Two Major 2026 Reviews Went Looking for Cannabis Trials in Depression. One Found None at All.

The Wilson meta-analysis in The Lancet Psychiatry set out to test cannabinoids as the primary treatment for diagnosed mental disorders. Across 54 randomized trials and 2,477 participants, it reported an outright absence of randomized trial evidence for the treatment of depression.

A second 2026 review, by Bharat and colleagues in Psychiatry Research, cast a wider net and did pool depression data. Its result was a standardised mean difference of -0.20 (95% CI -0.43 to 0.04), which the authors describe as trivial to no effect, drawn largely from depression scores measured as secondary outcomes in trials run for other conditions.

Depression is one of the most frequently cited reasons people give for using cannabis, and one of the most frequently listed qualifying conditions in state programs that allow provider discretion. The evidence supporting that practice is not weak. On the specific question of cannabis as a treatment for a depressive disorder, it is close to nonexistent.

That distinction matters at the bedside. A clinician who says the evidence is mixed is describing a different situation from one who says the trials were never run. The second statement is accurate, and it leads to a more honest conversation about what a patient is and is not signing up for.

Jack Wilson and colleagues at the Matilda Centre, University of Sydney, published their systematic review in The Lancet Psychiatry in the April 2026 issue, with online publication on March 16, 2026. They searched Ovid MEDLINE, PsycINFO, the Cochrane Central Register of Controlled Trials, the Cochrane Database of Systematic Reviews, and Embase for peer-reviewed randomized trials published between January 1, 1980 and May 13, 2025. The inclusion criterion was narrow and deliberate: the cannabinoid had to be tested as the primary treatment for the mental disorder or substance use disorder in question.

Fifty-four trials met that bar, covering 2,477 participants. None of them studied depression. The authors state it plainly in their findings: there was an absence of randomized controlled trial evidence for the treatment of depression.

Chrianna Bharat and colleagues at the National Drug and Alcohol Research Centre, UNSW Sydney, took a broader approach in Psychiatry Research, published August 11, 2026. They included 82 experimental and 118 observational studies and allowed symptom-level outcomes rather than requiring the disorder to be the trial’s primary indication. That wider frame did produce pooled depression data, and the pooled result was a standardised mean difference of -0.20 with a confidence interval crossing zero.

The Bharat depression estimate deserves a careful read rather than a headline. A standardised mean difference of -0.20 sits at the boundary of what is conventionally called a small effect, and the 95% confidence interval runs from -0.43 to 0.04. An interval that includes zero means the data are compatible with no benefit. Heterogeneity was extreme at I-squared 91.6%, meaning the individual studies disagreed with each other far more than chance would predict.

The authors also flag where those depression scores came from. Much of the anxiety and depression evidence, they write, was measured as secondary outcomes in trials whose primary condition was something else entirely, such as chronic pain or multiple sclerosis. A depression scale administered to a pain trial population is not the same instrument, in the same hands, as a depression scale administered to people recruited for a depressive disorder.

Overall certainty across the Bharat review was rated predominantly very low. That is the lowest rung on the GRADE ladder, and it means the true effect could be substantially different from the estimate.

Seven years earlier, Nicola Black and colleagues published a systematic review in the same journal, The Lancet Psychiatry, in December 2019. They searched the literature from January 1980 to April 2018 and included 83 studies, 40 of them randomized trials covering 3,067 participants. Forty-two of those studies addressed depression, including 23 randomized trials with 2,551 participants, yet the trials were overwhelmingly conducted in people with other primary medical conditions.

Black and colleagues concluded that there was scarce evidence to suggest cannabinoids improve depressive disorders and symptoms. The 2026 reviews, working with a literature that grew for another seven years, reached the same place. One found nothing qualifying at all under a stricter definition, the other found a pooled estimate compatible with no effect.

Three review teams, three search strategies, three publication dates spread across seven years, and one consistent answer. These groups overlap in membership, which is worth naming, but the consistency across changing inclusion rules is itself a finding.

The Wilson review was not uniformly negative, and reading it as a blanket verdict on cannabinoids misses its structure. A combination of cannabidiol and delta-9-tetrahydrocannabinol reduced cannabis withdrawal symptoms (SMD -0.29, 95% CI -0.57 to -0.02) and weekly grams of cannabis used (SMD -1.00, 95% CI -1.69 to -0.30) in people with cannabis use disorder. It reduced tic severity in tic or Tourette’s syndrome (SMD -0.68, 95% CI -1.03 to -0.34).

Any cannabinoid type increased sleep time in people with insomnia, measured both by electronic device (SMD 0.54, 95% CI 0.14 to 0.95) and sleep diary (SMD 0.55, 95% CI 0.01 to 1.09). Autistic traits decreased in autism spectrum disorder (SMD -0.36, 95% CI -0.66 to -0.07).

The pattern is worth noting. Where cannabinoids showed measurable effects, the targets were physical or behavioral and reasonably objective: tics, sleep duration, grams consumed. The conditions where the evidence went quiet were the ones defined by internal mood state.

Both 2026 reviews found a consistent tolerability signal. Wilson and colleagues reported higher odds of all-cause adverse events with cannabinoids compared with control (OR 1.75, 95% CI 1.25 to 2.46), with a number needed to treat to harm of 7, and no higher odds of serious adverse events or study withdrawal. Bharat and colleagues found the single significant safety result to be increased withdrawals due to adverse events with THC (OR 2.78, 95% CI 1.66 to 4.65).

Wilson and colleagues also reported that cannabinoids increased cocaine craving in people with cocaine use disorder (SMD 0.69, 95% CI 0.22 to 1.15) compared with placebo. That is a directional harm signal in a specific population, and it belongs in any honest summary of this literature.

A treatment with a measurable adverse event burden and no demonstrated benefit for a given indication is not a neutral option for that indication. It carries cost without documented return.

As a description of what randomized trials exist, this is strong evidence. Two independent groups, using different inclusion rules and different databases, searched the same forty-five year window and produced compatible answers about depression. A well-conducted systematic search is close to the best available instrument for the question of whether a body of trial evidence exists.

As evidence about whether cannabinoids help depression, it is close to silent, and the silence runs in one direction only. Wilson and colleagues found no qualifying trial. Bharat and colleagues found a pooled estimate compatible with no effect, assembled mostly from secondary outcomes and rated at very low certainty. Neither result establishes that cannabis fails in depression. Both establish that the question has not been properly tested.

The two reviews are not independent in the way the word usually implies. Several authors appear on both, and both groups draw on the same Australian research infrastructure, the same databases, and overlapping search strategies. Agreement between them is reassuring but is not the same as replication by an unconnected team.

The Wilson inclusion rule, requiring the cannabinoid to be the primary treatment for the primary diagnosis, is defensible for regulatory purposes and unrepresentative of clinical practice. Almost no patient takes cannabis instead of an antidepressant with nothing else in the plan. Reading the absence of qualifying depression trials as a verdict on adjunctive use would be reading past what the review measured.

Both reviews also carry the standard limits of the underlying literature. The Wilson sample was 69% male with a median age of 33.3 years, which does not match the demographic profile of treated depression in most clinics. Heterogeneity in the Bharat depression pool was 91.6%, which means the individual trials were measuring meaningfully different things.

Neither review shows that cannabis is ineffective for depression. An absence of trials and a confidence interval that crosses zero are statements about the evidence base, not about the drug. The honest description is that the trials have not been done at adequate quality or scale.

Neither review addresses adjunctive use, patient-directed titration, or cannabis used to treat a symptom such as insomnia or pain in a person who also has depression. Those are the situations most patients are actually in, and they sit outside what these designs measured.

Neither review tells a clinician what to do with a patient who is already using cannabis and reports that their mood is better. That question has no trial answer at present.

The contrast between depression and insomnia inside the same review is the most useful part of this literature. Sleep duration is measurable by device. Tic severity is countable. Grams of cannabis consumed per week is a number the participant can report and the investigator can check. Depression is measured by scales that depend heavily on self-report and expectation, in a trial design where the active drug announces itself through its psychoactive effects.

That is not an argument that cannabis works for depression and the instruments are too crude to detect it. It is an argument that the trials required to answer the question would need to be designed with unusual care about blinding, expectation, and outcome selection, and that so far nobody has funded them at the necessary scale.

Meanwhile the 2019 Black review, the 2026 Wilson review, and the 2026 Bharat review form an unusually clean sequence. The literature grew. The answer did not change.

When a patient tells me cannabis helps their depression, I believe them about their experience. What I cannot do is tell them the trials support it, because the trials do not exist. That is a specific and uncomfortable thing to say out loud, and I think saying it is part of the job.

What I have learned over two decades in this work is that the conversation usually gets more productive once the evidence is on the table accurately. Many patients who describe cannabis as helping their depression are describing something more particular when you ask: they are sleeping through the night for the first time in years, or their pain has stopped consuming the whole day, or they are less anxious in the evening. Those are downstream effects with better evidence behind them, and naming them correctly makes the treatment plan easier to build and easier to monitor.

The part I find genuinely frustrating is the funding picture. Forty-five years of literature, and not one randomized trial that tested a cannabinoid as a treatment for a depressive disorder well enough to qualify for a systematic review. That is not a scientific verdict. That is a research agenda nobody paid for.

There is no randomized trial evidence that cannabis treats depression as a primary indication, and the broader pooled estimate is compatible with no effect. Patients using cannabis and reporting mood improvement deserve to be taken seriously and asked what specifically changed, because sleep, pain, and anxiety have better evidence behind them and are easier to track.

Read this as a map of what has been studied, not a verdict on what works. The finding to carry forward is that depression is the largest unfilled hole in cannabinoid psychiatry and that three reviews across seven years agree about it. The finding not to carry forward is any claim that cannabis has been shown to fail in depression, because a trial that was never run cannot fail.