Control Group Outcomes in Trials of Psilocybin, SSRIs, or Esketamine for Depression: A Meta-Analysis

Antidepressant efficacy is typically demonstrated by comparing a candidate treatment with a control intervention over 4 to 12 weeks, and responses to control treatments in depression trials are known to vary substantially. Psilocybin has shown large acute-phase antidepressant effects in prior work, often reportedly exceeding effect sizes seen with conventional antidepressants, but some psilocybin trials have also exhibited unexpectedly low responses in their control arms. Functional unblinding (where participants or staff infer treatment allocation because of obvious drug effects) and expectancy differences have been proposed as possible contributors to such low control responses. Hieronymus and colleagues therefore set out to examine whether control-arm outcomes differ systematically between trials of psilocybin, selective serotonin reuptake inhibitors (SSRIs), and esketamine for depression. The investigators performed an indirect comparison using conventional meta-analytic techniques (rather than network meta-analysis) focussing on change in Montgomery-Åsberg Depression Rating Scale (MADRS) scores up to 6 weeks after randomisation, with the aim of assessing whether control treatment efficacy varied by the active drug under study and to what extent that might influence between-group effect size estimates for the active interventions.

This study is a meta-analysis of double-blind trials in adults with major depressive disorder (MDD) or treatment-resistant depression (TRD) that included a relevant control arm and used MADRS for symptom measurement. Trials were identified from two prior meta-analyses and one US Food and Drug Administration review published between March 2019 and December 2024. Included control conditions comprised inert placebo, low-dose psilocybin, or niacin. Exclusion criteria were samples restricted exclusively to individuals younger than 18 years or older than 65 years, crossover designs, and study durations shorter than 2 weeks. Three authors (F. H., E. L., and H. W. S.) independently extracted data, resolving disagreements by consensus. Data synthesis used random-effects meta-analysis. The primary within-group metric was the standardized mean change (SMC) in MADRS from baseline to up to 6 weeks, calculated with raw score standardisation and normalised to each arm's end‑point standard deviation; between-group effects were expressed as standardised mean differences (SMDs) normalised to the pooled end‑point SD. The pre–post correlation was conservatively set to 0 for all trials; one psilocybin arm with missing baseline MADRS had its pretest score imputed as the mean of the other psilocybin arms. Meta-regressions tested study population (psilocybin, esketamine, or SSRI) as the sole moderator, yielding pooled effect sizes, an Omnibus Test of Moderators (QM), and amount of variance explained (R2). The investigators also calculated pooled MADRS response rates (≥50% reduction) and dropout rates descriptively. Analyses were performed in R (version 4.3.3) using the metafor package, and two-sided P <.05 was considered statistically significant.

Seventeen trials met inclusion criteria: four psilocybin trials (total n = 373), two esketamine trials (n = 573), and eleven SSRI trials (n = 4014). Pretreatment to posttreatment SMCs for active arms were 1.21 (SEM 0.15) for psilocybin, 1.43 (0.15) for esketamine, and 1.28 (0.06) for SSRIs. Corresponding control-arm SMCs were 0.50 (0.15) for psilocybin trials, 1.12 (0.17) for esketamine trials, and 1.00 (0.08) for SSRI trials. The pre–post SMC differences (active minus control) were therefore 0.71 for psilocybin, 0.29 for esketamine, and 0.28 for SSRIs, which aligned with between-group SMDs of 0.70 (0.12) for psilocybin, 0.30 (0.12) for esketamine, and 0.27 (0.05) for SSRIs. Meta-regression showed that study population significantly moderated between-group SMDs (QM = 10.7; df = 2; P = .005) and pre‑to‑post control-treatment SMCs (QM = 10.4; df = 2; P = .005) but not pre‑to‑post active-treatment SMCs (QM = 1.21; df = 2; P = .55). Models including study population explained 40.9% of variance in control-arm outcomes, 34.8% of variance in between‑group outcomes, and 0% of variance in active-arm pre–post outcomes. MADRS response rates in control arms were reported as being 14 percentage points lower in psilocybin trials compared with SSRI trials and 23 percentage points lower compared with esketamine trials (table cited in the paper). In a post hoc analysis of esketamine trials conducted in participants with depression and acute suicidality (studies ASPIRE 1, ASPIRE 2, and SUI2001), the mean MADRS decrease for control arms was 22.9 points, corresponding to an SMC of 1.87 (0.12). The authors also reported that dropout/noncompletion rates did not provide a simple explanation for the differing control responses: esketamine trials were typically shorter than psilocybin trials and had comparable dropout rates yet showed larger control responses, while SSRI trials had higher noncompletion rates (likely reflecting older trial eras) but nonetheless larger control responses than psilocybin trials.

Hieronymus and colleagues interpret the findings to indicate that control-arm improvement in psilocybin trials was substantially smaller than in SSRI and esketamine trials and more similar to outcomes seen in waiting-list or care-as-usual controls from psychotherapy trials than to pill-placebo responses commonly reported in antidepressant trials. Active-arm pre–post improvements did not differ significantly across the three study populations, suggesting that the larger between-group differences for psilocybin were driven chiefly by poorer control-arm response rather than by inflated active-arm change. Two broad explanations are proposed for the divergent control outcomes: trials of psilocybin may have enrolled participants less likely to respond to control treatments, or methodological features of psilocybin trials (for example, difficulties maintaining blinding such that control treatment functions more like a waiting-list control) may suppress control-arm improvement. The indirect nature of the comparisons and the limited number of psilocybin and esketamine trials precluded determination of which specific factors account for the differences. The authors note that unblinding rates were not assessed or reported in the included trials, so the extent to which unblinding contributed is unknown. Temporal and other potential confounders also cluster with study population (for example, most SSRI trials were conducted in the 1980s–1990s while all psilocybin trials were published after 2020), making it difficult to disentangle study-era effects from treatment-type effects. The discussion acknowledges that excluding esketamine trials in acute suicidality was deliberate because those populations differ from typical psilocybin and SSRI trial populations; the post hoc inclusion of those studies yielded an even larger control-arm SMC for esketamine (1.87). The authors argue that dropout differences are unlikely to account for the main findings and note that sensitivity analyses (for example, exclusion of a psilocybin study that included an SSRI comparator) did not change the overall result. Overall, the low control-arm response observed in psilocybin trials leads the investigators to caution that psilocybin’s effectiveness for depression may be overestimated if trial methods systematically reduce control-arm improvement.

Participants randomised to control treatments in psilocybin trials showed significantly less improvement in depression ratings than those in control arms of SSRI or esketamine trials. The authors conclude that the low control-arm response in psilocybin trials raises concern that psilocybin may not be as broadly effective for depression as current between‑group estimates suggest. They recommend that future studies investigate moderators of control-arm outcomes in psilocybin research, for example by including multiple control conditions or by enrolling participants with positive expectations for the control treatment, to better understand and mitigate possible bias.