# Lesson 6: Safety, Screening and Risk Management

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

---

**Sarah:** Picture a man at a screening visit for a hypothetical ibogaine study. He's nervous, and his heart is running at a hundred beats a minute. His QT interval is three hundred sixty-five milliseconds. Under one standard correction for heart rate, he's excluded. Under the other, he's eligible.

**Kiffer:** The tracing, the man and the cut-off are all the same. The only thing that changes is the arithmetic, and the two formulas involved are the corrections in most common use.

**Sarah:** Which makes me wonder how many of the thresholds in screening are firmer on paper than they are in the evidence.

**Kiffer:** Quite a few of them rest on less evidence than their precision suggests. A cut-off is a decision that someone made, usually a cautious one, and it's worth asking what each one rests on.

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

**Kiffer:** And I'm Kiffer. That gap between a firm-looking line and the evidence underneath it runs all the way through Lesson six, Safety, Screening and Risk Management, so let's start with that man's heart.

**Sarah:** Question one. When a protocol sets a limit for the corrected QT interval, why does it matter which correction you use?

**Kiffer:** It matters because the two common corrections agree only at sixty beats a minute. The QT interval shortens as the heart speeds up, so it's corrected to a standard rate. Bazett's formula divides the measured QT by the square root of the RR interval, the time between beats in seconds, and Fridericia's divides it by the cube root. A scientific statement from the American Heart Association and the American College of Cardiology Foundation notes that Bazett's formula tends to give values that are too long at fast heart rates, particularly above about eighty-five beats a minute. The Dutch hospital study of ibogaine excluded men with a corrected QT above four hundred fifty milliseconds, although the paper doesn't say which correction it used for that limit.

**Sarah:** So here's one for everyone, using the man from the start, who is made up. A QT of three hundred sixty-five milliseconds at a hundred beats a minute. What are his two corrected values, and which side of four hundred fifty does each one fall on? Take a few seconds, and pause the audio if you'd like more time.

*(Pause)*

**Sarah:** Okay, I'll try it. First the RR interval. That's the heart rate divided by sixty, so a hundred divided by sixty is 1.67 seconds. The square root of 1.67 is about 1.29. Three hundred sixty-five divided by 1.29 is about two hundred eighty-three milliseconds. So by Bazett he's far below the line.

**Kiffer:** Before you go on to Fridericia, compare it with the measured value. Which way should a correction move the QT when the heart is beating fast?

**Sarah:** It should go up. A fast heart shortens the QT, so the corrected value should come out longer than the three hundred sixty-five we measured, and I've made it shorter. And a gap of more than a second between beats would mean a rate below sixty. I flipped the fraction. The RR interval is sixty divided by the heart rate, so sixty divided by a hundred, which is 0.6 seconds.

**Kiffer:** Good catch. Try it again with an RR interval of 0.6 seconds.

**Sarah:** The square root of 0.6 is about 0.775, and three hundred sixty-five divided by 0.775 is about four hundred seventy-one. The cube root of 0.6 is about 0.843, and three hundred sixty-five divided by that is about four hundred thirty-three. So Bazett puts him twenty-one milliseconds over the line, and Fridericia puts him seventeen under it.

**Kiffer:** And notice the direction, because it's worth remembering. At a fast rate Bazett gives the higher number, so a clinic that uses it would tend to exclude more people whose hearts are racing at screening. With Fridericia's correction, this man passes a four hundred fifty limit; with Bazett's, he fails it.

**Sarah:** Can we push it a step further? Suppose he's enrolled. What happens to that comparison once he's had the drug?

**Kiffer:** That's where it gets harder. In cell studies, ibogaine and its metabolite noribogaine block a potassium channel that the heart needs to recover after each beat, and noribogaine has a half-life of twenty-eight to forty-nine hours. In the Dutch study the largest lengthening in a typical participant was about ninety-five to one hundred milliseconds, and half of the fourteen participants passed five hundred. Ibogaine can also slow the heart. So imagine that the next morning his heart rate is sixty and his QT is four hundred ninety-five. What's his corrected value?

**Sarah:** At sixty beats a minute the RR interval is one second, so both formulas leave it alone. It's four hundred ninety-five. That's under five hundred, and that cardiology statement calls a value above five hundred highly abnormal.

**Kiffer:** It is. The same statement also calls for prompt action when the corrected QT rises sixty milliseconds or more above its value before the drug, so the next step is to compare it with his baseline. Here's the trap. Suppose his screening report listed the Bazett value.

**Sarah:** Then the change looks like four hundred ninety-five minus four hundred seventy-one, which is twenty-four. That looks like nothing to worry about. With the Fridericia baseline it's four hundred ninety-five minus four hundred thirty-three, which is sixty-two. That crosses the sixty millisecond rule.

**Kiffer:** So a baseline inflated by a nervous heartbeat can hide most of a real change. Using Bazett at both time points wouldn't help, because at sixty beats a minute it also gives four hundred ninety-five. The practical rule is to pick a correction that holds up at fast heart rates, such as Fridericia, use it at every time point, and repeat a baseline taken with a racing heart once the heart has settled. And even with careful methods, the Dutch authors note that their ECGs were taken at intervals and could have missed brief episodes of a dangerous rhythm.

**Sarah:** Question two. A clinic or a trial reports no serious adverse events. What has it actually shown?

**Kiffer:** It has shown less than it seems to. Start with the pooled evidence. A 2024 meta-analysis of one hundred fourteen studies of classic psychedelics, with three thousand five hundred four participants, found no serious adverse events among healthy participants, and serious adverse events in about four percent of participants who had a neuropsychiatric condition. It also found that only sixteen of the sixty-eight studies published since 2005 described a systematic assessment of adverse events.

**Sarah:** So fewer than a quarter of recent studies say they looked systematically.

**Kiffer:** Right, and that changes how to read a zero. Here's a made-up case. A clinic advertises that none of its first two hundred patients had a serious adverse event, and its records cover the dosing day only. What's the most the true risk could plausibly be, and what does that figure cover? Take a few seconds.

*(Pause)*

**Sarah:** By the rule of three, the upper limit of the ninety-five percent confidence interval is about three divided by the number of people, so three divided by two hundred, which is 0.015, or one and a half percent. So zero events among two hundred patients is still compatible with a true risk of up to one and a half percent.

**Kiffer:** And what about the second half of the question, what that figure covers?

**Sarah:** It only covers the dosing day. Anything that happened a week or a month later was never counted, so the clinic's zero says nothing about it.

**Kiffer:** That's the part that's easy to miss. A count of zero describes the events someone looked for, in the time window they looked, among the people they enrolled. The rule of three gives the bound for that question only.

**Sarah:** Then there's the other window, the population studies, which seem to tell a different story. The Ontario cohorts found much higher rates of psychotic and manic disorders after hospital care involving hallucinogens.

**Kiffer:** They did. Among more than nine million Ontarians aged fourteen to sixty-five with no history of psychosis, five thousand two hundred seventeen had an emergency visit involving hallucinogens, and within three years about four percent of them were diagnosed with a schizophrenia spectrum disorder, compared with 0.15 percent of the general population. A hazard ratio compares the rate of new diagnoses in two groups over time. Adjusted only for age and sex, it was 21.3. After adjustment for other substance use and mental health conditions, it was 3.53.

**Sarah:** So most of that crude association came from other things these people had going on. But 3.53 is still more than three times the rate. Doesn't that show that hallucinogens cause psychosis?

**Kiffer:** It shows that people who had an emergency visit involving hallucinogens were diagnosed more often afterwards. Two things limit the causal reading. Some of those visits may have been the first sign of an illness that was already developing, which would put the illness before the visit. And an emergency visit marks a severe reaction. In their companion study of mania, the same research team cautions that results like these may not apply to people who never needed acute care.

**Sarah:** So the trial window sees screened people and misses rare events, and the population window sees severe reactions and has trouble sorting out cause.

**Kiffer:** Yes, and neither one describes a screened patient in a supervised session followed over the long term. When someone quotes either kind of study, the first question to ask is which people it describes.

**Sarah:** Before question three, a short content note, because this part discusses suicidal thoughts and behaviour reported 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 three. When suicidal behaviour turns up in one arm of a psilocybin trial, what does it tell us?

**Kiffer:** Let's take the COMP360 phase 2b trial, in which two hundred thirty-three adults with treatment-resistant depression received a single dose of twenty-five, ten or one milligram with psychological support, and the one milligram dose served as the control. Suicidal ideation, suicidal behaviour or self-injury occurred in all three groups. In the trial registry's list of serious adverse events, suicidal behaviour appears in three of seventy-nine people at twenty-five milligrams and in none of the others, while intentional self-injury appears in every group.

**Sarah:** And the sponsor's analysis added the timing.

**Kiffer:** It did. The company reported that everyone with these events had described suicidal thoughts before the trial, that most serious events came at least a week after the session, and that all the suicidal behaviours came at least a month after it, in people who had not responded to treatment.

**Sarah:** So three at the high dose and zero at the control dose. That looks like a fairly clear comparison to me.

**Kiffer:** Let's test that. There were zero cases among the seventy-nine people at one milligram. Using the rule of three, how high could the true risk in that group plausibly be, and how does that compare with what was seen at twenty-five milligrams? Take a few seconds.

*(Pause)*

**Sarah:** Three divided by seventy-nine is about 0.038, so the upper limit is about 3.8 percent. Oh, and three of seventy-nine at twenty-five milligrams is also about 3.8 percent. So the zero in the control group is compatible with a true risk as high as the rate actually seen at the top dose.

**Kiffer:** Right. A zero in a group of seventy-nine leaves room for a risk as large as the one observed in the other group. It also helps to know that in one of the later phase 3 trials, the company reported one serious suicidal behaviour event, and it was in the one milligram group. Depression carries this risk whatever the treatment, including a control dose.

**Sarah:** So what's left of the signal?

**Kiffer:** Two readings fit the timing. One is that the drug can set off a delayed harm in some people. The other is that people who didn't improve went on living with severe depression after a treatment they had hoped would help, which fits the course of the illness itself. With fewer than eighty people per group, the trial can't tell those apart, and the company's report that its monitoring board found no clinically meaningful imbalance in the phase 3 trials remains company-reported until the data are published. My reading is that the question is still open, and that the practical response is the same under either reading.

**Sarah:** And that response is follow-up that lasts.

**Kiffer:** Yes. Health Canada's 2022 expectations for trials call for check-ins in the days and weeks after dosing. The timing in the phase 2b trial argues for carrying that over months, with each contact asking about mood and suicidal thoughts and checking whether the person has responded, so that someone who hasn't improved gets a plan for further care.

**Sarah:** Along with a safety plan built before the session. The evidence for that comes from a different setting. In United States veterans' emergency departments, a safety planning intervention with at least two follow-up calls was associated with suicidal behaviour in about three percent of patients over six months, compared with about five percent with usual care.

**Kiffer:** It comes from a different population, so it gives no estimate of any benefit after psilocybin. I'd still read it as a good reason to build a written plan with follow-up contact into a protocol.

**Sarah:** Question four, and this is where we disagree. Should a family history of bipolar disorder keep someone out of a psilocybin trial?

**Kiffer:** Let me set out the rule first. The 2008 safety guidelines from Johnson, Richards and Griffiths recommend excluding anyone with a current or past psychotic disorder or bipolar one or two disorder, and anyone with a first- or second-degree relative who has one of these disorders. Second-degree relatives are grandparents, aunts and uncles, nieces and nephews, half-siblings and grandchildren.

**Sarah:** That's a lot of people. Can we get a feel for how many it catches?

**Kiffer:** Here's a made-up exercise. Suppose a person has fifteen first- and second-degree relatives, and suppose, with an invented number, that each one independently has a two percent chance of having a psychotic or bipolar disorder. What's the chance that at least one of them does? Take a few seconds.

*(Pause)*

**Sarah:** The chance that a given relative doesn't is 0.98, so the chance that none of the fifteen does is 0.98 multiplied by itself fifteen times, which is about 0.74. So the chance that at least one does is about 0.26, roughly one person in four.

**Kiffer:** That comes from invented inputs, so the twenty-six percent is no estimate for any real population. And real disorders cluster in families, which concentrates affected relatives in fewer families, so with the same two percent, clustering would make the share somewhat lower than the independence assumption gives. Even so, the exercise shows how quickly a second-degree rule grows, because the number of relatives does most of the work.

**Sarah:** I'll take the cautious side anyway. The main concern with a family history of bipolar disorder is mania, and the evidence on genetic vulnerability isn't reassuring. In Swedish adolescent twins, the link between psychedelic use and manic symptoms depended on genetic vulnerability to schizophrenia or bipolar one disorder. An episode of mania can do lasting damage to someone's relationships, money and work, because judgement is what it impairs. I find it hard to see how consent can cover a risk that's still so poorly measured.

**Kiffer:** That's a serious argument. I lean the other way, toward the stratified assessment that Downey and colleagues proposed in 2024. They suggest weighing how closely related the relative is, the type of bipolar disorder, the age at which the relative became ill and the participant's own age. Under their approach, a person in their fifties whose aunt was diagnosed late in life would be assessed differently from a twenty-year-old whose parent became ill at eighteen. And a cross-sectional survey of five hundred forty-eight people found no association between psychedelic use and psychotic symptoms after adjustment, including among people with a family history.

**Sarah:** That's one survey, measured at a single point in time, with self-reported histories. People who had a frightening reaction may never have answered it.

**Kiffer:** That's fair, and I wouldn't change practice on that survey alone. My position is narrower. I'd keep excluding people with a personal history of psychotic or bipolar one disorder, where the evidence of harm is clearest. For family history, I'd want trials that enrol some people with more distant or later-onset family histories under closer monitoring, because otherwise the question can never be answered, and the rule keeps out people who have never been ill.

**Sarah:** And I'd want those trials to report mania scores at every visit and for months afterwards before the rule changes in practice. I'm more cautious than Kiffer, but we agree on the practical rule. For any family history, write down which relative, which disorder and the age at onset, state in the protocol why the criterion is there, and make sure that anyone who is excluded leaves with a referral to other care.

**Kiffer:** That last part matters for fairness. An exclusion protects a person from one risk and leaves them with their illness, so it should always come with a path to treatment. Let's pull it together with three things to take away.

**Sarah:** First, a threshold is only as firm as the measurement behind it. For the corrected QT, know which formula a cut-off was written for, use one that holds up at fast heart rates at every time point, and compare each value with the person's own baseline.

**Kiffer:** Second, a report of zero events gives an upper bound. By the rule of three it's about three divided by the number of people, and it applies only to the events someone looked for, in the window they looked, among the people they enrolled.

**Sarah:** And third, every safety result describes particular people. Trials leave out the people at highest risk, population studies of emergency visits see the most severe reactions, and some serious events come weeks after a session, so follow-up has to run for months.

**Kiffer:** If you'd like more practice, rework the screening ECG problem from this episode at a heart rate of one hundred twenty, and check that both corrected values come out longer than the measured QT.

**Sarah:** Next time, it's Lesson seven, The Therapeutic Process: Preparation, Dosing Sessions and Integration, where the consent, safety plans and follow-up we touched on today become part of the therapy itself.

**Kiffer:** Take care, everyone.

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