The Study Said Psychosis Rates Were Equal to Placebo. Most Outlets Left That Part Out.

On 16 March 2026, The Lancet Psychiatry published what the researchers describe as the largest-ever systematic review and meta-analysis of randomised controlled trials evaluating medicinal cannabinoids for mental health and substance use disorders. The analysis, led by Dr Jack Wilson, a postdoctoral research fellow at the University of Sydney's Matilda Centre for Research in Mental Health and Substance Use, reviewed 54 RCTs involving 2,477 participants, drawing on evidence spanning 1980 to 2025.

The headline finding, that the researchers found insufficient evidence to justify cannabinoids as a routine treatment for most psychiatric conditions, is significant and was widely covered. What the coverage failed to report, with near-uniform consistency, was a specific safety finding from the same dataset, one that materially complicates the prevailing media narrative around cannabis and psychosis.

What the Researchers Found

Writing in their own plain-language summary in The Conversation, the study authors stated that cannabis medicines were no more effective than a placebo at treating symptoms of psychotic disorders, anxiety, PTSD, anorexia nervosa, or opioid use disorder. Notably, the researchers found no randomised controlled trials at all examining cannabinoids for depression, despite depression being among the most common indications for which these medicines are prescribed in Australia.

On safety, the picture is more nuanced than most coverage conveyed. Participants receiving cannabinoids experienced significantly more adverse events overall compared with those receiving placebo, with the researchers calculating that one additional participant would experience an adverse event for every seven treated with cannabinoids. However, and this is the finding that received least attention: serious adverse events, including psychotic episodes, occurred at no higher rate in the cannabinoid group than in the placebo group. That distinction between all-cause adverse events and serious adverse events was not well communicated in most of the media coverage that followed.

The Pharmaceutical Journal noted that the analysis found no significant effects on psychotic disorders among the conditions examined. But noting the absence of a therapeutic effect on a condition is not the same as reporting the equivalence of serious adverse event rates between cannabis and placebo. The former describes what the medicine failed to treat. The latter describes its safety profile under controlled conditions relative to an inert control. Both findings are in the paper. Only one was consistently reported.

The Substitution

The psychosis claim that did appear in most coverage was drawn from a different body of literature entirely.

The NPR piece, syndicated across multiple public radio affiliates, referenced a well-documented link between cannabis use at a young age and an increased risk of psychosis. That claim is sourced to a separate review published in JAMA Internal Medicine earlier the same month, drawing on a broader mix of study designs including observational data. It is not drawn from the Lancet Psychiatry RCT dataset. The two are presented in close proximity in a manner that invites conflation. A reader who did not parse the passage carefully would reasonably conclude that the cannabis-psychosis link was supported by the controlled trial evidence under discussion. It was not. The controlled trial evidence under discussion found no significant difference in serious adverse events, including psychotic episodes, between cannabis and placebo.

The Pharmaceutical Journal covered the study with reasonable breadth, reporting positive findings alongside the negative ones and including a counter-perspective from Mike Barnes of the Medical Cannabis Clinicians Society, who argued that the RCT model is poorly suited to evaluating a pharmacologically complex plant and that real-world evidence would yield different conclusions. That is a legitimate methodological critique and its inclusion represents a more complete piece of coverage than most. The Pharmaceutical Journal article nevertheless did not report the equivalence of serious adverse event rates between cannabis and placebo, the same omission present across the broader coverage.

The clearest illustration of the substitution comes from within the study's own media cycle. In the University of Sydney press release accompanying the paper, Dr Wilson is quoted as saying: "Though our paper didn't specifically look at this, the routine use of medicinal cannabis could be doing more harm than good by worsening mental health outcomes, for example a greater risk of psychotic symptoms and developing cannabis use disorder, and delaying the use of more effective treatments."

The comment is clearly labelled as outside the scope of the paper, and Dr Wilson's general caution is understandable in a broader clinical context. But it handed media outlets a quotable psychosis concern that could be used without engaging with what the study's own controlled trial data showed. The result was that speculative comment explicitly placed outside the study's scope received prominent placement across the coverage, while the finding inside the study's scope, that serious adverse event rates including psychotic episodes were equivalent between cannabis and placebo, did not appear at all.

This is the structural problem. The study's methodological credibility as gold-standard RCT evidence is invoked to support the "cannabinoids are ineffective" finding, while a lower-grade evidentiary basis, including comment that explicitly acknowledges falling outside the study's scope, sustains the "cannabis causes psychosis" narrative.

What the Study Also Found

A balanced reading requires acknowledging where the researchers found positive signals, and where the study's own limitations are genuine.

The researchers found meaningful reductions in cannabis use and withdrawal symptoms among people with cannabis use disorder, with the effect driven specifically by pharmaceutical-grade mixed CBD and THC formulations administered orally. This is consistent with a harm-reduction model: patients substituting a standardised oral formulation for smoked high-THC cannabis may use less overall, with associated reductions in related health risk. Notably, the researchers also found that cannabinoids increased cocaine cravings among people with cocaine use disorder, a cautionary finding that qualifies any broad reading of the substance use disorder evidence.

The researchers identified a significant reduction in tic severity among those with Tourette syndrome, again only for combined CBD and THC formulations, not for either cannabinoid administered alone. Sleep time showed improvement among those with insomnia, though the researchers note this finding lost significance when high-risk-of-bias studies were removed from the analysis, placing it on less certain ground than some coverage implied. A reduction in autistic traits was observed, but subgroup analysis revealed that neither CBD alone nor combined CBD and THC showed individual significance, and both contributing studies had a high risk of bias.

The researchers characterise these as promising signals warranting further investigation, particularly given the scarcity of effective pharmacological options for cannabis use disorder and Tourette syndrome, but the certainty of evidence across these positive findings was rated as low to very low under the GRADE framework.

Two methodological caveats from the paper itself are directly relevant to Australian readers. First, most included trials used registered cannabinoid products, such as Sativex, rather than the unregistered high-THC products that dominate the Australian market. The safety and efficacy profile observed in the trials may not reflect what Australians are actually receiving. Second, the average treatment duration across the included trials was approximately five weeks, which the researchers acknowledge as a significant constraint: long-term harms are unlikely to be captured within that window.

The Australian Regulatory Context

The study lands at a consequential moment for Australian regulation. According to the paper, over one million prescription applications have been approved in Australia, and mental health conditions account for six of the ten most common prescribing indications. Around 700,000 Australians used cannabis for health purposes in the past year, with sales increasing four-fold since 2022. The majority of products prescribed are not registered with the TGA, meaning they have not undergone rigorous pre-market testing. The researchers explicitly state that their findings should inform the TGA's current review of medicinal cannabis prescribing.

Editorial note: The Conversation article co-authored by Wilson and Stockings states that sales have increased "four-fold" since 2022. The University of Sydney press release issued on the same day describes "a tripling of sales...over the past four years." The Lancet paper does not provide a sales multiplier figure. Both figures originate from the same research team and the discrepancy has not been resolved. This article uses the figure from the Conversation piece, as it carries the authors' names directly, but readers should be aware of the inconsistency pending clarification.

Two co-authors on the paper hold declared interests relevant to Australian regulatory context. Wayne Hall and Myfanwy Graham received consultation fees from the World Health Organisation; Hall additionally received payment for expert testimony on the risks of cannabis use. Graham is a member of the Medicinal Cannabis Expert Working Group at the Australian Department of Health, Ageing and Disability, and has received TGA funding for independent evidence reviews on medicinal cannabis. These disclosures, noted in the paper itself, do not impugn the study's findings. They are part of the complete public record. Most media coverage did not mention them.

The Evidentiary Pattern

The most important observation is the selective application of evidentiary standards.

When RCT evidence supports a cautionary or restrictive interpretation of cannabis medicine, it is presented as definitive. When the same RCT evidence contradicts a harm narrative established by decades of observational epidemiology, it is absent from the coverage and replaced, without acknowledgement, by observational evidence or speculative comment that explicitly falls outside the study's scope.

The relationship between cannabis use and psychosis in population-level data is documented across multiple cohort studies and should not be dismissed. However, population-level associations capture confounding variables that controlled trials are specifically designed to remove. The controlled trial data, on the specific question of whether pharmaceutical cannabinoids administered under clinical conditions produce serious adverse events including psychotic episodes at a higher rate than placebo, found no statistically significant difference. That is a finding. It belongs in the same coverage that cites the study's methodological credibility.

The source material is publicly available under a Creative Commons Attribution licence. Readers are in a position to assess the coverage for themselves.

Citation: 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. '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, 2026. DOI: 10.1016/S2215-0366(26)00015-5. Published under Creative Commons Attribution (CC BY 4.0) licence.

Editorial transparency note: This article was prepared with the assistance of Claude (Anthropic), an AI system used to locate and retrieve media coverage of the study, cross-reference claims against the primary Lancet Psychiatry paper and the authors' own Conversation summary, identify factual discrepancies between drafts and source material, and flag unsupported claims for removal across multiple revision cycles. All editorial judgements, interpretive framing, and final decisions regarding content and accuracy remain the responsibility of the human author. The primary source material, including the full text of the Lancet paper, was provided directly and verified against all substantive claims prior to publication.

This article discusses published research for informational and media literacy purposes. It does not constitute medical advice and should not be used as the basis for any clinical decision. Patients seeking information about medicinal cannabis treatments should consult a qualified healthcare professional.