The experimental effects of psilocybin on symptoms of anxiety and depression: A meta-analysis

Research into new treatments for anxiety and depressive disorders has expanded to include psychedelic compounds after a long hiatus. Earlier work suggested potential therapeutic benefits of psilocybin, a 5-HT2A receptor agonist, for conditions including anxiety and substance use, but quantitative synthesis of clinical trials testing psilocybin for anxiety and depression had not been conducted. Goldberg and colleagues set out to fill that gap by performing a meta-analysis of clinical trials in which psilocybin was administered to people with clinically elevated symptoms of anxiety and/or depression. The aim was to estimate pooled effects on anxiety and depression, assess heterogeneity and publication bias, and characterise risks of bias within and across studies to inform future research planning.

This was a meta-analysis of clinical trials that administered psilocybin to samples with clinically elevated anxiety and/or depression. Eligibility required administration of psilocybin, a clinical diagnosis or interview indicating elevated symptoms, and provision of outcome data allowing computation of effect sizes. The literature search used terms combining psilocybin with depression, anxiety, mood, or psychological distress and included ClinicalTrials.gov; the search date was 6 October 2018 and no restrictions were placed on language, date, or publication status. Data extraction was performed using standardised spreadsheets by doctoral-level coders experienced in meta-analysis. Extracted items included measures needed to compute effect sizes, study design features, adverse events, assessment timing, dose, behavioural interventions, control condition, demographics, retention, intent-to-treat use, and location. Risk of bias within studies was assessed following Cochrane domains: selection bias, performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias, and reporting bias. Effect sizes were computed as Hedges' g to correct for small-sample bias. The investigators calculated within-group pre–post effect sizes for psilocybin conditions (including single-group and controlled studies), between-groups effect sizes for controlled trials reflecting the difference in pre–post change between psilocybin and control groups, and pre-to-follow-up within-group effects using the last available follow-up (noting that crossover trials meant all participants received psilocybin by follow-up). Random-effects meta-analytic models were employed in R (metafor), with studies weighted by the inverse variance of effect sizes. Heterogeneity was indexed by I2. Publication-bias assessments included fail-safe N (FSN), funnel plots, and trim-and-fill analyses, which the authors treated as exploratory given the small number of studies. Study selection yielded 864 citations and a final set of five reports representing four unique trials.

Four unique trials met inclusion criteria: one single-group study and three randomized, placebo-controlled crossover trials. Across trials, psilocybin was delivered alongside preparatory and post-session behavioural support and in-session support during administration; dosing and placebo conditions varied but specific dose details are reported in the supplemental material. Samples were small (mean sample size 29.25, SD = 19.14) and demographically limited (mean 58.25% female; mean 86.33% White). Three trials were conducted in the United States and one in the United Kingdom. Post-treatment assessments occurred on average at 3.75 weeks (SD = 2.75), with follow-up assessments at six months. Retention averaged 94.0% at post-treatment and 79.5% at follow-up. Transient adverse reactions were reported but resolved within hours; no serious or persistent adverse events were reported. Risk-of-bias assessments found high risk for performance and detection bias across studies because effective blinding was difficult to achieve given psilocybin's acute effects; attrition bias was high in three of four studies due to lack of intent-to-treat analyses, while selective reporting risk was judged low. Three of the four trials included participants with terminal cancer diagnoses, which the authors note limits generalisability. Meta-analytic findings indicated large within-group pre–post reductions in both anxiety and depression. Across four studies (n = 69 for this analysis), within-group pre–post effect sizes were large for anxiety (Hedges' g = 1.38, 95% CI [0.78, 1.99]) and for depression (g = 1.47, 95% CI [0.72, 2.21]); heterogeneity for these within-group effects was high (I2 approximately 74.5% to 82.4%). In the three randomized, placebo-controlled trials (n = 97 for the between-group pre–post comparison), the placebo-controlled pre–post effect sizes favoured psilocybin with large effects (g ≈ 0.82 to 0.83); heterogeneity for these controlled effects was low (I2 = 0.00%) though confidence intervals were wide. For pre-to-follow-up within-group effects across four studies (n = 117 for this analysis), reductions in anxiety and depression remained large at six-month follow-up, with depression reported as g = 1.17 (95% CI [0.80, 1.53]); heterogeneity for follow-up effects was high (I2 ≈ 57.2% to 82.3%). Assessments of publication bias produced mixed indications. Fail-safe N values for within-group pre–post and pre-to-follow-up effects were large (FSNs = 117 to 141), implying many null unpublished studies would be required to nullify those results. However, FSNs for the placebo-controlled pre–post effects were small (FSNs = 12 to 14), below the commonly recommended threshold, suggesting those controlled effects were not robust to potential unpublished null studies. Funnel plots and trim-and-fill analyses did not indicate clear asymmetry in this sample, but the authors treated these tests as exploratory given the small number of studies.

Goldberg and colleagues conclude that available data tentatively support psilocybin administered with preparatory, in-session, and post-session behavioural support as producing large reductions in anxiety and depression symptoms in clinical samples. The authors emphasise that effect sizes were large both at post-treatment and at six-month follow-up, and that the three randomized, placebo-controlled trials produced between-group effects of roughly g = 0.82–0.83 favouring psilocybin versus placebo. At the same time, the investigators underscore important limitations. Only four trials (N = 117 across analyses) were available, samples were small, and study designs and effect sizes were heterogeneous. Risk of bias was high in several domains, notably performance and detection bias because effective blinding was generally not feasible, and attrition bias due to limited use of intent-to-treat analyses. The small FSNs for placebo-controlled effects indicate those particular findings may not be robust to publication bias. Generalisability is also limited because participants were self-selected and willing to receive a Schedule I substance, racial/ethnic diversity was low, and three trials enrolled people with terminal cancer diagnoses rather than broadly representative clinical populations. The authors recommend further research, including larger placebo-controlled trials and moderator analyses to examine study features such as psilocybin dose and types of behavioural support. They also advise caution in interpreting results as showing benefits of psilocybin in the absence of behavioural interventions, since all included trials combined drug administration with psychosocial support. In summary, the paper presents preliminary evidence that psilocybin plus behavioural support may reduce symptoms of anxiety and depression, but stresses that additional rigorous trials are needed to confirm efficacy, clarify mechanisms, and establish generalisability.