Invisible Variables: what we don't yet understand about psychedelic therapy

On 13–14 June I was at the Psychedelic Science 2026 conference at the University of Warsaw with a talk called “Invisible Variables”. This text is its expanded version – for those who were there and want to revisit the argument at their own pace, and for those who weren’t. I’ve embedded the slides below, but everything that matters is in the text.

Let me start by admitting where I speak from, because it changes the sense of everything I write next. I’m not a sceptic watching from the sidelines. I’m a psychologist and psychotherapist, working day to day close to two worlds that rarely talk to each other – clinical psychology and personality-disorder therapy on one side, clinical research on psychedelics on the other. I’ve spent hundreds of hours preparing patients for sessions, assisting during the sessions themselves, and on integration. I’m involved in research and in supervision. So I speak from the inside, not from a podium.

The title is “Invisible Variables”, though it would be more honest to say “unnoticed”. Because these variables aren’t inaccessible. They can be seen. We simply don’t look at them. And that’s my question – not “do psychedelics work”, but what we don’t yet understand, and what happens beneath the surface of the numbers we use to talk about it.

Let’s assume they work

Let’s make an assumption that spares us a barren argument. Let’s assume they work.

We have grounds for it. The first classic psychedelic already has phase-three data. In the COMP005 trial, psilocybin at a 25 mg dose produced an improvement of about 3.6 MADRS points over placebo at week 6 (p<0.001), and in COMP006 – 3.8 points over the 1 mg dose. Earlier, phase 2b (Goodwin et al., 2022) showed a difference of about 6.6 MADRS points at week 3 and remission in roughly one third of patients with treatment-resistant depression. Efficacy is being demonstrated, and much suggests these drugs will enter wide use.

This isn’t the only data. An earlier trial comparing psilocybin with escitalopram found no significant difference on the primary endpoint, though on some secondary measures psilocybin came out more favourably (Carhart-Harris et al., 2021), and a six-month follow-up of the same trial suggests the effect persists in some people (Erritzoe et al., 2024). The field is heterogeneous, and that’s exactly why it’s worth reading this data carefully, not through headlines.

So I’m not asking whether they work, or whether they should enter use. I’m asking something else. If we assume they work and will spread, then what do we not yet know that we should know before they reach wide access? Because this won’t stay in the hands of a handful of experienced practitioners who have worked on it for years. It will open up to a wider circle – and to conditions far from the ideal conditions of a trial.

Statistics is a tool, not a king

First, something that sounds trite yet keeps getting lost. Statistics is a tool, not a king. Using research doesn’t change reality itself – it doesn’t remove biases, it only changes their form. And I’ll say it plainly. There are no perfect studies. We wrestle with methodology in many ways – from what we actually measure, to functional unblinding, because participants usually know what they got (after a psychedelic it’s hard not to guess), which probably overestimates the observed effects (functional unblinding, J Clin Psychiatry 2025).

It’s worth pausing on one thread here, because it leads straight to the heart of the matter. In a classic study, neither the patient nor the researcher knows who received the drug and who received placebo – this is blinding, which lets us separate the effect of the substance from the effect of the mere hope that “this will help me”. After psilocybin, this mechanism doesn’t work in practice. The experience is so distinctive that a participant almost always knows which group they’re in. That’s a serious question about how much of what we see in the results is an effect of pharmacology, and how much is the effect of being convinced I received something groundbreaking. And – most important to me – it means that expectation is one of the active variables of treatment, not a disturbance you can “subtract out”. Placebo and nocebo here aren’t noise, they’re signal – proof of how much what the patient brings and believes matters. I’ll come back to this, because it’s one of our invisible variables.

I won’t give a lecture on methodology, though. I’ll take one thread, the most important for the consulting room. What we actually measure when we say “improvement”. Because the point isn’t treatment that looks good in a table – it’s the most effective, least burdensome one that produces lasting change.

What is “improvement” actually made of?

If we say “depression decreased by so many points”, we ought to know what those points are about. Depression research most often uses two scales. MADRS – ten items. The Hamilton scale (HAM-D) – seventeen. The clinician (or the patient) rates the severity of individual symptoms, and the points are summed. Sadness, tension, sleep, appetite, concentration, anxiety, somatic symptoms, libido, weight, guilt.

<iframe src="https://lukaszwarchol.com/embed/hamd-scale/" width="100%" height="500" style="border:none;border-radius:8px;" loading="lazy" title="HAM-D Scale – the numbers in context"></iframe>

Notice what “measuring the severity of depression” actually means. We sum points from a narrow list of symptoms. The symptoms themselves – largely sleep, appetite, anxiety, the body. Not a single item about meaning, about relationships, about identity, about the durability of change. And what does “minus 3.6 points” mean? Roughly the distance on this scale between moderate and mild depression. A real – but narrow shift. And precisely in those items. This is the window we look through. The rest of the person stays outside its frame.

There’s one more concept worth knowing here, because it orders the whole conversation about “efficacy” – the minimal clinically important difference (MCID). This is the threshold below which a change exists in the data but the patient doesn’t feel it in their life. For HAM-D, experts proposed a threshold of about 3 points. Imagine two people with depression. One wakes at three in the morning and can’t fall back asleep, the other sleeps poorly but makes it through the night. On the scale that’s maybe 2 points on one item. Only a difference of about 3 points on the whole scale is roughly the boundary where a patient starts saying “I feel that something has changed”. Below it, the change is statistical, not human. Let’s keep this in mind, because when we compare treatments “point for point”, we’re often talking about differences that fall below the threshold of felt experience.

And something that’s rarely said outright. We’ve learned to talk about depression through the lens of scales – so many points on MADRS, so many on HAM-D. It’s useful, because it gives us a shared language and comparability across studies. But depression isn’t a score on a scale. It’s waking with the feeling that the day has no meaning before it’s even begun; the inability to enjoy the presence of your own child; suffering that no questionnaire captures in full. Maybe one day we’ll reach a methodology that grasps it better. For now we have what we have – and we need to know how to read it.

From the average to three stories

And now the most important thing in this whole text. Let’s come down from the level of the average to the level of the person.

We have a group mean – say those minus 3.6 points.

But a mean has a spread; it’s not a point, it’s a cloud. And beneath the spread is scatter – and every dot in that scatter is a particular person.

Let’s zoom in on three dots lying right next to each other. The same score – MADRS 32. Identical. And behind it – three completely different stories. We don’t know the real ones, so these are vignettes; all of them are about depression.

Before we go further, two words of explanation. “MADRS 32” is the severity of significant depression, the same score in all three people. “Group average” and “spread” mean that behind a single number stands a whole cloud of results – some improved by four or five points, others by two, some not at all. And the vignettes below are clinical composites, not real patients.

First. A woman, 38. For a few months the world has dimmed – work, her relationship, half-asleep nights. Before that she coped well. This is an acute episode on stable ground.

Second. A man, 27. Emptiness and fatigue for as long as he can remember. Closeness burns, sleep slips away, sometimes he vanishes from his own body. Here depression is the tip, not the whole mountain – underneath is something far more complex.

Third. A woman, 61. After her husband’s death the days blurred, the phone fell silent. She says it’s probably just old age. Sadness and loneliness – is that already an illness?

The same score. Not the same suffering. Not the same story. And the question I want to leave here. Will each of these people need the same amount of preparation? The same number of sessions? The same guidance during and after? Probably not. And if so, then it isn’t the same treatment either, even though on the group-effect bar it looks identical.

The galaxy of variables

Three stories – and yet there are far more than three variables. Behind that one score stands a whole galaxy of factors that abstracts barely mention – personality structure, identity integration, attachment patterns, affect regulation, readiness to confront difficult material, set and setting (mindset and environment), life history, trauma history, resources, support network, expectations.

And here’s something easy to miss. Each of these variables in itself holds enormous variability. “Personality structure” isn’t a label – it’s the organisation of our whole inner world, which during a psychedelic experience can disintegrate. That’s why people with psychotic states are excluded from trials. And at the same time – in some Indigenous South American traditions, differentiation has long been practised – some states are worked with ceremonially, some aren’t. I don’t offer this as a recommendation – only as proof that distinguishing is as old as these practices themselves.

Let me show it concretely, because this is my ground. Preparing a person with a well-integrated structure and secure attachment is something entirely different from preparing a person with borderline organisation and disorganised attachment. For the first, the dissolving of the usual ego boundaries during a session can be revealing and integrating – there’s something to return to. For the second, that same dissolving can be flooding – there’s no stable core to return to after the experience, so what was meant to heal can destabilise. That doesn’t mean “off limits” – it means “differently, more slowly, with different preparation and a different safety net”. And it’s precisely this distinction that you can’t see in one averaged bar; you see it only when you look at structure, not just symptom. And beneath all of it there’s biology too – also varied.

What’s more – even what we do study doesn’t explain everything. We have data that a stronger mystical experience was linked to greater improvement. But in everyone? There are those who have a very strong experience and yet no improvement. What does it depend on? On dose? On the number of administrations? Or maybe on what we rarely deal with in psychiatric research and very often in psychotherapy – on who the person having that experience is, and what happens with them afterwards.

The measurement window: we measure early, change happens later

There’s one more dimension that changes the picture – time. We measure improvement at a very specific moment, usually after 3–6 weeks. The symptom drops quickly and fits within that window. But those are only the first weeks. Beyond the measurement window lie months and years – and we usually don’t check that.

Here I step onto my own ground. From the perspective of psychotherapy – especially work with personality structure – I know that symptomatic change and structural change move at different speeds. A symptom can subside in weeks. Structure – how someone regulates emotions, builds relationships, how their sense of self holds together – changes over months and years. And under the single word “symptom” hides a whole set of symptoms – sleep, anxiety, drive, mood – each moving at its own pace. Averaging them into a single score already loses a great deal.

And now the most important part. There are people who improve symptomatically but not structurally – and without that deeper change the symptom returns. A short measurement doesn’t see this. It sees the drop at week 6 and marks a plus.

And here I step onto ground I know from the side of statistics, not just the consulting room. Measuring the effect at a single, predetermined point in time is like judging a runner from one frame of film – you can’t tell whether you’re looking at a sprinter at peak form or a marathoner just getting going. And yet there are methods that let you model trajectories of change rather than static measurement points. In 2018, together with Ludmiła Zając-Lamparska and Monika Deja, I published a paper on latent growth curve modeling (LGCM) in Polskie Forum Psychologiczne – at the time I wasn’t thinking about psychedelics at all, they came to me two years later. But this very question of time in longitudinal data turned out to be crucial for this field. Such methods model the starting point and the rate of change separately, and most importantly – individual differences in these trajectories. We don’t ask “how much changed”, but “how does the change unfold and for whom does it unfold differently”. And individual participant data (IPD) meta-analyses would let us track these curves at the level of single people, not aggregated averages. This is the direction the field should take – and precisely the one where you can see whether our invisible variables are being captured or lost.

Four scenarios, one measurement

Let’s arrange this into a simple matrix. On the horizontal axis, what the study sees – whether the symptom dropped within the measurement window. On the vertical, what the study doesn’t see – whether structure moved.

• A good start. Symptom and structure move together, from the beginning, and the change holds. The study sees improvement – and it’s right.
• The same in the window, but without depth. The symptom dropped, structure didn’t budge. After the window the symptom returns. On the measure this story looks identical to the previous one – the study won’t tell them apart.
• The overlooked. In the window there’s no improvement – the study will say “no response”. And underneath, structure has already moved, something has opened, and the symptom catches up just behind it, only later. This person the study overlooks.
• No movement. Sometimes nothing really moves – and that too has to be recognised.

A short measurement goes wrong precisely where symptom and structure diverge – sometimes overstating, sometimes overlooking. That’s why it’s worth looking at structure, not just symptom. And if we can recognise which of these stories a given patient is in – and do it early – we can fit preparation, pace, and care to them, not to the average.

One person convinced me of this most of all. After six weeks there was no improvement, and honestly she felt worse – in the measurement she looked like a non-responder. I saw her many months later and had someone completely different in front of me. What mattered most wasn’t the experience itself or those six weeks, but what happened in the therapeutic work afterwards.

This isn’t just intuition – what we already know

At this point the usual objection appears – “it’s all nice, but soft”. Well, no. From psychotherapy and clinical psychology we know quite a lot about who benefits and when – and when they destabilise. Three threads.

Benefit has a profile

In one study, 158 psychedelic therapists were asked what they think most favours benefit. At the top were therapeutic alliance, social support, openness, the capacity to “surrender”, to let go of control, and secure attachment (Viljoen et al., 2026; cf. Viljoen et al., 2025). This is a qualitative ranking, not a table with exact values – but the direction is telling. At the top there’s no dose. There’s the relationship, readiness, and resources.

Risk has a face

Risk has a face too – and concrete numbers. In one cohort about 16% of participants were negative responders, and among people with a prior personality-disorder diagnosis – about 31%; the model indicated more than a fourfold raised risk of adverse reactions (Marrocu et al., 2024). These aren’t abstract percentages – they’re people with a less integrated sense of self and difficulties in emotion regulation, for whom a psychedelic can be destabilising, sometimes to the point of what’s been described as iatrogenic structural dissociation (Elfrink & Bergin, 2025). This isn’t an argument for fearmongering – it’s an argument for screening and preparation. For being able to recognise them earlier.

The relationship is the matrix, not the background

That’s why the relationship isn’t decoration around the substance. It’s the matrix in which all of this plays out. Set, setting, alliance, and the structure of the process are a co-active ingredient. In an analysis of the amount of therapeutic work, more preparatory therapy was linked to greater symptom reduction (Florineth et al., 2026). And one caveat that’s easy to forget. The same intensity that heals carries risk. The “peak therapy” model increases suggestibility. The relationship is then not only a healing factor, but also a safety factor.

”Psychedelic therapy” swells like a sack

Now let’s look at how we talk about this as a field. The term “psychedelic therapy” swells like a sack. We toss almost everything into it – TFP, IFS, ACT, somatic experiencing, mindfulness, holotropic breathwork, psychodynamic therapy, coaching, ceremony, “inner healing intelligence”. Each of these approaches pulls in a different direction. This is eclecticism – not integration.

Because if we lay these approaches out on an axis, we see a spectrum. At one end, pure presence, support, ceremony, harm reduction, integration. At the other, structural psychotherapy – trauma work, psychodynamic approaches, schema therapy. And here I’ll allow myself a personal reflection, because I don’t have a ready answer for it. What is psychedelic therapy actually meant to be? Psychotherapy assisted by psychedelics? Or is integration plus psychotherapy enough? This multiplicity delights me as much as it unsettles me. It delights because there’s plenty to choose from. It unsettles me because how do we know it should be one way and not another, if we don’t have a conceptualisation – an understanding of what we’re doing? How do we train for it – with a weekend of each method? Is that enough when we’re working with people who are ill?

Hence three things I want to say plainly.

1. There is no single “psychedelic therapy”. There are many very different practices under one banner.
2. Integration is not therapy. Accompanying someone in making sense of things after a session is different from psychotherapy – and the two are constantly confused.
3. And the most important question. Who decides whether someone needs psychotherapy or whether support is enough – and who has the competence to judge it?

In my view the real debate about competence hasn’t even begun.

Not another drug – a different paradigm

And here is the core of my unease. Because this isn’t another drug. It’s a different way of treating. In the drug model – a pill, a repeatable dose, effect after weeks, a short follow-up, the therapist off to the side. In the psychedelic model – one administration and six to eight hours of presence, preparation and integration as the core, a team, set and setting, the relationship. It’s a different organisation of the whole treatment.

You can put it more precisely. Psychedelic therapy is, as Muthukumaraswamy et al. (2025) argue, a complex intervention in which substance, psychotherapy, setting, and the patient’s expectations are inseparably intertwined. And the randomised-trial paradigm was designed to test the isolated action of a substance – you separate the drug from context and see what the drug alone does. In psychedelic therapy that separation is structurally impossible, and not because of researchers’ incompetence, but because the very object of study is indivisible. There’s even a tool (PRECIS-2) that measures how “laboratory-like” a study is versus how “real-world” it is – and psychedelic trials sit close to the laboratory pole – a room with dimmed light, music, two therapists, a session lasting hours. This has little to do with how this treatment would look in an average consulting room. In other words, we study it under conditions we won’t reproduce in reality – and yet from those conditions we draw conclusions about reality.

It’s worth drawing one distinction. In a trial we have everything – time, a team, mentoring, supervision, a protocol for preparation and integration – and carefully selected patients. In the consulting room we don’t have all of that – time pressure, often a single therapist, no ready framework, an unselected, real patient. Hence the honest questions that have to be asked before this spreads. Is this really the direction we’re training toward? How will we learn this practice? Will we require mandatory placements? It’s the same responsibility as in any other psychiatric and psychotherapeutic treatment – and we shouldn’t lose it in an otherwise fair attempt to honour cultural diversity and the roots of these practices.

There’s one more risk that’s rarely mentioned. When the substance reaches the market, the pressure of time and cost may push out of the equation exactly what’s most expensive and most difficult, that is the relationship and preparation. What’s left is the administration alone. That would be therapy without therapy, which is precisely what I’m warning against.

Eight hours

Let me say this through one story, because it conveys it better than any slide.

The first person I accompanied through an administration was a widow in her seventies. I was already a relatively experienced therapist, regularly using supervision. And even so, little of my prior experience could have prepared me for eight hours of presence with a person in her situation. That experience was entirely different from everything I knew from the consulting room – a different register of closeness, a different responsibility, a different kind of silence.

This is precisely the clinical space I keep writing about – the one you can’t see in an abstract or on a group-effect bar. And the one a weekend training can’t make up for.

Let’s change the question

That’s why I propose changing the question. “Do psychedelics work?” is like a broken record. The right question is – for whom? when? under what conditions? Not “what is the average”, but “what might the truth be at the level of this one person sitting across from me”.

What we really need

At conferences I often hear – “we need more studies to settle this”. I agree – we need a great many more. But we already have enough to say one thing. Psychedelics in psychiatric care are a direction worth developing in practice. So I won’t end with a call for more studies – but for more practitioners thinking about psychedelics in a clinical context.

And I’ll say it plainly. I’m against psychotherapists specialising in psychedelic work exclusively. For me it’s a bit like improvisation – to improvise well, you first have to be able to read music.

Hence a concrete idea. Instead of training a narrow group of “psychedelic therapists”, it’s worth building a broader network of specialists who can prepare a patient over a longer time. Alliance, openness, or the capacity to “surrender” can’t be built in three or five meetings before the session. It’s the work of months, closer to ordinary psychotherapy than to a weekend course.

Two caveats, so there’s no misunderstanding. First, I’m talking about clinical populations. Psychedelics can serve various ends – coaching is one thing, a personal-growth workshop another, psychotherapy yet another. But if we mean treating people struggling with psychiatric states, it takes nuance and the ability to see many variables – including those not obvious at first glance. Second, I’m not fearmongering and I’m not stripping these experiences of their beauty. I’m calling for a mature conversation – because it’s “just around the corner”, and the need for reflection is already real and will keep growing.

So I return to the two pillars – data and stories. Between them is an integration we still lack. In the consulting room I see how hard this is; in abstracts it seems simpler. The real challenge isn’t choosing one of these perspectives – it’s joining them. I don’t have all the answers. But I believe this knowledge will arise not only in research – also in consulting rooms. Just as psychotherapy develops not only in science, and perhaps above all outside it. And the foundations are three – alliance, preparation, understanding structure.

In closing – the galaxy returns

At the end our main image returns – that galaxy of variables. It is the true background of every “score”. The number on a slide isn’t the end of the story – it’s its beginning. If this text leaves you with one question instead of a ready answer – “for whom, when, and under what conditions?” – then it has done its job.

If the topic interests you, I’d be glad to talk – in the comments, by email, at the next conference. Because we’ll build this knowledge together or not at all.

References

Carhart-Harris, R. L., Giribaldi, B., Watts, R., et al. (2021). Trial of psilocybin versus escitalopram for depression. New England Journal of Medicine, 384(15), 1402–1411. https://doi.org/10.1056/NEJMoa2032994

Chmiel, J., & Rybakowski, F. (2026). Predictors of the effectiveness of psychedelics in treating depression: A scoping review. International Journal of Molecular Sciences, 27(5), 2202. https://doi.org/10.3390/ijms27052202

Compass Pathways. (2025, 2026). COMP005 and COMP006: Announcements of meeting the primary endpoint in phase 3 trials [press releases]. https://ir.compasspathways.com/

Elfrink, S., & Bergin, L. (2025). Psychedelic iatrogenic structural dissociation: An exploratory hypothesis on dissociative risks in psychedelic use. Frontiers in Psychology, 16, 1528253. https://doi.org/10.3389/fpsyg.2025.1528253

Erritzoe, D., Barba, T., Greenway, K. T., et al. (2024). Effect of psilocybin versus escitalopram on depression symptom severity in patients with moderate-to-severe major depressive disorder: Observational 6-month follow-up of a phase 2, double-blind, randomised, controlled trial. eClinicalMedicine, 76, 102799. https://doi.org/10.1016/j.eclinm.2024.102799

Florineth, G. A., Klima, I., Boeker, A. L., et al. (2026). Psychological therapy quantity and depressive symptom reduction in psychedelic-assisted therapy: A systematic review and meta-analysis. JAMA Network Open, 9(1), e2554843. https://doi.org/10.1001/jamanetworkopen.2025.54843

Functional unblinding in pivotal studies and the future of psychedelic medicine. (2025). Journal of Clinical Psychiatry. https://www.psychiatrist.com/jcp/functional-unblinding-in-pivotal-studies-and-the-future-of-psychedelic-medicine/

Goodwin, G. M., Aaronson, S. T., Alvarez, O., et al. (2022). Single-dose psilocybin for a treatment-resistant episode of major depression. New England Journal of Medicine, 387(18), 1637–1648. https://doi.org/10.1056/NEJMoa2206443

Kinahan, S., & Wilson, E. (2025). Perspectives of psychotherapists regarding psychedelic-assisted therapy. Counselling and Psychotherapy Research, 25(1), e12881. https://doi.org/10.1002/capr.12881

Marrocu, A., Kettner, H., Weiss, B., et al. (2024). Psychiatric risks for worsened mental health after psychedelic use. Journal of Psychopharmacology, 38(3), 225–235. https://doi.org/10.1177/02698811241232548

Muthukumaraswamy, S. D., Baggott, M. J., Schenberg, E. E., et al. (2025). Psychedelic-assisted therapy as a complex intervention: Implications for clinical trial design. Therapeutic Advances in Psychopharmacology, 15, 20451253251381074. https://doi.org/10.1177/20451253251381074

Royal College of Psychiatrists. (2025). Pharmacologically/psychedelic-assisted psychotherapy: Research guidance. https://www.rcpsych.ac.uk/docs/default-source/improving-care/better-mh-policy/position-statements/pharmacologically-assisted-psychotherapy-research-guidance.pdf

Viljoen, G., Bendau, A., Walter, H., et al. (2026). Therapist-rated predictors of response to psychedelic-assisted therapy. Nature Mental Health, 4(6), 951–961. https://doi.org/10.1038/s44220-026-00642-4

Viljoen, G., Walter, H., Bendau, A., et al. (2025). Predictors of therapeutic response to psychedelic-assisted therapy: A systematic review. Journal of Psychopharmacology, 40(5), 703–719. https://doi.org/10.1177/02698811251389581

Zając-Lamparska, L., Warchoł, Ł., & Deja, M. (2018). Modelowanie latentnych krzywych rozwojowych jako metoda analizy danych podłużnych w psychologii [Latent growth curve modelling as a method of longitudinal data analysis in psychology]. Polskie Forum Psychologiczne, 23(2), 342–362.

Zamaria, J. A., Fernandes-Osterhold, G., Shedler, J., et al. (2025). Psychedelics assisting therapy, or therapy assisting psychedelics? The importance of psychotherapy in psychedelic-assisted therapy. Frontiers in Psychology, 16, 1505894. https://doi.org/10.3389/fpsyg.2025.1505894