Psychiatry · Medical Literature
[Correspondence] Efficacy and safety of cannabinoids
Abstract
Jack Wilson and colleagues1 provide an important synthesis of randomised evidence on cannabinoids for mental and substance use disorders; their findings also highlight the limits of inferring real-world safety from trial-based adverse event data alone. The apparent absence of increased serious adverse events in trials should not be interpreted as reassurance, but rather as a consequence of measuring harm within a framework that does not reflect real-world exposure.
Clinical relevance
Does the absence of increased serious adverse events in randomized controlled trials of cannabinoids for mental and substance use disorders constitute meaningful real-world safety reassurance, or is this a methodological artifact of trial-based harm ascertainment frameworks? __SNAPSHOT__ {"studyDesign":"Correspondence letter (invited commentary) responding to a published systematic review and meta-analysis (Wilson et al., The Lancet Psychiatry, 2026). No original data are presented; this is an expert methodological critique.","setting":"Not applicable (correspondence/commentary; authors affiliated with academic institutions, country not reported in abstract)","sampleSize":"Not applicable — no primary data. The parent study (Wilson et al.) included 54 RCTs with 2,477 participants.","population":"The commentary addresses the broader population receiving cannabinoids for mental disorders and substance use disorders in real-world clinical settings — populations systematically underrepresented in the RCTs analyzed by Wilson et al.","intervention":"Cannabinoids (including CBD, THC, synthetic cannabinoids, and combined formulations) used as primary treatment for mental disorders or substance use disorders","comparator":"Placebo or active comparator, as reported in the parent RCTs; not applicable to the correspondence itself","primaryOutcome":"Not applicable — the correspondence does not define a primary outcome. Its central argument concerns the adequacy of trial-based serious adverse event (SAE) ascertainment as a surrogate for real-world harm.","followUp":"Not reported — correspondence does not specify follow-up. The parent RCTs were published 1980–2025.","fundingSource":"Not reported for this correspondence. The parent Wilson et al. study was funded by the Australian National Health and Medical Research Council (NHMRC)."}
Key findings
Yuan and colleagues do not report original quantitative findings. Their correspondence makes two central methodological arguments in response to Wilson et al. First, the apparent absence of increased serious adverse events (SAEs) across cannabinoid RCTs should not be interpreted as evidence of real-world safety. Second, this null SAE signal is characterized as a structural consequence of measuring harm within a framework — the short-duration, tightly controlled RCT — that does not adequately capture real-world exposure patterns (e.g., chronic use, higher-potency formulations, dose titration, comorbid vulnerability). For context from the parent study: Wilson et al. reported that cannabinoids were associated with significantly increased all-cause adverse events versus controls (OR 1.75, 95% CI 1.25–2.46), but that serious adverse events and study withdrawals did not differ significantly between cannabinoids and controls across conditions. The NNTH from a prior review in this area was reported as 7 for anxiety symptom reduction. No additional statistics are attributable to the Yuan et al. correspondence itself. NNT/NNH for the correspondence: not calculable (no original data).
Practice implications
This correspondence establishes a theoretically important but empirically unquantified concern: that the RCT framework systematically underestimates the real-world safety risks of cannabinoids because trials enroll selected, lower-risk populations; impose standardized products, fixed doses, and short durations; and use harm ascertainment methods that may miss delayed, cumulative, or context-dependent adverse effects. The correspondence does not establish what the true SAE rate is in real-world populations, does not provide new data, and cannot define the magnitude of the safety gap it describes. The findings apply conceptually to any clinician interpreting SAE data from cannabinoid RCTs — particularly psychiatrists, addiction medicine specialists, and prescribers managing patients with complex comorbidities who are routinely excluded from such trials. The correspondence appropriately cautions against the inferential leap from 'no increase in SAEs in trials' to 'cannabinoids are safe in clinical practice.' This limitation is additionally relevant given that the parent Wilson et al. review — while the largest RCT-only synthesis to date — covered only approximately 2,500 participants across 54 heterogeneous trials, limiting statistical power to detect rare but serious harms. Only an abstract and secondary sources were available for this correspondence; the full text may contain additional nuance, citations, or specific proposals not captured here. __LIMITATIONS__ 1. Correspondence format: no primary data, no statistical analysis, no formal systematic methodology. Arguments are necessarily speculative in the absence of real-world pharmacovigilance data to quantify the hypothesized safety gap. 2. The specific magnitude of real-world SAE risk above the trial-observed rate is not established or estimated. 3. The parent RCT database (Wilson et al.) itself has significant limitations: high heterogeneity across cannabinoid type, dose, and indication; predominantly male sample (69%); median participant age of 33 years; evidence quality rated as generally low by GRADE criteria; and virtual absence of trial data for some common prescribing indications (e.g., depression). 4. The correspondence does not distinguish between cannabinoid types (CBD vs. THC vs. synthetic) in its safety argument, despite materially different risk profiles. 5. No alternative harm-ascertainment method or study design is formally proposed or evaluated. 6. Generalizability of the concern is plausible but unproven: real-world registry data would be required to test the correspondence's central hypothesis empirically. __BOTTOM_LINE_CATEGORY__ potentially_practice_informing __BOTTOM_LINE__ Yuan et al. raise a methodologically valid and clinically important caution: the null serious adverse event signal in cannabinoid RCTs should not be treated as real-world safety reassurance, as trial designs structurally constrain the detection of chronic, dose-dependent, or population-specific harms. Clinicians should interpret the parent Wilson et al. findings — which already found no efficacy signal for anxiety, PTSD, depression, psychotic disorders, or opioid use disorder, and increased all-cause adverse events (OR 1.75) — in light of the likelihood that trial-based harm ascertainment underestimates risk in the patients most commonly prescribed cannabinoids. Routine prescribing of cannabinoids for these indications remains unsupported by current evidence and should prompt caution, not permissiveness.