Why Durability Drives the Economics

One of the most attractive stories about psychedelic therapy is also one of the easiest to overstate. A person receives one dose, has a meaningful experience, and is then described as better months or years later. Compared with medication taken every day (such as an antidepressant) or treatment delivered during regular clinic visits (such as maintenance Spravato), the contrast is powerful. It can start to sound like one treatment followed by a permanent recovery.

Ian Roullier, who participated in psilocybin trials and later co-founded the Psychedelic Participant Advocacy Network, has described the risk plainly: “one dose and you're fixed forever.” As he immediately added, “It sets expectations too high.” It is a compelling story. It is also too simple.

That is not what durable means. It certainly is not what the data allow us to promise.

Some participants improve substantially and remain well at later follow-ups. Some improve and then relapse. Some receive additional medication, psychotherapy, or another psychedelic treatment. Others do not respond in the first place. Even a study that reports a positive average result can contain all of those different paths.

I was reminded of the human side of this at the Interdisciplinary Conference on Psychedelic Research. In a patient-led panel I later wrote about for Lucid News, participants described what happened when the structure of a trial ended and ordinary life resumed. Leonie Schneider, another trial participant and PsyPAN co-founder, has said that after a psychedelic trial there can be “no community, or a place where you can go, to land.” That does not tell us whether a treatment effect was durable. It does show why the end of a protocol should not be confused with the end of a person's care.

This distinction is clinically important, but it is also central to the economics. Psychedelic therapy may require screening, preparation, a long administration session, trained staff, a suitable room, and follow-up care. Most of that cost arrives near the beginning. The health benefit, by contrast, accumulates only for as long as the person remains better than they would have been with the alternative. One week, twelve weeks, one year, and three years are not minor variations on the same model. If the health difference stays constant, three years produces 156 times as much health gain as one week.

That makes durability both the largest question mark and often the largest unresolved lever in the value story. Health economist Elliot Marseille captures the attraction as the potential for “durable responses after limited treatment courses”. Longer benefit is usually better for patients, health systems, and society. A manufacturer may earn revenue when treatment is repeated, so its incentives are not identical. Yet developers still need a schedule that patients can accept, clinics can deliver, and payers can justify. A product that must be given too frequently loses much of the practical advantage it is promising over existing care.

In other words, the entire economic case does not rest on a treatment working forever. It does rest on being clear about how long it works, for whom, and what happens next.

In the previous article in this series, we gave a fictional psychedelic pathway 0.5 additional quality-adjusted life years (QALYs). We then used that health gain to calculate the incremental cost-effectiveness ratio, or ICER.

The arithmetic was deliberately simple, but it quietly contained an assumption about time. A treatment does not produce half a QALY at the moment it is delivered. The gain accumulates because someone spends time in better health than they otherwise would have done.

Imagine that a person's health-related quality of life is 0.30 higher after the new treatment than it would have been with current care. If that difference lasts for one year, the treatment produces 0.30 additional QALYs. If it lasts for six months, it produces 0.15. If it lasts for only three months, it produces 0.075.

The treatment episode still costs the same in each scenario. Only the duration changes. Yet the cost per QALY changes dramatically.

That is the basic economic mechanism. Cost is concentrated at the start. Health gains (and any savings from avoided care) accumulate over time. When the additional benefit stops, the accumulation stops too.

Let us make that mechanism visible with one intentionally invented example. Assume that the new psychedelic pathway costs €10,000 more than current care. While its benefit lasts, health-related quality of life is 0.30 higher. For now, we will ignore savings elsewhere in the health system and assume that treatment is not repeated.

The four fixed scenarios look like this:

At one week, the €10,000 is spread across a very small health gain. At three years, the upfront cost has not changed, but the accumulated health gain is 156 times larger. The resulting cost per QALY is therefore 156 times lower. Nothing about the treatment day changed. The entire difference came from the duration we asked the model to assume.

This is not an estimate for any real psychedelic treatment. It is a way of isolating the duration assumption. A real evaluation would also consider the comparator, treatment response, relapse, additional care, adverse events, productivity, discounting, and the possibility of retreatment. Turning all of those on at once would make the explanation more realistic but less useful. Here, the point is to see how much work one assumption can do.

There is another important lesson in the numbers. A benefit does not need to last forever to matter. (In our fictional example, one year of benefit produces an ICER of about €33,000 per QALY. That sits near commonly used European decision ranges, although no threshold guarantees coverage.) For someone with severe illness, frequent crisis care, high health-care use, or no good remaining treatment, a meaningful six-month improvement may have considerable value. If the same pathway helps a broader group with lower unmet need or fewer costs that can be avoided, it may need to produce a larger or longer health gain to make the same economic case.

Cost-effectiveness is therefore not a cure test. It asks how much additional health a pathway produces, at what additional cost, compared with what would otherwise happen.

Cancer care makes the principle easier to see. A treatment can be valuable because it extends life, improves its quality, or gives someone more time before their illness progresses. It does not need to cure the cancer. At the same time, the comparison shows why we should not relax the evidence question. A 2024 study of cancer drugs granted accelerated approval in the United States found that 41% of the indications with more than five years of follow-up had not subsequently demonstrated an improvement in overall survival or quality of life. That does not mean a modest benefit is worthless. It means that a plausible benefit and a demonstrated benefit are not the same thing.

In principle, mental health should not sit on a lower rung. NICE begins from the position that QALYs receive equal weight, whether the gain comes from living longer or living better (a QALY makes that comparison visible). In practice, mental health has long been underfunded. The WHO Mental Health Atlas 2024 reports that mental health still accounts for only about 2% of health budgets globally. Better evidence on durability will not solve that imbalance by itself. It can make the health gained through psychiatric treatment harder to overlook, while holding psychedelic therapy to the same basic question as oncology and every other field: how much better are people, compared with the alternative, and for how long?

Before turning to the psychedelic literature, it helps to slow down over the word itself. Researchers can use durability to describe several related but different results.

The first is the average symptom trajectory. A group may still have a lower average depression or PTSD score at six or twelve months. That tells us something useful about the group, but it does not mean that every participant maintained the same response.

The second is continued response or remission. Researchers may report the proportion of participants who remain above a response threshold or below a remission threshold. The denominator matters here. “Most responders remained well” says something different from “most people who received treatment remained well.”

The third is time until an event, such as relapse, a new depressive episode, rescue medication, or loss of response. This can be closer to the question an economic model needs to answer, but studies do not all define these events in the same way.

The fourth is time until retreatment. A person may improve, deteriorate later, and then receive another dose or treatment course. That is not simply a failed durable response. It is a different care pathway with another possible benefit and another set of costs.

These measures cannot be collapsed into one universal number of “months durable.” A whole-group average can hide people who remain well and people who relapse. A responder analysis can leave out everyone who did not initially respond. A late follow-up can show where participants are at that point without proving that the original treatment caused everything observed in between.

A clinical trial and a health-economic model often run on different clocks.

A trial may have a primary endpoint after three or six weeks because that is when it is designed to test whether the treatment works. Researchers may then follow participants for six months or a year. A health-technology assessment may need to estimate costs and outcomes over several years or even a lifetime because relevant effects do not stop when data collection does.

That creates an unavoidable gap. NICE's methods guidance asks analysts to choose a time horizon long enough to capture the important differences in costs and outcomes. When the evidence ends earlier, they have to extrapolate. They might assume that the effect stops, gradually wanes, persists while treatment continues, or lasts beyond the observed period. Those choices should be explained and tested rather than hidden.

The United States does not have one national HTA body equivalent to NICE, but American value assessments still confront the same question. The Institute for Clinical and Economic Review (the organisation often called ICER, rather than the ratio we explained earlier) used a lifetime horizon in its 2024 exploratory model of MDMA-assisted therapy for PTSD, included the possibility of retreatment within five years, and then tested shorter horizons. Over a lifetime, the model described the treatment as less costly and more effective. At three years, the result was approximately $157,000 per QALY; at five years, approximately $81,000. The clinical inputs had not changed. The amount of future benefit the model was allowed to count had.

ICER is not itself an insurer, and its MDMA analysis was exploratory because it judged the underlying evidence too uncertain for a definitive conclusion. Still, the example answers the practical question. U.S. payers may not use one national threshold, but durability, retreatment, and the time horizon remain central to the value case placed in front of them.

A twelve-month follow-up therefore means that researchers looked at outcomes through twelve months. It does not automatically establish a twelve-month causal treatment effect. Participants may use antidepressants, start psychotherapy, receive ketamine, change their work or relationships, or seek another psychedelic experience. Some may leave the study. Blinding may no longer be possible. The comparison group may no longer receive clearly different care.

Longer follow-up is still valuable. It gives us more information than stopping at week six. We simply need to ask what kind of information it provides.

The evidence in 2026 is no longer limited to a handful of very short studies. Large depression trials are now collecting outcomes well beyond the primary endpoint, several published studies report results after six or twelve months, and a few small cohorts have been followed for years. The strength of the conclusion does not increase automatically with the length of the follow-up, however.

Reacting to the 2022 Phase IIb psilocybin trial, King's College London psychopharmacologist Anthony Cleare noted that “The effects did start to wear off by three months”. He pointed to psychological therapy or periodic repeat treatment as possibilities for preventing the return of depression. The comment is not a universal estimate for psilocybin. It is a reminder that recurrence was visible even in one of the field's most prominent trials.

The largest current example comes from the COMP360 programme in treatment-resistant depression. In July 2026, COMPASS Pathways reported 26-week results from COMP006, a Phase III trial with nearly 600 participants. The company reported that the randomized groups remained separated through week 26 and that some participants who later received retreatment entered remission. The COMP005 and COMP006 protocols also continue observation beyond the early endpoint and allow retreatment under defined conditions.

This is exactly the kind of middle-distance evidence an economic model needs. It is also important to describe it accurately. The 26-week findings are currently company-reported, the programme includes blinded and open-label periods, and a statement about maintaining benefit among early responders is not a statement about every treated participant. The trials will tell us much more than a six-week endpoint, but they cannot by themselves settle what happens during routine care over many years.

Published studies add different pieces. The COMP004 observational follow-up followed a subset of participants from an earlier psilocybin trial for up to a year. In the parent cohort, the median time to a new depressive event was measured in months rather than years, and most events occurred by week twelve. That is useful because it makes recurrence visible. It also resists the easy story that one dose creates the same lasting result for everyone.

The EPIsoDE long-term follow-up found improvements at six and twelve months after psilocybin-assisted psychotherapy for treatment-resistant depression. Yet the controlled phase had ended much earlier, participants could use other treatments, and the groups were no longer clearly different at later follow-up. The later observations are encouraging, but they are not equivalent to a controlled estimate that can simply be extended across a model.

Small studies provide some of the most memorable long-term findings. In a psilocybin-facilitated smoking cessation pilot, ten of fifteen participants were biologically confirmed as abstinent after twelve months and nine remained abstinent at a follow-up averaging about thirty months. A cancer-distress follow-up reported improvements among fifteen surviving participants more than three years after treatment. Longer-term MDMA-assisted psychotherapy studies have also reported sustained improvement in PTSD symptoms for many participants.

Taken together, these outcomes demonstrate that lasting improvement is possible and give researchers hypotheses worth testing in larger studies. Their small, selected samples, additional care, missing participants, and uncontrolled later periods mean that they cannot supply a single routine-care durability parameter. The emotional force of a person remaining well for years and the methodological caution needed to model that result can both be true.

The one-and-done story makes retreatment sound like an exception or a disappointment. It may turn out to be a normal part of care. Psychiatrist David Feifel now considers “the concept of one and done ... pretty much dead”, although he also stresses that the number and interval of repeat administrations remain unknown.

In an economic model, retreatment changes both sides of the calculation. Another treatment may restore health gains or prevent further deterioration. It also adds another medicine dose, more clinician and room time, follow-up care, patient time, and possible adverse events. The relevant question becomes less “Did one dose last?” and more “What treatment pathway keeps people well, how often is it needed, and what does that pathway cost?”

Ketamine offers a useful contrast. Its antidepressant effect can arrive rapidly but often fades, which is why repeated or maintenance treatment has been studied. A systematic review of maintenance ketamine found promising but methodologically limited evidence. Economically, maintenance dosing is not just the original response stretched into the future. Each return visit brings possible benefit and another use of clinical resources.

Developers are already thinking in these terms. In the Wall Street Journal video “The Race to Commercialize the World's Most Powerful Psychedelic” , the CEO of ataiBeckley contrasted a possible 5-MeO-DMT pathway with weekly Spravato visits: “you get discharged, but instead of coming in every week, you're coming in every two months, three months.” That is not evidence that this schedule will work in routine care. It is a clear statement of the proposed economic advantage: treatment may recur, but much less frequently than an existing clinic-based alternative.

The current COMP360 trials are beginning to show how retreatment might be studied prospectively. Participants can be observed for loss of benefit and receive another dose according to the protocol. That can help answer who returns, how long they went before returning, and whether they respond again. Routine care will still be different. Eligibility rules, appointment capacity, patient preference, clinician judgement, reimbursement, and access to other treatment will all influence what happens.

In the coaching work I have done, I have seen people return after some time. Their first experience was not necessarily a failure. Life had continued, circumstances had changed, or there was more work they wanted to do. That makes episodic care feel more intuitive to me than the idea of one final intervention. It is still personal experience, not a basis for setting a reimbursed retreatment schedule for depression or PTSD.

The uncertainty is not mysterious. Researchers and health systems can collect the information that economic models currently have to assume.

They can measure quality of life alongside symptom scores, because a change in a depression scale does not directly tell us how many QALYs were gained. They can record medication, psychotherapy, hospital care, crisis care, work absence, informal care, and other resource use during follow-up. They can report how many participants did not respond, how many remained well, how many relapsed, and how many were retreated. They can make clear what happened after retreatment rather than counting only the new cost.

They can also preserve the timeline. A result at twelve months is more useful when we know whether the participant remained well throughout, relapsed at month four and recovered after other care, or returned for another psychedelic treatment at month nine. The same endpoint can hide very different clinical experiences and very different economic pathways.

Real-world evidence will eventually matter as much as longer trials. Trials have motivated participants, specialist teams, carefully specified sessions, and more support than many clinics will be able to reproduce. Routine data can show whether benefits and retreatment patterns survive when care is delivered across ordinary services, to a broader group of patients, with real capacity constraints.

None of this requires us to dismiss the long-term signals already available. It asks us to use them for what they can genuinely tell us.