Assessing the risk of symptom worsening in psilocybin-assisted therapy for depression: a systematic review and individual participant data meta-analysis

The investigators conducted an IPD meta-analysis reported in line with PRISMA guidance and preregistered on the Open Science Framework; the IRB at UW–Madison determined the project to be exempt. They searched PubMed, PsycINFO, Embase and the Cochrane Library from inception through 28 March 2022 using the search term "psilo*", with no restrictions on language or publication status. Bibliographies of recent meta-analyses were also screened. Eligible studies used psilocybin as the primary intervention and reported outcomes on standardised depression measures; both controlled and uncontrolled studies and clinical and non‑clinical populations were eligible. For outcome measurement the team calculated standardised mean difference (SMD) scores for depression measures by computing pre–post change (post minus pre) divided by the baseline standard deviation of each measure. The authors defined clinically relevant symptom worsening as SMD ≥ 0.24. Analyses used a one‑step random effects approach implemented as random intercept multilevel models, nesting participants within study ID in line with recommended practice. Multilevel linear regression was used for continuous SMD outcomes and multilevel logistic regression for the dichotomous worsening outcome. Five baseline demographic predictors that were available across the included studies were examined: age; gender (male versus female); race/ethnicity (White versus non‑White); education (undergraduate degree or higher versus other); and employment status (unemployed versus other). Models of treatment response included the psilocybin arms from all three trials. Analyses were conducted in R; when an empty cell occurred for a demographic predictor (no non‑White participants reported worsening after psilocybin) the investigators used bias‑reduced penalised likelihood logistic regression (logistf) as a sensitivity approach. Deviations from the preregistration and further methodological detail were reported in Supplemental Materials.

Three eligible studies were included; each had two dosing sessions and focused on populations with depressive disorders. Across these studies 102 participants completed both baseline and six‑week follow‑up depression assessments: 62 received psilocybin‑assisted therapy, 29 received escitalopram, and 11 were assigned to waitlist. On average, participants in the psilocybin and escitalopram arms showed large reductions in depressive symptoms at post‑test, with SMDs of −2.38 (SD = 1.69) for psilocybin and −1.56 (SD = 1.36) for escitalopram. Participants in the waitlist condition showed a mean worsening of symptoms (SMD = 0.26, SD = 1.06). The proportion of participants meeting the study’s pre‑specified threshold for clinically significant worsening was 9.7% in the psilocybin arm and 10.3% in the escitalopram arm, whereas 63.6% of waitlist participants met the worsening threshold. When analyses were restricted to the two studies that included control conditions, assignment to the psilocybin arm was associated with a lower likelihood of symptom worsening relative to waitlist; the reported odds ratio comparing worsening in waitlist versus psilocybin was 13.30 (95% CI [3.02, 70.74], p = .001), which the authors interpret as greater odds of worsening in the waitlist group. There was no significant difference in likelihood of symptom worsening between psilocybin and escitalopram (OR = 0.88, 95% CI [0.17, 3.89], p = .865). Including the five demographic covariates did not alter these test results (psilocybin vs waitlist remained p < .001; psilocybin vs escitalopram p = .833 in the adjusted model as reported). None of the five baseline demographic variables were significantly associated with treatment response or with likelihood of symptom worsening in the psilocybin arm. The authors also report that treatment‑resistant depression status was not associated with frequency of symptom worsening (5.3% for treatment‑resistant depression vs 18.1% for other studies; OR = 0.42, 95% CI [0.02, 3.30], p = .446). The authors note the small sample sizes—particularly the waitlist group (n = 11)—and the occurrence of an empty cell for race in the worsening analysis, which prompted use of penalised logistic regression as a robustness check.

Simonsson and colleagues interpret their findings as indicating that clinically relevant worsening of depressive symptoms occurred in a minority of participants receiving psilocybin‑assisted therapy (approximately 10%), a rate comparable to that reported for psychotherapy (~7%) and similar to the rate observed with escitalopram in this pooled dataset. By contrast, a majority of participants in the waitlist condition met the worsening threshold (63.6%), a proportion the authors note is higher than a prior psychotherapy waitlist meta‑analysis (17.4%) but may reflect differences in the chosen worsening cutoff (their conservative SMD ≥ 0.24 versus a larger threshold used elsewhere) and potential nocebo effects when participants do not receive a desired treatment. The pooled analyses that included control trials suggested that psilocybin confers a lower risk of symptom worsening than being assigned to waitlist and a similar risk to an FDA‑approved antidepressant (escitalopram). No baseline demographic variables among the five examined (age, gender, race/ethnicity, education, employment) were associated with either treatment response or likelihood of worsening, and treatment‑resistant status was not linked to higher rates of worsening in these data. The authors acknowledge several important limitations: the overall sample size was relatively small, limiting power to detect smaller effects and making some subgroup comparisons unreliable; the waitlist sample was especially small; the study operationalised worsening only as change in depressive symptom scores and did not capture other clinical deteriorations such as increased suicidality; included trials were heterogeneous in design; only five baseline demographics were available, precluding analysis of psychological or genetic predictors; participant samples lacked diversity in race and ethnicity; and follow‑up was limited to six weeks, preventing assessment of longer‑term trajectories. They conclude that the results are preliminary but suggest psilocybin‑assisted therapy is not associated with higher short‑term risk of clinically relevant symptom worsening compared with standard pharmacological treatment, and that replication in larger, more diverse, and longer‑duration studies would strengthen these safety inferences.