RCTs vs real world evidence: what the Lancet cannabis meta-analysis misses

By Prof Michael Lynskey

A recent meta-analysis of randomised controlled trials, published in Lancet Psychiatry, concluded there was little evidence that cannabinoids help treat mental health conditions (Wilson et al, 2026). It got plenty of media coverage, much of which made the classic mistake of treating 'lack of evidence of benefit' as 'evidence of lack of benefit'. They're not the same thing.

Why does that distinction matter? Because the absence of good trials is not the same as trials showing a treatment doesn't work. When headlines declare that cannabis 'doesn't help' mental health conditions, that message reaches patients, prescribers and policymakers alike and it shapes decisions about treatment, funding and research.

We've already published a blog post picking apart the study's methodological weaknesses. We've now followed it up with an editorial looking at the limitations of the Randomised Controlled Trials (RCTs) included in the meta-analysis, and arguing that 'real world' approaches, like our own Project T21, may tell us more about whether cannabinoids can help with mental health conditions.

Read our full publication here

What the real-world data shows

In the editorial we used anxiety as an example, comparing real world data from 176 people using cannabis-based medicinal products (CBMPs) for generalised anxiety disorder (GAD) with the meta-analysis results.

For anyone unfamiliar with our research, T21 (formerly Project Twenty21) was the UK's largest non-profit-led medical cannabis patient registry. It ran from 2019 to the end of 2024 and gathered longitudinal data on more than 4,500 patients prescribed CBMPs for conditions including chronic pain, anxiety, PTSD and epilepsy. Patients accessed medicines at discounted prices in exchange for contributing data, building real world evidence on how these treatments perform outside the tightly controlled world of clinical trials.

Wilson et al found no significant difference between cannabinoids and placebo for anxiety symptoms (SMD = -1.88 [95% CI -4.79, 1.03]). In Project T21, CBMP use was associated with substantial reductions in anxiety (Cohen's d = 0.90 (0.73-1.08)), along with improvements in quality of life, general health, mood and sleep, with effect sizes from 0.40 to 0.87, all clinically significant.

Why the results differ

So why the gap? A few methodological differences stand out.

Sample size

Wilson et al found only four RCTs testing cannabinoids for anxiety with enough information to include. Between them, those studies covered 292 people, just 150 of whom actually received a cannabinoid rather than placebo. In Project T21, 273 people (5.9% of the sample) reported generalised anxiety disorder as the main condition they were treating with medicinal cannabis, 68.9% male, 31.1% female, 1.1% non-binary, average age 36.3 (range 21-75). Three-month follow-up data were available for 176 of them. In other words, every RCT to date combined has studied fewer people receiving cannabinoids than one real world study. Small samples matter because they produce enormous uncertainty. Look at that confidence interval again: it stretches from a very large benefit to a modest harm. A result that wide tells you almost nothing either way.

Illness severity

In the UK, medicinal cannabis patients must have an established diagnosis and a history of treatments that haven't worked. Three of the four RCTs recruited people who weren't seeking treatment at all. That matters. People who aren't seeking treatment tend to have milder symptoms and less room to improve, making any benefit harder to detect. UK patients, by contrast, are treatment-resistant almost by definition, a very different population.

Comorbidity

RCTs tend to apply strict exclusion criteria, while people seeking CBMP treatment often have multiple conditions. Applying typical trial exclusion criteria to the T21 patients with GAD would have ruled out between 30.2% and 82.4% of them. The trials studied unusually 'clean' samples that bear little resemblance to the people actually being prescribed these medicines, which limits how far their findings can be generalised.

The intervention itself

Three of the four RCTs looked at a single dose of CBD. T21 patients used an average of 2.63 different CBMPs, 88.1% of which were THC-dominant flower. A one-off dose of CBD and an ongoing course of THC-dominant treatment are simply not the same intervention, so it shouldn't surprise anyone that they produce different results. There's also a well-known problem with blinding in cannabis trials: THC has noticeable psychoactive effects, so participants can often tell whether they've received the active drug, which undermines the placebo comparison.

Where the two agree: safety

The one place the two lined up was safety. In T21, 4.5% of patients reported adverse effects, and 77.8% of those were rated mild. Wilson et al similarly found no significant impact of cannabis on adverse events (OR = 1.48 [0.37, 5.86]).

Beyond anxiety

We focused on anxiety for convenience, but the same pattern held across the mental health conditions in the meta-analysis: few studies, small samples, and little resemblance to how medicinal cannabis is actually used. Many trial participants wouldn't be eligible for prescribed cannabis in the UK because they had no formal diagnosis, while many real-world patients would be excluded from the trials because of comorbid conditions. And single-dose studies simply don't reflect how cannabis is used medicinally.

Put simply, the small number and poor quality of existing RCTs mean they can't tell us much about whether CBMPs help with anxiety or other mental health conditions.

Why real world evidence matters

Real world data can help close the gap between the lack of high-quality RCT evidence and the widespread, and growing, use of cannabinoids for mental health. Studies like ours have found medicinal cannabis use is associated with moderate to large reductions in symptoms. Findings from these and other novel study designs, including Bayesian approaches, deserve more weight when evaluating both the effectiveness and safety of cannabinoids.

Regulators are increasingly open to this. Real world evidence is playing a growing role in how medicines are evaluated internationally, and the data collected through T21 will continue to support the case for NHS funding of medical cannabis. Meanwhile, thousands of UK patients are already paying privately for these treatments. They deserve an evidence base that reflects how the medicines are actually used, not one built on a handful of small, short trials that were never designed to answer the question.

Reference

Wilson J, Dobson O, Langcake A, et al. (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.