Published on 16 September 2026
Psilocybin in treatment-resistant depression: the EPIsoDE trial misses its primary outcome
In brief
The German EPIsoDE trial randomised 144 adults with treatment-resistant depression, triple-blind and with federal funding. The primary outcome, response on the 17-item Hamilton Depression Rating Scale at week six, was not met: 17.0% on psilocybin 25 mg, 12.5% on psilocybin 5 mg, 10.6% on nicotinamide, with an adjusted odds ratio of 1.73 for the 25 mg versus nicotinamide comparison, 95% CI 0.53 to 6.23, p = 0.19. The key secondary outcome, change in Hamilton score at six weeks, favours psilocybin, with an estimated difference of 4.60 points, 95% CI 2.18 to 7.01. This divergence between primary and secondary outcomes cannot be resolved by simply keeping the second. Two elements shape the reading: among participants given 25 mg, 86% correctly guessed that they had received the high dose, and suicidal ideation was reported as an adverse event more often on dosing days, in 4% of 25 mg administrations against 1 to 2% of comparator administrations. The authors themselves describe the trial as inconclusive. Its value today lies in its safety data and in its method.
The context
For some years psilocybin has held an unusual place in the debate, one where public expectation came before demonstration. The authors point out that earlier data are limited by often small samples, insufficient pharmacovigilance, functional unblinding and expectancy effects. A larger trial does exist in the same indication, however: the one by Goodwin and colleagues, published in 2022 in the New England Journal of Medicine with 233 participants. EPIsoDE is therefore not the largest trial in the field, it is one of the most methodologically complete.
What sets it apart is three choices: blinding at three levels, investigator, participant and rater; a four-arm architecture that allows dosing schedules to be compared; and an ethical guarantee that every participant would receive at least one 25 mg dose during the trial. The formal assessment of blinding integrity, rarely reported in this field, is its most original contribution.
The study at a glance
| Population | |
| 144 adults aged 25 to 65 years with moderate to severe treatment-resistant depression, defined by a score of at least 17 on the 17-item Hamilton scale at screening and by the absence of adequate improvement despite at least two antidepressant treatments from different pharmacological classes, each given for at least six weeks at the minimum effective dose in the current episode. All monoaminergic medication had to be stopped at least two weeks before the first dose, five weeks for fluoxetine. Mean baseline Hamilton score 22.1, mean age 42.6 years, 59% men, 98% White participants, 89% psychedelic-naive. | |
| Intervention | |
| Oral synthetic psilocybin, two doses six weeks apart, within a structured psychotherapeutic programme of seven preparation and integration sessions, fourteen hours in total, with two therapists present, a dosing session of six to eight hours followed by an overnight hospital stay. | |
| Randomisation arms | |
| Four arms, ratios 2:2:1:1: nicotinamide 100 mg then psilocybin 25 mg; psilocybin 5 mg then 25 mg; psilocybin 25 mg then 5 mg; psilocybin 25 mg twice. Up to the primary outcome, the comparison therefore involves three levels, 25 mg versus 5 mg versus nicotinamide, with 48 participants per level. | |
| Comparators | |
| Two distinct comparators. Nicotinamide 100 mg replaced niacin, which the authors state is not approved in the European Union, and they specify that nicotinamide produces fewer acute effects, “functioning largely as an inert placebo”. It is psilocybin 5 mg that plays the part of active comparator, chosen as a non-therapeutic dose producing mild subjective effects. | |
| Primary outcome | |
| Response, defined as a reduction of at least 50% in the 17-item Hamilton score, six weeks after the first dose and one day before the second. Pre-specified fixed-order testing: 25 mg versus nicotinamide first, then 25 mg versus 5 mg, the second test being contingent on the success of the first. | |
| Key secondary outcomes | |
| Change in Hamilton score from baseline at six weeks, change and response on the Beck Depression Inventory at six weeks. The Clinical Global Impression and the Global Assessment of Functioning are exploratory outcomes. | |
| Design | |
| Phase 2b randomised controlled trial, triple-blind, investigator-initiated, sponsored by the Central Institute of Mental Health in Mannheim, conducted at two German university centres, Mannheim and the Charité in Berlin, each of which randomised 72 participants, from June 2021 to February 2024. Registration NCT04670081 and EudraCT 2019-003984-24. |
The quality check
| Criterion | Judgement |
|---|---|
| Primary outcome | Not met |
| Finding17.0% against 10.6%, adjusted odds ratio 1.73, 95% CI 0.53 to 6.23, p = 0.19. The hierarchical testing stops there: the 25 mg versus 5 mg comparison, odds ratio 1.52, interval 0.47 to 4.85, no longer carries confirmatory value. | |
| Blinding integrity | Compromised |
| FindingAmong participants given 25 mg, 86% correctly guessed that they had received the high dose, with a mean certainty of 77.6%. Across all participants, 74% correctly identified that they had received psilocybin. On the therapists’ side, the 25 mg dose was correctly identified in 82% of cases. Rater blinding was not assessed. | |
| Divergence between outcomes | Needs careful interpretation |
| FindingThe primary outcome is negative while the continuous key secondary outcome is favourable. Keeping the second would amount to choosing the result after seeing it. The authors acknowledge this and explicitly write that the divergence “warrants cautious interpretation”. | |
| Safety signal | Present |
| FindingSuicidal ideation reported as an adverse event on dosing days in 4% of 25 mg administrations against 1 to 2% of comparator administrations. Four serious adverse events in total, two of them judged related to psilocybin 25 mg. | |
| Source of funding | Public |
| FindingGrant 01EN2006 A/B from the German Federal Ministry of Education and Research, open access publication funded by the German Center for Mental Health. The funders declare that they had no role in the conduct of the study or in the decision to publish. The protocol also provided for additional funding from private investors and sponsors of the MIND Foundation, and many authors declare personal ties to the sector. | |
| Statistical power | Overestimated |
| FindingThe sample size calculation rested on expected response rates of 50%, 20% and 10% for 25 mg, 5 mg and nicotinamide, that is 43 participants per arm, raised to 48 to allow for 10% dropout, 144 in total. The observed rates for psilocybin fall well short of that assumption. The authors themselves cite this overestimation of the expected effect as a major limitation, and point out that the calculation was not based on a pre-specified minimal clinically important difference, which they call “an omission for a phase 2b trial”. | |
The findings
| Outcome | Value |
|---|---|
| Primary outcome, response at week six | 17.0% against 12.5% and 10.6% |
| Reading8 responders out of 47 on 25 mg, 6 out of 48 on 5 mg, 5 out of 47 on nicotinamide. Adjusted odds ratio 1.73, 95% CI 0.53 to 6.23, p = 0.19. The interval includes no effect, a reduction in the response rate and a sixfold increase in the odds of response. The result is undecided, not a refutation. | |
| Key secondary outcome, change in Hamilton score at six weeks | 4.60 points in favour of psilocybin 25 mg |
| ReadingEstimated mean difference against nicotinamide, 95% CI 2.18 to 7.01, p < 0.001. Against psilocybin 5 mg, 3.09 points, interval 0.69 to 5.50, p = 0.02. These p values are nominal and uncorrected: since the confirmatory procedure stopped at the primary outcome, they do not control the type I error. | |
| Beck Depression Inventory at six weeks | 23.4% response against 10.6% |
| ReadingResults consistent with the Hamilton scale. Odds ratio against nicotinamide 2.64, interval 0.87 to 9.05. Change in score: 7.21 points in favour of 25 mg, interval 2.56 to 11.86. A self-reported outcome, and therefore directly exposed to unblinding. | |
| Response at week one | 34.0% against 10.4% and 6.4% |
| Reading16 responders out of 47 on 25 mg, odds ratio against nicotinamide 7.60, interval 2.29 to 34.75. A secondary outcome pre-specified in the statistical analysis plan. The early gap is large, but the interval is very wide, and whether this is a rapid antidepressant effect or a postacute afterglow phenomenon remains an open question. | |
| Remission at six weeks | 10.6% against 0% and 2.1% |
| ReadingHamilton score below 8. Group sizes of around five patients: no odds ratio can be interpreted at that scale. | |
| Centre effect | Response on nicotinamide: 0% and 21.7% |
| ReadingNo responders out of 24 on nicotinamide at the first centre, 5 out of 23 at the second. The Breslow-Day test gives weak evidence of a difference between centres. The authors list this centre effect among the limitations and describe it as unexplained. | |
| Course at twelve weeks | Reduction of 7.50 points, all groups combined |
| Reading95% CI 6.31 to 8.69. Measured after every group had received at least one 25 mg dose, which removes the comparison. No difference between the four arms, no p value reported, the second phase being entirely exploratory. It cannot be read as evidence of a treatment effect. | |
| Suicidal ideation on dosing days | 4% against 1 to 2% |
| ReadingSix administrations out of 160 of psilocybin 25 mg, one out of 72 of psilocybin 5 mg, one out of 48 of nicotinamide: the denominators are administrations, not patients. In parallel, the Columbia scale analysis finds no differential worsening of suicidal ideation up to week six, 34.0%, 33.3% and 34.0% across arms, p = 0.996. Patients at high suicide risk were excluded. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Randomisation | Low risk |
| FindingPermuted blocks stratified by centre, generated with an independently managed online tool. Identical nicotinamide capsules formulated by the pharmacy of the University of Mainz, blinded treatment labelling. Baseline characteristics comparable across groups. | |
| Protocol deviations and blinding | High risk |
| FindingFunctional unblinding was measured and is massive in participants and in therapists. It was not assessed in raters, and patients’ expectations were not measured: the extent to which expectancy effects contributed to the observed differences therefore cannot be determined, and the authors say so. One descriptive analysis nonetheless runs against the simplest hypothesis: among participants given 25 mg, the response rate at six weeks is 15.8% in those who thought they had received the high dose and 33.3% in the six who thought otherwise. The authors conclude that functional unblinding of the first dose “played a negligible role” in the week-six outcomes, a conclusion that the size of the subgroups makes fragile. | |
| Missing data | Low risk |
| Finding143 of the 144 randomised participants received at least one administration, 137 reached the primary outcome per protocol, 142 are included in the primary efficacy analysis, with comparable discontinuation rates across groups. Estimand: a while-on-treatment strategy in the sense of the ICH E9 (R1) addendum. No imputation of missing values, the dropout rate remaining below the planned 10% threshold. The principal stratum sensitivity analysis gives consistent results, odds ratio 1.92, interval 0.55 to 7.73. | |
| Measurement of the outcome | Moderate risk |
| FindingGerman version of the Hamilton scale, rated by trained local investigators not involved in therapy, with sharing of information about dosing sessions discouraged. This arrangement limits rater unblinding without documenting it. The authors also note that the Hamilton scale, with its emphasis on somatic symptoms, might be less sensitive to the emotional and cognitive effects of psilocybin. | |
| Selection of the reported result | Moderate risk |
| FindingA negative primary outcome alongside favourable secondary outcomes creates a risk of selective emphasis. The confirmatory testing procedure was also modified during peer review, departing from the filed statistical analysis plan. To their credit, the authors publish the change log, report the secondary p values as nominal and uncorrected, present the second phase as entirely exploratory and write that “overall this constituted an inconclusive trial”. | |
| Conflicts of interest | Numerous and declared |
| FindingInstitutional funding is public, but personal declarations are abundant: co-founders or shareholders of OVID Health Systems, OVID Clinic Berlin, OVID Tagesklinik and the MIND Foundation among several authors including the last author; study support or travel expenses from Beckley Psytech, MindMed, Compass Pathways, Cybin; fees from Takeda, Medice, Boehringer Ingelheim, Bristol Myers Squibb, Johnson & Johnson, Lundbeck, Otsuka, Roche and ROVI. These ties are declared as outside the submitted work. | |
| External validity | Limited |
| FindingA socioeconomically and ethnically homogeneous sample, 98% White participants, 35% with postgraduate education, probable self-selection bias. Psychiatric comorbidity in 78% of participants, including 19% with personality disorders, which the authors themselves judge “unusually high”. No follow-up beyond twelve weeks in this publication. Patients at high suicide risk or with very low functioning were excluded. | |
What the trial does not tell us
Three questions lie outside the scope of this trial. No comparison with a conventional antidepressant was made: no conclusion about comparative efficacy is possible. Durability beyond twelve weeks is not reported here, the six- and twelve-month assessments being planned in the analysis plan but not published in this article. Finally, the respective contributions of the molecule and of the psychotherapeutic setting, fourteen hours of structured support with two therapists, cannot be separated: every arm received the same support, which neutralises that variable in the comparison but rules out attributing the result to the molecule alone.
Level of evidence
Oxford level of evidence 1b, individual randomised trial. This classification describes the design, not the precision: level 1b assumes a narrow confidence interval, and that of the primary outcome, from 0.53 to 6.23, is plainly not narrow. Confidence is reasonable on two points: the fact that the trial did not demonstrate an effect on its primary outcome, and the description of adverse events, collected here more systematically than in earlier trials. It is low on the existence of a real antidepressant effect, in either direction: the confidence interval remains compatible with a clinically useful benefit, power was calculated on an overly optimistic effect assumption, and unblinding prevents us from saying what belongs to the molecule and what belongs to expectation.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“This is one of the best-built trials in the field, and it does not reach a conclusion. When eighty-six per cent of those who get the high dose can guess it, you are no longer quite sure what you are measuring. What I take away is the dosing-day signal, and the fact that they went looking for it.”
What this means in practice: to a patient who asks whether psilocybin is going to become a treatment option, one can answer that the best trial conducted so far did not reach its objective, and that the question remains open rather than settled.
What you can do with this
- Be ready to answer a patient who has read an enthusiastic article: the most rigorous trial available did not show superiority on its primary outcome, and its authors present it as inconclusive.
- Do not conclude from this that psilocybin is ineffective. A confidence interval that includes no effect also includes an appreciable benefit, and that symmetry holds here for psilocybin as for its comparators.
- Keep in mind the dosing-day signal and the nature of the setting: a six to eight hour session, an overnight hospital stay, two therapists, fourteen hours of preparation and integration. This is an intensive care arrangement, not a prescription.
- Know the exclusions chosen by the investigators, which show where the risk was judged unacceptable: personal or first-degree family history of psychotic or bipolar disorder, cluster A or borderline personality disorder, post-traumatic stress disorder, recent active suicidal ideation with intent, suicidal behaviour within the past year, current lithium, electroconvulsive therapy within the past twelve months, ketamine or esketamine within the past six months, pregnancy.
- Use this trial as a textbook case of the blinding problem in studies of substances with marked subjective effects. The difficulty applies to other fields too, from brain stimulation to behavioural interventions.
- A pharmacological reminder, outside the scope of the trial: the nicotinamide used here at 100 mg is a form of vitamin B3, used as a comparator and not as a treatment for depression.
Frequently asked questions
Does the EPIsoDE trial show that psilocybin does not work for depression?
No. It fails to demonstrate that it works, which is different. The confidence interval of the primary outcome remains compatible with a real effect, and the authors themselves describe their trial as inconclusive rather than negative in the sense of a refutation.
Why not rely on the significant secondary outcome of the psilocybin trial?
Because it was not the trial’s decision criterion. The pre-specified testing procedure stops as soon as the primary outcome fails: the secondary p values are reported as nominal and uncorrected, and therefore do not control the risk of a false conclusion.
Why did EPIsoDE use two different comparators, nicotinamide and low-dose psilocybin?
Because they do not serve the same purpose. Nicotinamide, chosen because niacin is not approved in the European Union according to the authors, functions in their words largely as an inert placebo. It was psilocybin 5 mg that was meant to produce mild subjective effects and make allocation harder to guess. The result shows that this arrangement was not enough.
What does the improvement seen at twelve weeks mean?
Nothing about comparative efficacy. By then, every participant had received at least one 25 mg dose, and there was no longer a reference group. The mean reduction of 7.50 points is real but cannot be attributed to the treatment rather than to spontaneous course, to psychotherapeutic support or to regression to the mean.
Does functional unblinding invalidate the results of psilocybin trials such as EPIsoDE?
It weakens the result without cancelling it. The authors’ descriptive analysis even suggests that unblinding of the first dose weighed little on the week-six outcomes, but it rests on very small subgroups, one of them of six participants. The right reading is uncertainty, not the opposite certainty.
Annotated bibliography
Source study. Mertens LJ, Koslowski M, Betzler F, et al. Efficacy and Safety of Psilocybin in Treatment-Resistant Major Depression: The EPIsoDE Randomized Clinical Trial. JAMA Psychiatry. 2026;83(5):448-460. doi:10.1001/jamapsychiatry.2026.0132. Published online 18 March 2026, open access under a CC-BY-NC-ND licence. Trial registered as NCT04670081 and EudraCT 2019-003984-24.
Protocol and analysis plan. Mertens LJ, Koslowski M, Betzler F, et al. Methodological challenges in psychedelic drug trials: Efficacy and safety of psilocybin in treatment-resistant major depression (EPIsoDE), rationale and study design. Neuroscience Applied. 2022;1:100104. doi:10.1016/j.nsa.2022.100104. A design paper published before the trial, useful for checking what was pre-specified.
Reference trial in the same indication. Goodwin GM, Aaronson ST, Alvarez O, et al. Single-Dose Psilocybin for a Treatment-Resistant Episode of Major Depression. N Engl J Med. 2022;387(18):1637-1648. doi:10.1056/NEJMoa2206443. Larger sample, primary outcome assessed earlier, and antidepressants resumed by 17% of participants in the 25 mg arm by week six, which complicates direct comparison with EPIsoDE.