Primary Outcome, Negative. Subgroup, Promising. Now What?

A press release crossed my feed announcing promising results from a psychedelic trial. Reading the actual paper, the pre-specified primary outcome was negative. The optimism rested on a secondary finding in a subset of participants. Both the press release and the paper were accurate. They were just answering different questions, and only one of them was the question the trial was built to answer.

This happens constantly, and clinicians are handed the headline version without ever being taught to read past it. Here are the moves that do most of the work, each illustrated with a real study from the psychedelic and ketamine literature that you can pull up yourself.

Every well-designed trial names its primary outcome in advance, along with how many participants it needs to detect an effect. That pre-registration is what protects against fishing through the data for anything that looks good. An exploratory or secondary outcome is a hypothesis for next time, not a result you can bank.

The psilocybin cluster-headache work is a clean teaching case. The randomized controlled trial's efficacy outcomes were negative, owing in part to its small size, yet the coverage leaned on the more encouraging exploratory and follow-up signals. The same pattern showed up in a psilocybin depression trial where the primary endpoint was not statistically significant while secondary measures were reported as clinically meaningful. When a press release quietly shifts from the primary outcome to a secondary one, that shift is the story.

An open-label study, where everyone knows they got the drug, can generate a strong-looking result that evaporates under control. The magnesium-ibogaine veteran data are a good example: a prospective study reporting large symptom reductions, with 30 participants, no control group, and people who had sought out and paid for the treatment. That is a hypothesis worth testing, not a demonstrated effect. It is also the same uncontrolled study that gets cited as evidence ibogaine is safe, which is a separate problem I take up in ibogaine's cardiac risk profile.

The stronger designs use a placebo, and the best use an active placebo that mimics the drug's noticeable effects so participants cannot easily tell which arm they are in. The recent migraine trial did exactly that, comparing psilocybin against diphenhydramine as an active placebo. The placebo group also improved, which is precisely the information an inert-sugar-pill design would have hidden. I read that trial in more detail in a piece on the psilocybin migraine results.

A trial can show a real benefit that is not caused by what you think. Mediation analysis asks what the effect actually runs through. The sharpest example in this literature: in a cohort of 329 chronic-pain patients, pretreatment depression was the sole mediator of ketamine's pain relief. The pain improved, but the analysis suggested ketamine was working largely by treating depression rather than by acting on pain directly. Same outcome, different mechanism, and very different clinical implications, which I follow into the exam room in what ketamine might actually be treating in chronic pain.

Psychedelics are extraordinarily hard to blind, because people usually know whether they are having a profound experience. When the blind breaks in practice, it is called functional unblinding, and it lets expectancy inflate the result: people who know they got the active drug, and who often signed up because they believe in it, report more improvement.

This is not hypothetical. When the FDA declined to approve MDMA-assisted therapy for PTSD in 2024, its reviewers and advisory committee flagged functional unblinding and expectancy bias as central problems, noting that many participants had prior MDMA experience and could tell they were in the treatment arm. A large effect in an unblinded-in-practice trial is not necessarily a large drug effect.

Trials stopped before their planned end tend to overestimate effects, because they often stop on a striking interim result that regression to the mean would have softened. Early stopping is sometimes the right ethical call, but it changes how you read the number. A trial testing whether naltrexone blocks ketamine's antidepressant effect was halted at interim analysis because the blockade was so pronounced that continuing seemed unjustified. The finding is important, and it rests on a small, truncated sample. Both things are true, and I unpack what that early stop does and does not license in what naltrexone reveals about ketamine's mechanism.

When the next trial crosses your feed, run these before you update your practice:

1. Was the reported result the pre-specified primary outcome, or a secondary or subgroup finding?
2. Was there a control group, and a blind that plausibly held?
3. If psychedelic, how did they handle functional unblinding and expectancy?
4. Does a mediation or secondary analysis suggest the effect runs through something other than the drug?
5. What was the sample size, and did the trial stop early?
6. Who wrote the summary you are reading, and what do they gain if you believe it?

That last question is not cynicism. The hype economy around psychedelics rewards the optimistic read, because attention and investment follow it. Methods literacy is how a clinician stays useful to patients in that environment: not credulous, not dismissive, but able to say what a study actually established.

Get the free clinician guide for the conversations that follow when a patient acts on a headline.

The primary outcome is the single result a trial is designed and powered to test, declared before the study begins. Exploratory or secondary outcomes are additional analyses that generate hypotheses for future research. A positive exploratory finding alongside a negative primary outcome is a reason to study further, not a demonstrated effect.

Functional unblinding happens when participants can tell whether they received the active drug despite the study's blinding, which is common with psychedelics because the subjective effects are obvious. This lets expectancy inflate results, since people who know they got the treatment, especially those who believe in it, tend to report more improvement. It was a central FDA concern in the 2024 MDMA decision.

Trials halted before their planned endpoint tend to overestimate the true effect, because they often stop on a striking interim result that would have moderated with more data. Early stopping can be ethically correct, but it leaves a smaller, truncated sample. Treat the effect size as provisional and look for replication before changing practice.

An active placebo produces noticeable bodily effects, such as diphenhydramine causing drowsiness, so participants cannot easily guess whether they received the real drug. This preserves the blind better than an inert sugar pill, which participants in psychedelic or ketamine trials can often identify by the absence of any effect. Active-placebo designs give a more honest estimate of the drug's specific benefit.