# Lesson 8: Psilocybin, Ketamine and Depression, Anxiety and End-of-Life Distress

*Companion-podcast transcript • Sarah & Kiffer*  
*Office Hours episode to listen to after working through the lesson*

---

**Sarah:** Two research teams pooled overlapping sets of trials of psychedelics for depression and found almost the same effect, a little over 0.6 on a standardized scale. One team rated the certainty of that evidence as moderate. The other rated it as very low.

**Kiffer:** Both teams were reading many of the same trial reports. Much of the gap between their ratings seems to come down to one judgement, which is how much it matters that many people in these trials can tell whether they've had the drug.

**Sarah:** So the two estimates of the effect agree, and the confidence we should place in them is what's in dispute.

**Kiffer:** That's a fair way to put it, and it's an argument Sarah and I come at from slightly different directions.

**Sarah:** Welcome back to Office Hours. I'm Sarah.

**Kiffer:** And I'm Kiffer. How far to trust a result is a question that runs all through Lesson eight, Psilocybin, Ketamine and Depression, Anxiety and End-of-Life Distress, so let's start with the trials in people facing the end of life.

**Sarah:** Question one. The two psilocybin trials in people with cancer, published in 2016, reported that most participants were still doing better six months later, and a follow-up of the New York group found benefits about four and a half years on. What do those long-term numbers show?

**Kiffer:** Both trials were crossovers, so the randomized comparison lasted five weeks at Johns Hopkins and seven weeks at New York University. After that, everyone had received psilocybin. When Griffiths and colleagues report that seventy-eight percent had a depression response at six months, there's no untreated group left to compare with. The figure shows that improvement was maintained in treated people. It can't show how many would have improved over those months anyway.

**Sarah:** Why choose a crossover, then, if it ends the comparison so early?

**Kiffer:** Because the participants had serious and often advanced cancer. Asking someone with limited time to wait months for the active treatment, or never receive it, is a hard ethical ask. The crossover gave everyone the active treatment. I think that was a defensible choice, and its price is that the question of how long the effect lasts was left open by design.

**Sarah:** Still, improvement lasting four and a half years after a single dose sounds remarkable, however you got there.

**Kiffer:** It's encouraging, so let's check who was there to be counted. The New York trial randomized twenty-nine people. By the follow-up, sixteen were still alive, and fifteen took part. At four and a half years, about sixty to eighty percent of those fifteen met criteria for a clinically significant response, depending on the measure. How many people is that, and what share of the original twenty-nine? Take a few seconds, and pause the audio if you'd like more time.

*(Pause)*

**Sarah:** Sixty percent of fifteen is nine, and eighty percent is twelve. Nine out of twenty-nine is about thirty-one percent, and twelve out of twenty-nine is about forty-one percent. So a headline about nearly five years of relief rests on nine to twelve people, roughly a third to two fifths of the people who were randomized.

**Kiffer:** And those fifteen are the people who survived and agreed to take part. Four years is also long enough for other treatments and life events to matter. In the Johns Hopkins trial, six participants had already started an antidepressant or an anti-anxiety medication before the six-month visit. What I think the data support is this. After one high dose with a lot of support, improvement was usually maintained, and these designs can't say how much of that maintenance psilocybin caused.

**Sarah:** Meanwhile, some people in Canada have been allowed to use psilocybin legally at the end of life since 2020. If Health Canada approves a request, hasn't it judged that the drug works?

**Kiffer:** That's a common reading, and Health Canada addressed it in writing. Its notice of 27 February 2023, issued under the earlier rules, says that an authorization "is not a broad statement or opinion that the drug is safe, efficacious or of high quality". It describes case-by-case decisions for people whose conventional treatments have failed, are unsuitable or are unavailable.

**Sarah:** And the routes have changed over time. From August 2020 the federal Minister of Health granted section fifty-six exemptions, starting with four people with incurable cancer whose applications TheraPsil supported. CBC News reported seventy-nine patient exemptions by February 2022. An exemption removed the legal barrier and didn't supply any psilocybin, so patients had to find their own.

**Kiffer:** Then from January 2022, practitioners could request it through the Special Access Program, with the drug coming from an approved manufacturer. The Federal Court recorded that psilocybin had been authorized for one hundred sixty-one patients by November 2023, and since the first of October 2026 the program has operated under the new Controlled Substances Regulations. Each of those authorizations is a decision about particular patients' circumstances. Whether psilocybin works for distress at the end of life is a question for trials, and in June 2023 the federal government announced nearly three million dollars for three psilocybin trials, one of them in people with advanced cancer.

**Sarah:** Question two. When a company announces that a phase three trial succeeded, what do we actually know?

**Kiffer:** Let's use Compass Pathways' psilocybin product, COMP360. In the published phase two b trial in 2022, a single twenty-five milligram dose lowered scores on the Montgomery-Asberg depression scale by 6.6 points more than a one milligram dose, at week three. Then came two phase three trials, both company-reported and neither peer-reviewed as of October 2026. For COMP005, reported on 23 June 2025, the difference compared with placebo was 3.6 points at week six. For COMP006, reported on 17 February 2026, it was 3.8 points compared with two one milligram doses.

**Sarah:** So "succeeded" means each trial met its primary endpoint. The confidence intervals stayed clear of zero.

**Kiffer:** Right, and that's a real finding. It says the difference is probably above zero, and on its own it says little about how large or how lasting the benefit is. The company said on 5 August 2026 that the United States Food and Drug Administration's rolling review was under way, so the regulator's own analysis isn't public yet either.

**Sarah:** The phase three differences are a bit more than half the phase two b difference. Is the drug weaker than people hoped?

**Kiffer:** Maybe, and several other explanations are in play at once. Small early trials tend to overestimate effects, and these programs changed a lot between phases.

**Sarah:** Right. The phase three trials enrolled more participants at more sites, the time point moved from week three to week six, one comparator became an inert placebo, and COMP006 gave two doses. Any of those could move the difference.

**Kiffer:** For a point of comparison, esketamine nasal spray, which is approved, showed a difference of 4.0 points in TRANSFORM-2. So the phase three differences sit close to what an approved treatment showed, and well below what the early trials suggested.

**Sarah:** Then there's the thirty-nine percent in the COMP006 release.

**Kiffer:** Yes. The release said that thirty-nine percent of the twenty-five milligram group had a reduction of at least twenty-five percent, which is half the usual threshold for a response. It gave no conventional response or remission rates and no efficacy result for the ten milligram arm.

**Sarah:** Let me see what that threshold catches. A hypothetical participant starts at thirty-two and ends at twenty-three. That's a drop of nine points, and nine divided by thirty-two is about twenty-eight percent. So she's counted in the thirty-nine percent. A response would need a fifty percent drop, to sixteen or lower, so she's a long way from one.

**Kiffer:** And without the same figure for the comparison arm, the thirty-nine percent can't give you a number needed to treat at all. The LSD results from Definium Therapeutics, the company called MindMed until January 2026, are easier to work with. In the Voyage trial in generalized anxiety disorder, company-reported on 12 August 2026, a single one hundred microgram dose lowered Hamilton anxiety scores by 5.4 points more than placebo at week twelve.

**Sarah:** And that release gave categories. Response was forty-three percent compared with sixteen percent, which works out to a number needed to treat of about four. Remission was fourteen percent compared with four percent. So here's one for everyone. What's the number needed to treat for remission? Take a few seconds.

*(Pause)*

**Kiffer:** Fourteen percent minus four percent is ten percentage points, or 0.10. One divided by 0.10 is ten. So about ten people would need the dose for one more person to reach remission, compared with about four for one more response. The release gives only rounded percentages, so with roughly one hundred seven people per arm we can only approximate the confidence interval, and it's wide, from about five to thirty or more.

**Sarah:** Same trial, same drug, and the number needed to treat more than doubles depending on the outcome you pick.

**Kiffer:** Which is why a release that reports one favourable category, or a non-standard category, deserves careful reading. For any topline result, I'd want the comparator, the time point, conventional response and remission rates in every arm, and the date of the announcement, and then the peer-reviewed paper when it appears.

**Sarah:** Question three is about ketamine. The ELEKT-D trial compared intravenous ketamine with electroconvulsive therapy, or ECT, in people with treatment-resistant depression. Response was 55.4 percent with ketamine and 41.2 percent with ECT. Did ketamine outperform ECT?

**Kiffer:** The trial was designed to ask a narrower question, whether ketamine was worse than ECT by more than ten percentage points. It was open-label, and the outcome was self-rated. There's also a withdrawal problem. Four hundred three people were randomized, two hundred to ketamine and two hundred three to ECT. Before treatment started, five withdrew from the ketamine arm and thirty-three from the ECT arm. The response rates come from the people who began treatment, which works out to one hundred eight of one hundred ninety-five and seventy of one hundred seventy.

**Sarah:** I'd like to try something with those numbers.

**Kiffer:** Go ahead. This is a teaching reanalysis we're doing ourselves from the published counts, so the figures we get are our own. Count everyone who was randomized, and treat the thirty-eight people who withdrew as people who didn't respond. What are the two response rates, and what's the difference? Everyone else, try it too. Take a few seconds.

*(Pause)*

**Sarah:** Okay. Ketamine is one hundred eight out of two hundred, which is fifty-four percent. ECT is seventy out of two hundred three, about 34.5 percent. The difference is about 19.5 points, and one divided by 0.195 is about 5.1, so a number needed to treat of six. So the cautious analysis makes ketamine look even better.

**Kiffer:** Counting withdrawals as failures often gets called the cautious choice. Cautious for which arm, though?

**Sarah:** Oh, I see. Thirty-three of the thirty-eight withdrawals were in the ECT arm, so scoring every one as a failure mostly penalizes ECT. That's only cautious when both arms lose about the same number of people. The cautious case for ketamine runs the other way. The five ketamine withdrawals fail, and all thirty-three ECT withdrawals respond.

**Kiffer:** That assumption is deliberately unfair to ketamine, which is the point of a worst case. Try it.

**Sarah:** Ketamine stays at one hundred eight of two hundred, fifty-four percent. ECT becomes seventy plus thirty-three, which is one hundred three of two hundred three, about 50.7 percent. So the difference falls to about 3.3 points.

**Kiffer:** Now ask the question the trial was built for. Could ketamine be more than ten points worse? Using the Wilson score method from the lesson, the confidence interval for that worst-case difference runs from about minus 6.4 to plus 12.9 percentage points. The lower limit stays above minus ten, so ketamine still meets the noninferiority criterion. The noninferiority conclusion survives an assumption chosen to hurt ketamine, and the fourteen-point lead shrinks to a difference the interval can't separate from zero.

**Sarah:** And the open-label design and the self-rated outcome are still there. ECT's 41.2 percent is also well below the sixty-five to seventy-five percent response that CANMAT, the Canadian Network for Mood and Anxiety Treatments, cites for ECT.

**Kiffer:** So I'd read ELEKT-D as good evidence that intravenous ketamine is a reasonable alternative for people referred for ECT, and as weak evidence that it's better.

**Sarah:** That fits CANMAT's 2023 update, which places intravenous ketamine and esketamine at second line, and oral, nasal and intramuscular racemic ketamine at third line, because protocols vary and results are mixed. And that ranking matters for the clinic market. A 2026 study found two hundred thirty-three clinics advertising ketamine in the New York area, and 36.5 percent of them prescribed it for use at home.

**Kiffer:** That's where I'd be most careful. The trials behind ketamine's second-line place gave it intravenously or as a supervised spray, with monitoring after each dose. Oral ketamine taken at home differs in route, dose and supervision, so that evidence doesn't carry over to it automatically.

**Sarah:** Before question four, a short content note, because this part discusses suicidal thoughts and behaviour in clinical trials. If you'd like support, you can call or text nine eight eight, the Suicide Crisis Helpline, at any time. The Hope for Wellness Help Line for Indigenous people is at one eight five five, two four two, three three one zero. So, question four. When a small trial reports no serious adverse events, how reassured should we be?

**Kiffer:** Let's start with the 2016 cancer trials. Between them, eighty people received at least one session, and neither trial reported a serious adverse event attributed to psilocybin. A zero in a small sample still leaves room for real risk. There's a handy rule of thumb, sometimes called the rule of three. When nothing has happened in a group, you can be about ninety-five percent confident that the true rate is below three divided by the number of people. For eighty people, that's three divided by eighty, about 3.75 percent.

**Sarah:** So apply it to the New York trial on its own, with twenty-nine people. What's the highest rate that zero events would still leave plausible? Take a few seconds.

*(Pause)*

**Kiffer:** Three divided by twenty-nine is about 0.10, so about ten percent. Zero serious events in twenty-nine people is compatible with a true rate as high as about one person in ten.

**Sarah:** And the larger trials did record serious events. Why wouldn't the small ones?

**Kiffer:** Let's turn the rule around with a hypothetical. Suppose some serious event had a true rate of one percent. What's the chance that eighty people would show none at all?

**Sarah:** Each person has a ninety-nine percent chance of no event, so for eighty people it's 0.99 multiplied by itself eighty times. That comes to about 0.45.

**Kiffer:** About forty-five percent, so close to a coin flip. Now look at the larger trials. In the published phase two b trial of COMP360, suicidal ideation or behaviour or self-injury occurred in every dose group, and the trial's registry record lists serious adverse events of suicidal behaviour in three participants in the twenty-five milligram group. For the phase three trials, the company reported on 17 February 2026 that serious adverse events of suicidal ideation occurred in fewer than one percent of participants across both trials, with one serious event of suicidal behaviour in COMP006, in the one milligram arm.

**Sarah:** So an event rate under one percent, across more than eight hundred people, is the kind of thing a pair of trials with eighty people between them would miss about half the time or more.

**Kiffer:** Roughly, yes, with the caveat that the populations differ. The trials also screened out people at higher risk. The trial of psilocybin and niacin in one hundred four adults, for example, excluded people with active suicidal ideation with intent. Lesson six looked at these events and at screening in detail.

**Sarah:** Suicidality can also be the outcome a trial measures. Esketamine's approval in the United States mentions suicidal ideation.

**Kiffer:** On 3 August 2020 the FDA added an indication for depressive symptoms in adults with major depressive disorder with acute suicidal ideation or behaviour. The ASPIRE trials behind it enrolled two hundred twenty-six and two hundred thirty adults who were hospitalized with active suicidal ideation with intent, all receiving standard care. Depressive symptoms fell more with esketamine at twenty-four hours, by 3.8 and 3.9 points. The severity of suicidality fell quickly in both groups and didn't differ significantly between them.

**Sarah:** So the finding is faster relief of depressive symptoms in people with acute suicidal ideation, and the trials couldn't separate the two groups on suicidality itself.

**Kiffer:** Yes, and the manufacturer's own announcement in 2020 said that the effectiveness of esketamine in preventing suicide, or in reducing suicidal ideation or behaviour, had not been demonstrated. When suicidality appears in an indication, check whether it describes the population or the outcome.

**Sarah:** Okay. Now the part where we disagree, which takes us back to those two meta-analyses.

**Kiffer:** Let me set out the first one. Metaxa and Clarke pooled randomized trials of psilocybin for depression. After the BMJ issued an expression of concern about calculation errors in May 2024, they corrected the analysis, and the corrected version, published in 2025, gives a standardized effect of 0.66 from seven trials and four hundred thirty-six participants, with moderate certainty.

**Sarah:** And the second is the living review that Højlund and colleagues keep, searched to July 2025. They judged 83.3 percent of thirty trials to be at high risk of bias, found a pooled effect of about 0.62 for psilocybin, ayahuasca and LSD in major depressive disorder, and rated the certainty as very low using the GRADE approach.

**Kiffer:** So the two pooled effects are close, and the ratings are two steps apart. Where do you land, Sarah?

**Sarah:** I think very low is closer to right. In the published LSD phase two b trial, 92.5 percent of people given one hundred micrograms had visual perceptual changes, compared with about ten percent on placebo. In the Panorama trial, company-reported on 14 September 2026, sixty-eight percent of the one hundred microgram group reported illusions on the dosing day. People can usually tell what they got. Metaxa and Clarke's own analysis found larger effects with self-rated scales and with more participants who'd used psychedelics before, which looks like expectancy. And the effects have shrunk as the trials have grown.

**Kiffer:** Those are fair points, and I'd still say very low undersells the evidence. The effect has appeared with very different comparators, a waiting list, niacin, a low dose of psilocybin and an inert placebo, and several trials used raters who didn't know the allocation. In the escitalopram trial, participants expected more from psilocybin, yet expectancy predicted improvement with escitalopram and didn't predict it with psilocybin. My reading is a real effect of modest size, smaller than the early trials suggested.

**Sarah:** The blinded raters help less than they seem to. The rater is unaware of the allocation. The participant usually knows, and the participant is the one describing their symptoms. And that expectancy analysis comes from one trial of fifty-nine people, by members of the original research group.

**Kiffer:** That's right, and I'd treat that analysis as a hypothesis for independent trials to test. Where we may meet is CANMAT's position. Its 2023 update classified psilocybin-assisted psychotherapy as investigational, citing the lack of blinding and inadequate assessment of expectancy. I think that's the right clinical label for now, whichever certainty rating you prefer.

**Sarah:** I'm more skeptical than Kiffer, but we agree on the practical rule. When you cite this evidence, give the pooled effect with both certainty ratings, say that they seem to differ mainly over functional unblinding, and look for trials that asked participants to guess what they received or that added a lower active dose, as Panorama did.

**Kiffer:** Let's pull it together with three things to take away.

**Sarah:** First, in a crossover trial the randomized comparison ends at the crossover. Later figures describe maintained improvement in people who have all been treated, and a long follow-up counts only the people still there to be counted.

**Kiffer:** Second, a company-reported topline tells you that a primary endpoint was met, as of a given date. To judge the size of the benefit, you need the comparator, the time point, conventional response and remission rates in every arm, and eventually the peer-reviewed report.

**Sarah:** And third, look at what the denominators can hide. Unequal withdrawals can make a cautious-sounding analysis favour one arm, and zero serious events in eighty people still leaves room for a rate of almost four percent.

**Kiffer:** If you'd like more practice, rework the ELEKT-D withdrawal problem from this episode with the opposite assumption, where every ketamine withdrawal responds and every ECT withdrawal doesn't, and say what that best case can and can't tell you.

**Sarah:** Next time, it's Lesson nine, MDMA-Assisted Therapy for PTSD: Evidence, Regulation and Research Integrity, where questions about blinding and expectancy come back in the FDA's 2024 decision on MDMA.

**Kiffer:** Take care, everyone.

**Sarah:** See you in Lesson nine.
