% CI 0.14 to 0.95) on device-recorded sleep time, moderate GRADE certainty; +0.55 (95% CI 0.01 to 1.09) on sleep diary, low certainty. No effect on sleep quality or on sleep-onset latency |
| PEB readingDoes not survive sensitivity analysis excluding the high-risk-of-bias trials, the device measure drops to 0.44 (95% CI -0.10 to 0.98) and loses significance. The self-reported measure has a lower bound of 0.01. The endpoint is sleep time, not insomnia severity or daytime functioning, and the adverse-event odds ratio reaches 11.04 (95% CI 4.37 to 27.92) | |
| Autism spectrum disorder | Autistic traits, 2 trials, 210 participants: -0.36 (95% CI -0.66 to -0.07), very low GRADE certainty. Both trials are at high risk of bias. Taken in isolation, no subgroup by cannabinoid type reaches significance |
| PEB readingThe most fragile result in the review two trials, both at high risk of bias, very low certainty, and a surrogate endpoint: the autistic-traits score, measured by the Social Responsiveness Scale and the Autism Treatment Evaluation Checklist, is neither a measure of autonomy, nor of schooling, nor of quality of life. After the correction, both trials belong to the combination, not to cannabidiol alone. This is the indication where the risk of public misinterpretation is highest | |
| Cocaine use disorder | Craving, 3 trials, 199 participants: +0.69 (95% CI 0.22 to 1.15) versus placebo, in the direction of an increase, very low GRADE certainty. Excess adverse events: odds ratio 3.76 (95% CI 1.70 to 8.33) |
| PEB readingA harm signal the interval excludes zero, the direction is unfavourable. The abstract says nothing about the molecule, which does not exonerate cannabidiol in the least: in the full text’s table, all three trials for this indication are classified as cannabidiol alone. That is precisely the molecule patients buy freely | |
| Anxiety, anorexia nervosa, psychotic disorders, post-traumatic stress disorder, opioid use disorder | No significant effect on the associated endpoints. Anxiety, 6 trials: -1.88 (95% CI -4.79 to 1.03). Psychotic disorders, 8 trials: total PANSS score -0.14 (95% CI -0.39 to 0.11). Post-traumatic stress disorder, 3 trials: -0.16 (95% CI -0.82 to 0.49). Opioid use disorder, 4 trials: withdrawal -0.63 (95% CI -1.41 to 0.14), craving -0.06 (95% CI -0.70 to 0.59). Anorexia nervosa, 2 trials, qualitative synthesis with no difference |
| PEB readingAbsence of proof of efficacy to be distinguished from proof of inefficacy, and the distinction reads in the intervals. Anxiety’s, from -4.79 to 1.03, excludes neither a major benefit nor a harmful effect: it says nothing. That of psychotic disorders, from -0.39 to 0.11, is narrow, and comes close to a true null result | |
| ADHD, bipolar disorder, obsessive-compulsive disorder, tobacco use disorder | Insufficient data to meta-analyse |
| PEB readingAn empty zone trials exist, they do not allow pooling | |
| Depression | No randomised trial meeting the inclusion criteria, namely a cannabinoid given as the primary treatment for the disorder, across a search from 1 January 1980 to 13 May 2025 |
| PEB readingNo data a category still different from the previous two: nothing to pool, nothing to discuss | |
| All-cause adverse events | 28 trials pooled. Odds ratio 1.75 (95% CI 1.25 to 2.46) versus control group, number needed to harm of 7 (95% CI 5.4 to 10.8) for one additional event. By subgroup: combination 2.60 (95% CI 1.11 to 6.08) across 8 trials, cannabidiol alone 1.66 (95% CI 1.01 to 2.72) across 12 trials, delta-9-tetrahydrocannabinol alone 1.41 (95% CI 0.96 to 2.07) across 10 trials, not significant |
| PEB readingA clear and consistent excess the figure means that around 7 people need to be treated to observe one more adverse event than with the comparator, not that one person in 7 will experience an adverse effect. Note that cannabidiol alone, a product with a reputation for being harmless, is among the subgroups with a significant excess | |
| Serious adverse events and study discontinuations | Serious events, 12 trials: odds ratio 1.12 (95% CI 0.51 to 2.47). Discontinuations, 30 trials: 0.91 (95% CI 0.72 to 1.14) |
| PEB readingNo detected excess a reassuring but underpowered result: with trials whose median size is 31.5 participants, a moderate excess risk of a rare event would remain invisible | |
Three reading notes. First, the values above come from the full text, freely available under a CC BY 4.0 licence, not from the abstract alone; they are converted from the source’s decimal notation with no other transformation. Second, several of the figures highlighted in the abstract are subgroup estimates, whereas the GRADE grading applies to the corresponding pooled estimate: this is a persistent source of confusion, and it is why both levels are given here side by side. Third, no adjustment for multiple comparisons is mentioned, even though the review explores around ten indications and several endpoints per indication. Query a heterogeneous dataset that many times, and a few intervals end up barely excluding zero, which is exactly the shape of several of the favourable estimates.
Critical appraisal
| Domain | Judgement |
|---|---|
| Level of evidence of the synthesis | Solid |
| Finding Systematic review with meta-analysis of randomised trials, pre-registered protocol, dual screening, standardised tools. This is the method expected at the top of the hierarchy | |
| Quality of the raw material | Severe |
| Finding 44% of trials at high risk of bias, and very low GRADE certainty on four of the seven graded favourable or unfavourable endpoints. A rigorous synthesis of fragile trials produces a precise estimate of an uncertain reality | |
| Maintenance of blinding | Caveat |
| Finding A point not addressed by the authors, but inherent to the subject: the psychoactive effects of delta-9-tetrahydrocannabinol often let the participant guess their group. On self-reported endpoints such as craving, reported sleep or anxiety, this de facto unblinding pushes the result toward the active arm. An expert appraisal, not to be mistaken for a finding of the review | |
| Heterogeneity of interventions | Caveat |
| Finding The “any type of cannabinoid” category pools molecules with differing pharmacology. Useful for statistical power, costly for clinical interpretation: one does not prescribe a category | |
| External validity | Caveat |
| Finding A predominantly male population, median age 33.3 years, ethnicity data absent. The patients who raise the question in consultation do not necessarily resemble those who were included | |
| Duration of observation | Severe |
| Finding Mean treatment duration is 4.98 weeks (standard deviation 4.41), and the longest follow-up exceeds one month in only a minority of trials: four of four insomnia trials stop at one month or less, two of eight psychotic-disorder trials and two of six anxiety-disorder trials are limited to a single day. Yet the principal stakes of long-term cannabinoid treatment, tolerance, dependence, cognitive impact, are judged beyond a few weeks. These trials cannot see them | |
| Match between claim and evidence | Solid |
| Finding The authors’ conclusion stays calibrated: “rarely justified”, not “ineffective”, not “to be avoided”. They themselves flag the need for higher-quality research and document the unfavourable signal on cocaine | |
What remains is to sort out what these data allow us to say. What is demonstrated fits into two points, and they must be stated precisely. First, across the indications tested, no benefit of cannabinoids could be shown for anxiety, anorexia nervosa, psychotic disorders, post-traumatic stress disorder or opioid use disorder. This is a failure to demonstrate, not a demonstration of an absence of effect: the nuance is not rhetorical, it changes what can be told to a patient. Second, there is a clear excess of all-cause adverse events, found for both the combination and for cannabidiol given alone. What is suggested, without being established: a small-to-moderate benefit in four clinical situations, of which two, cannabis withdrawal and tic severity, are attributed by the full text to the combination of cannabidiol and delta-9-tetrahydrocannabinol, while for sleep time and autistic traits the review does not distinguish between molecules; and an increase in craving in people with cocaine use disorder, where all three pooled trials used cannabidiol alone. What amounts to expert opinion, including our own: the idea that unblinding accounts for part of the observed self-reported effects, and the idea that repeated statistical testing accounts for the fragility of several intervals.
What cannot be said either deserves to be spelled out. This review does not show that cannabidiol is dangerous in anxiety disorders: it shows that there is no demonstrated effect, and that adverse effects are more frequent under cannabinoid treatment, across all products pooled. Nor does it show that medical cannabis is useless in the indications where it is regulated in a given jurisdiction: those indications are not psychiatric and are not the object of this work. Finally, the absence of a randomised trial in depression says nothing about efficacy, in either direction. It says that between January 1980 and May 2025, no trial meeting this review’s inclusion criteria, a cannabinoid given as the primary treatment for the disorder, was conducted.
Authorisation, availability, and reimbursement: three different questions
A cannabinoid medicine can be authorised on paper and unobtainable in practice; it can be authorised and marketed, yet reimbursed for one indication and not another; and a reimbursement decision, wherever it is taken, never means that a treatment is ineffective, only that a health system has ruled on its collective value for money for a given indication. These three planes, marketing authorisation, actual commercial availability, and collective coverage, are frequently conflated in public discussion, and keeping them apart is what allows an honest answer to a patient. France offers one health system where all three can be checked against primary regulatory sources rather than described from memory, and it serves here as a worked example, not as the point itself.
The elements below were checked on 10 August 2026 against the websites of the French national medicines agency (Agence nationale de sécurité du médicament et des produits de santé), the European Medicines Agency, the French National Assembly and the French official legal database Légifrance.
| Product or scheme | Verified status | Authorised indication |
|---|---|---|
| Nabiximols (Sativex) | French marketing authorisation granted in 2014. The summary of product characteristics on the French medicines agency’s database specifies that each 100-microlitre spray contains 2.7 mg of delta-9-tetrahydrocannabinol and 2.5 mg of cannabidiol. The product has never been marketed in France. The Transparency Committee concluded on 22 October 2014 that its added clinical benefit was low, with no improvement in actual benefit, a conclusion upheld on re-evaluation on 20 July 2022, and price negotiations between the manufacturer and the Economic Committee for Health Products did not result in an agreement. A ministerial reply published in the Official Journal on 28 February 2023 referred to renewed pricing discussions; no actual marketing could be traced as of 10 August 2026 | Treatment of symptoms related to moderate-to-severe spasticity due to multiple sclerosis, in adults with an inadequate response to other antispastic agents. A neurological, not a psychiatric, indication. Authorised on paper, unavailable in practice |
| Cannabidiol (Epidyolex) | European marketing authorisation of 19 September 2019, held by Jazz Pharmaceuticals Ireland Limited. Orphan medicine. Marketed in France since January 2023, available in pharmacies on initial hospital prescription | Two distinct indications, not to be confused. Adjunctive treatment of seizures associated with Lennox-Gastaut syndrome or Dravet syndrome, in combination with clobazam, in patients aged 2 and older. Adjunctive treatment of seizures associated with tuberous sclerosis complex, in patients aged 2 and older, with no requirement for combination with clobazam. Neurological, not psychiatric, indications |
| French medical cannabis pilot programme | Started 26 March 2021. No further enrolment possible since 27 March 2024. The French medicines agency states the pilot ended on 31 December 2024 | Five clinical situations, as named by the French medicines agency: neuropathic pain refractory to accessible therapies; certain drug-resistant forms of epilepsy; certain refractory symptoms in oncology related to cancer or its treatment; palliative care situations; painful spasticity from multiple sclerosis or other central nervous system disorders. No psychiatric indication |
| Aftermath of the pilot programme | The health ministry extended coverage for already-enrolled patients beyond 31 March 2026, until three months after the publication of the national health authority’s opinion, an opinion still pending as of 10 August 2026. The permanent scheme for “cannabis-based medicines” is not operational: the three legal texts meant to establish it were notified to the European Commission on 20 March 2025 and are still not published in the Official Journal, a search of Légifrance finding no such text more recent than 2024. The health minister had confirmed to Agence France-Presse, in June 2026, an expected publication in July 2026: it did not happen. The medicines agency states that access will be strictly restricted, as a last-line treatment on initial hospital prescription | Not applicable to psychiatry to date |
The conclusion to draw from this table is a simple one. No psychiatric indication appears among those verified here, neither for the two authorised cannabinoid medicines, one of which is not marketed in France, nor among the five clinical situations of the French pilot scheme. A patient who requests a cannabinoid for anxiety, depression or post-traumatic stress disorder falls outside any authorised indication, for every product verified here. This does not mean their request is unlawful: cannabidiol-containing products sold outside the medicines pathway are legally marketed, and an off-label prescription is not legally impossible. It means that no authorised indication supports it, independently of what the trials show. Readers in other jurisdictions should expect a comparably narrow, largely neurological set of authorised psychiatric indications for cannabinoid medicines, and should verify their own system’s specifics rather than assume this French example transfers directly.
Level of evidence
PEB assessment: very low confidence on cannabis withdrawal, on tics, on autistic traits and on cocaine craving; low confidence on cannabis use and on diary-reported sleep; moderate confidence on device-measured sleep time alone, noting that this estimate does not survive the removal of high-risk-of-bias trials. Moderate confidence, however, on the overall message, namely the absence of demonstrated efficacy in the major psychiatric indications and the existence of an excess of non-serious adverse events. The scientific score of 74 reflects a gap between the quality of the synthesis, which is high, and that of the synthesised trials, which is not. A methodologically flawless piece of work does not manufacture evidence where none exists: it takes stock of it, and that stocktaking is itself considerable. The editorial score of 90 reflects how often this question arises in consultation, and the clarity of the answer this review allows.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Fifty-four trials, two thousand four hundred patients, and on depression, zero trials. For anxiety and post-traumatic stress, nothing significant. What moves on cannabis withdrawal and on tics is the combination with THC, not cannabidiol alone. On sleep and on autism, the review does not distinguish between molecules, and for autism the certainty is very low with two trials. On the other side of the ledger, one extra adverse event for every seven patients treated. I can answer my patient without falling out with them. ”
What this means in practice: when a patient asks about a product, the first useful question is which one exactly, and for which symptom. This review’s answers change depending on both.
What you can take from this on Monday morning. No prescription changes, since none was possible for these indications. What changes is the quality of the conversation.
- Get the product named before discussing substance. Oil bought over the counter, resin, flower, capsules, medical cannabis under a controlled scheme, an authorised medicine: these situations share neither composition nor status, and this review does not say the same thing about each. One question suffices: what exactly are you taking, and at what dose.
- Answer with figures rather than with a position. On depression, no randomised trial meets this review’s criteria. On anxiety, anorexia nervosa, psychotic disorders, post-traumatic stress disorder and opioid use disorder, no significant effect. The favourable signals concern four clinical situations, cannabis use disorder, tics, sleep time in insomnia and autistic traits, that is six estimates, four of very low certainty and only one of moderate certainty.
- Document current use rather than comment on it. Quantity, frequency, potency if known, route of administration, duration, sought-after motive, effect obtained. This record is worth more than agreement or refusal, and it helps identify an established use disorder.
- Use the safety figure correctly. The number 7 means that roughly seven people treated produce one more adverse event than under the comparator. It does not mean that one person in seven will have an adverse effect. Against that, no excess of serious events or of premature discontinuations was detected, which is also worth stating, for honesty’s sake.
- Mention the cocaine signal when relevant. In a person with cocaine use disorder, craving increases under cannabinoid treatment in these trials, all three of which used cannabidiol alone. This is the review’s only clearly unfavourable-direction result, and it concerns precisely the molecule with a reputation for being harmless.
- Set the regulatory picture out without moral judgement. The two authorised cannabinoid medicines verified here are approved for spasticity in multiple sclerosis, for a product never marketed in France, and for three rare forms of epilepsy. The French scheme covers five clinical situations, none psychiatric, and its pilot phase ended on 31 December 2024. Stating this calmly avoids turning a clinical question into a conflict.
- Redirect toward what has evidence for the targeted symptom. For chronic insomnia, for an anxiety disorder, for post-traumatic stress disorder, first-line treatments are set out in guidelines outside this review, and those are what should be offered in response to the request. The review analysed here does not evaluate them.
Frequently asked questions
Does cannabidiol relieve anxiety?
Across the six trials pooled in this review, no significant effect is found on anxiety-related endpoints. The confidence interval measures the extent of our ignorance: from -4.79 to 1.03, it excludes neither a considerable benefit nor a harmful effect. An absence of a significant result at this scale therefore does not prove an absence of effect. In practice, there is no demonstration of efficacy, and there is an excess of non-serious adverse events, including in the cannabidiol-alone subgroup.
What about depression?
There is no randomised trial to analyse, across a search covering January 1980 to May 2025, the inclusion criterion being a cannabinoid given as the primary treatment for the disorder. This differs from a negative result: there is nothing to interpret, in either direction.
Do the favourable results concern cannabidiol alone?
None of them do, but the answer varies by indication. For cannabis withdrawal and tic severity, the full text shows that only the subgroup combining cannabidiol and delta-9-tetrahydrocannabinol reaches significance, with neither cannabidiol alone nor delta-9-tetrahydrocannabinol alone doing so. For sleep time and autistic traits, the review does not distinguish between molecules: the category used is “any type of cannabinoid” for sleep, and for autism no subgroup taken in isolation is significant. One point deserves note in the opposite direction: the three cocaine use disorder trials, where craving increases, all used cannabidiol alone.
Is a product bought over the counter equivalent to what was tested?
Nothing supports that claim. The trials use products with controlled composition and a defined dose. A product outside the medicines pathway is not subject to the same requirements for content, purity and stability, and its actual composition is not guaranteed. Transposing a trial result to a bottle bought elsewhere is not justified.
Is medical cannabis authorised in France for a psychiatric indication?
No, according to the official sources checked on 10 August 2026. The five clinical situations of the French scheme are refractory neuropathic pain, certain drug-resistant epilepsies, certain refractory symptoms in oncology, palliative care situations, and painful spasticity from multiple sclerosis or other central nervous system disorders. The pilot programme ended on 31 December 2024 and the permanent scheme is still not operational: the three legal texts meant to establish it, notified to the European Commission on 20 March 2025, were not published in the Official Journal as of 10 August 2026, despite a publication announced for July 2026.
How should the figure of 7 be interpreted?
It is the number of people who need to be treated to observe one additional adverse event compared with the comparator, with a 95% confidence interval running from 5.4 to 10.8. It is not an individual probability, and it is not a raw frequency. An important precision: this excess concerns all-cause adverse events. No excess of serious adverse events (odds ratio 1.12, 95% CI 0.51 to 2.47) or of study discontinuations (0.91, 95% CI 0.72 to 1.14) was detected, with the caveat that such an excess would be hard to detect in trials whose median size is 31.5 participants.
Does this review settle the question of the link between cannabis and psychosis?
No, that is not its object. It evaluates cannabinoids as a treatment. The question of cannabis as a risk factor belongs to a different literature, and a 2026 publication in the same journal finds credible evidence for a causal contribution of daily use to psychosis. The two questions are answered separately.
Annotated bibliography
Wilson J, Dobson O, Langcake A, Mishra P, Bryant Z, Leung J, Dawson D, Graham M, Teesson M, Freeman TP, Hall W, Chan GCK, Stockings E (2026). The efficacy and safety of cannabinoids for the treatment of mental disorders and substance use disorders: a systematic review and meta-analysis. The Lancet Psychiatry, 13(4), 304–315. DOI 10.1016/S2215-0366(26)00015-5 · PMID 41856154. Source study analysed here. Open access under a CC BY 4.0 licence. Full text consulted on 10 August 2026 on the publisher’s site, in its corrected version of 16 June 2026: endpoint-by-endpoint GRADE grading, subgroups by cannabinoid type, sensitivity analyses, tables 1 and 2 of characteristics, acknowledgements and conflict-of-interest statement. The electronic appendix, which contains the list of products tested trial by trial and the detailed forest plots, was not reviewed. Funding: National Health and Medical Research Council. PROSPERO protocol CRD42023392718, search from 1 January 1980 to 13 May 2025.
Correction (2026). Correction to Lancet Psychiatry 2026; 13: 304–15. The Lancet Psychiatry, 13(8), e11. DOI 10.1016/S2215-0366(26)00176-8 · PMID 42302809. Correction text consulted on 10 August 2026. It corrects the x-axis spacing of figure 2 and, in table 1, the classification by cannabinoid type: anxiety disorders, 1 trial of delta-9-tetrahydrocannabinol alone and 5 of cannabidiol alone; autism spectrum disorder, 0 trials of cannabidiol alone and 2 of the combination, both with follow-up beyond one month. These data were also corrected in the appendix. Corrections applied to the online version on 16 June 2026. No pooled estimate is changed.
Black N, Stockings E, Campbell G, Tran LT, Zagic D, Hall WD, Farrell M, Degenhardt L (2019). Cannabinoids for the treatment of mental disorders and symptoms of mental disorders: a systematic review and meta-analysis. The Lancet Psychiatry, 6(12), 995–1010. DOI 10.1016/S2215-0366(19)30401-8 · PMID 31672337. An earlier systematic review, cited as reference nine by the 2026 publication among prior syntheses, and not presented by its own authors as the work that the 2026 review updates. Only two authors are shared between the two publications, Emily Stockings and Wayne Hall. Cited here as bibliographic antecedent, not as a source of figures.
Hall W, Yimer TM, Lees Thorne R, Hoch E, Burke C, Gridley A, Hurd YL, Wilson J, Leung J, Freeman TP (2026). Relationships between cannabis use and mental disorders: assessing the coherence of evidence from studies with different methodologies. The Lancet Psychiatry, 13(7), 604–614. DOI 10.1016/S2215-0366(26)00086-6 · PMID 42309106. A companion publication from the same journal, on the distinct question of cannabis as a risk factor for mental disorders. Cited to delimit the scope of the main analysis. Abstract consulted by index reconstruction, without reading the full text, which calls for caution about its exact wording.
Agence nationale de sécurité du médicament et des produits de santé (French national medicines agency). Medical cannabis, thematic file. ansm.sante.fr. Page updated 20 April 2026, consulted 10 August 2026. Source of the following elements: end of the pilot programme on 31 December 2024, enrolment stopped since 27 March 2024, extension of coverage for enrolled patients beyond 31 March 2026 until three months after the national health authority’s opinion, notification of three legal texts to the European Commission on 20 March 2025, future access restricted to last-line treatment on initial hospital prescription.
Agence nationale de sécurité du médicament et des produits de santé. Medical cannabis: the French medicines agency publishes the patient breakdown for each indication retained in the pilot programme. ansm.sante.fr. Consulted 10 August 2026. Source of the list of the five retained indications, cited verbatim in the regulatory table.
Agence nationale de sécurité du médicament et des produits de santé. Summary of product characteristics, SATIVEX oromucosal spray, solution. agence-prd.ansm.sante.fr. Consulted 10 August 2026. Source of the composition, 2.7 mg of delta-9-tetrahydrocannabinol and 2.5 mg of cannabidiol per 100-microlitre spray, and of the authorised indication in multiple sclerosis spasticity.
French National Assembly. Written question no. 75, marketing of Sativex in France for multiple sclerosis, and the reply of the ministry of health and prevention. questions.assemblee-nationale.fr. Reply published in the Official Journal on 28 February 2023, consulted 10 August 2026. Source of the 2014 French marketing authorisation, of the Transparency Committee’s opinion of 22 October 2014 finding a low added clinical benefit with no improvement in actual benefit, of its confirmation on 20 July 2022, and of the absence of a pricing agreement with the Economic Committee for Health Products.
European Medicines Agency. Epidyolex (cannabidiol), product page and summary of product characteristics. ema.europa.eu. Consulted 10 August 2026. Source of the European marketing authorisation date, 19 September 2019, and of the exact wording of the two authorised indications, taken from section 4.1 of the summary of product characteristics.
Vidal. Epilepsy: EPIDYOLEX (cannabidiol) available in pharmacies. vidal.fr. Published 5 January 2023, consulted 10 August 2026. Source of the availability in pharmacies on initial hospital prescription and of the reimbursement of both indications.
Légifrance (French official legal database). Search of the Official Journals for the term cannabis, sorted by descending signature date. legifrance.gouv.fr. Consulted 10 August 2026. No text establishing the permanent framework for cannabis-based medicines appears as of this date, the most recent relevant text dating to 2024.