What would show a lasting benefit?

Begin with the outcome, not the image

A colourful brain image can show an interesting observation. It does not, on its own, tell us whether someone’s symptoms, functioning or quality of life improved. Before interpreting a result, identify what researchers actually measured and why that outcome matters.1

The studied treatment is more than a name

Psilocybin research commonly involves selected participants, a defined preparation, supervised sessions and psychological support. A result from that setting is not automatically evidence for an unsupervised product or an entirely different protocol. Selection can exclude people with important medical or psychiatric risks, and the support package can affect what participants experience and report. The profile separates the compound from these contexts.2

What should it be compared with?

A comparator helps estimate what might have happened without the intervention being studied. Psychedelic trials have a particular difficulty: noticeable effects may reveal which treatment someone received, even when the design is described as blinded. Expectations and other parts of the treatment setting can then complicate interpretation. A low-dose or active comparator may preserve some expectation of receiving an active intervention, but it brings its own assumptions; no comparator removes every source of bias. Researchers should report whether participants and assessors guessed assignment and how that information was handled.34

Put the finding on a timeline

Ask whether the key outcome was measured after days, weeks or months. Then look at what happened at later visits, including to people who did not complete follow-up. A short-term primary result and sustained improvement are distinct claims. In the 2022 phase 2 trial of a proprietary synthetic formulation, adults with treatment-resistant depression received a single 25-mg, 10-mg or 1-mg control dose with psychological support. At three weeks, the 25-mg group had a larger mean reduction in MADRS score than the 1-mg group, but the pattern at 12 weeks did not establish the same strength of sustained response. In the 2026 German EPIsoDE trial, the primary six-week response comparison between 25 mg and nicotinamide was not statistically significant; secondary mean symptom changes favored 25 mg but were exploratory after the primary result. A later Stockholm single-site trial (35 people) found a day-8 MADRS difference, but its blinded-rater contrast was no longer statistically significant at day 365. Our 2022 study explainer shows why these timelines should not be collapsed into a promise of durable benefit.51012

Read harms with the same attention

A clinical story is incomplete when it reports benefit carefully but treats harms as an afterthought. In the 2022 trial, adverse events occurred in all dose groups and included headache, nausea and dizziness; suicidal ideation, suicidal behaviour or self-injury occurred in all groups. These observations do not prove that every event was caused by psilocybin, but they show why denominators, comparison groups and follow-up matter. In the 2023 Usona/JAMA trial, no serious treatment-emergent adverse events occurred, but overall and severe events were more frequent with psilocybin; the report cautions that psychedelic-specific psychological harms may be missed by routine capture. The 2026 Stockholm study reported two participants with persistent severe anxiety requiring medical attention. EPIsoDE reported two serious adverse reactions after 25 mg, including a case of hallucinogen persisting perception disorder. Psilocybin can produce distressing psychological experiences and physical adverse effects. Clinical supervision and participant selection matter, but they do not remove every risk.291012

What would strengthen a lasting-benefit claim?

Look for a clear definition of improvement, an appropriate comparison, sufficiently long follow-up, transparent handling of missing data and replication. Also ask how the studied population relates to the people for whom a benefit is being claimed. Psychedelic trials face a masking problem because noticeable effects can reveal assignment; psychological support, preparation, medication restrictions and follow-up contacts can also be part of the tested treatment package. FDA’s guidance discusses expectancy, safety monitoring and durability, but a guidance document is a framework for research rather than evidence that a compound works or is approved. None of these features can be replaced by an attractive mechanism animation.18

The question stays open in a useful way

A responsible explanation can describe encouraging observations without presenting an entire treatment approach as settled. A lasting-benefit claim becomes more credible when independent studies reproduce a patient-important outcome, follow people long enough to detect loss of effect and delayed harms, report denominators and withdrawals, and include populations closer to routine care. The 2023 JAMA study was multi-site but sponsored by Usona. It enrolled a selected group of people with major depressive disorder (MDD); only 12.5% met its definition of treatment-resistant depression.

In the 2026 Stockholm study, 16/17 psilocybin and 17/18 niacin participants contributed final self-reports, but masking was weak: 16/17 psilocybin participants correctly guessed assignment at day 365, as did all 16 evaluable niacin participants; the blinded-rater day-365 contrast was not statistically significant.

The 2026 NHS study was single-site and explicitly a feasibility trial: 75 screened, 60 randomized and 59 completed MADRS at all visits. The study reported 287 nonserious adverse events: 164 with psilocybin and 123 with placebo. These are counts of events, not the numbers of people affected. Four serious adverse events occurred across the randomised and open-label phases. None was judged treatment-related among people receiving psilocybin. Its six-week follow-up can inform short-term feasibility and symptom trajectories, but cannot establish late durability or delayed harms. The next experiment should address these uncertainties, rather than merely repeat the most compelling headline.91011