Psychedelic drug development has reached the stage where its evidence will be read by regulators, payers and clinics, not only by the researchers its early results excited. Compass Pathways announced on November 9 that it had completed its new drug application for COMP360 psilocybin in treatment-resistant depression. If FDA accepts it, those readers will ask questions that small academic studies were never built to answer: what dose, how often, for whom, measured how, and delivered by whom.
Tanya Ramey has worked on those questions from three sides. At Pfizer she worked in conventional CNS drug development, including new antidepressants that combined serotonergic mechanisms, new antipsychotics and anxiolytics. She then spent about a decade as a medical officer and program director in the National Institute on Drug Abuse’s Division of Therapeutics and Medical Consequences, overseeing medication-development research for addiction. Until mid-2026 she was chief medical officer of the Usona Institute, the nonprofit studying psilocybin as a potential treatment for major depressive disorder, where she describes leading clinical and regulatory strategy through Phase 3 and preparation for a New Drug Application. She is now co-founder and chief medical officer of Telesti Therapeutics, a preclinical-stage company developing neuroplastogens.
She draws one lesson from all three.
“Each of these settings has the same blind spot: the evidence that makes a molecule exciting is usually not the evidence that gets it approved,” she told Behavioral Wire.
What follows is her account of what that means in practice, set against the trial record. In places the record supports her strongly. In others it is unsettled, and in a few it pushes back.
Gates, not momentum
Conventional CNS programs, in Ramey’s account, are built as a chain. The strongest programs, she said, “connect exposure, measurable target engagement if possible, pharmacodynamic effects, and clinical outcomes.” She added: “Each study should answer a defined question and inform the next development decision.”
Psychedelics arrived by a different route. The field, she said, drew first on a long history of traditional and folk use and only later merged that knowledge with rational drug development. “There wasn’t a clear, steady progression of expected studies to move the candidate through the ‘gates’ of decision-making,” she said. “Not surprisingly, some mechanistic questions have not been answered to this day,” even though, she noted, several psychedelics are in Phase 3 or have submitted applications to FDA.
Her prescription is to agree early, for classic psychedelics and for the newer neuroplastogens, on what would count as proof of viability, proof of mechanism and proof of clinical benefit, and on which uncertainties have to be resolved before money is spent on the next stage.
Ramey would start from the target product profile, developed first: the indication, the dosing paradigm, the monitoring requirement, the likely REMS and the payer argument. Each trial is then built to produce evidence for one of those claims. For a new chemical entity, she adds, the differentiation hypothesis has to be testable early and monitored; one example is a shorter acute duration that changes clinic economics.
She is equally specific about a practice she considers counterproductive: collecting safety data unevenly while failing to characterize the acute psychedelic experience itself. Protocols, she argues, need field-wide rules for deciding which perceptual, emotional and cognitive changes count as adverse events and which are expressions of the drug’s mechanism.
“A patient-reported benefit may not automatically be classified as an adverse event simply because it represents a change from baseline,” she said. “Equally, an effect should not be excluded from safety reporting simply because it is expected, reflects the drug’s pharmacology, or is experienced as meaningful by the patient. Those characteristics can coexist.”
That is not an abstract concern. When FDA declined to approve Lykos Therapeutics’ MDMA-assisted therapy in August 2024, the first deficiency in its complete response letter was that the company “did not collect important information on events that the participant, therapist, or study physician considered ‘positive’ or ‘favorable,’” and that inspections had found unreported adverse events at two or more sites. FDA had earlier advised that abuse-related events be recorded even when they seemed desirable, naming euphoria as an example. (Lykos now operates as Resilient Pharmaceuticals.)
Phase 2 should change a decision
Ramey’s case for Phase 2A rests on what it is for. “Phase 2A is fundamentally a learning phase in drug development,” she said. “Its value does not depend on directly supporting a labeling claim; it depends on whether it reduces uncertainty about the next development decision.”
That frees a sponsor to organize an early study around a measurable symptom domain, such as anhedonia, not a DSM diagnosis alone. “We shouldn’t assume that a DSM diagnosis is the best starting point,” she said. A domain can cut across several diagnoses, while patients with the same diagnosis can differ widely on it. Her approach is to characterize patients more precisely and record the DSM diagnosis alongside.
On who pays, sponsors should fund exploratory studies that improve decisions about their own drug, she said; public agencies, foundations and academic partners are the right funders for methods the whole field can use, such as new designs and biomarkers.
For a board, the test is the decision the study enables. Before funding, Ramey wants a credible biological rationale, reliable measures, a defined population, and prespecified criteria for advancing, redirecting or stopping. After it, she looks for an efficacy signal of clinically meaningful size and a dose and regimen that are practical and commercially viable.
She also wants a comparator arm in Phase 2, even one not powered for a formal comparison, and says many companies avoid one because it can raise questions at a stage when a compound is most vulnerable. Psychedelic trials, she notes, often carry a built-in option: a low dose of the same drug, used as a pseudoplacebo to protect the blind.
She ties that to how the field’s first results were produced. Early academic psilocybin studies, she said, reported very large effects in small, self-referred, often psychedelic-experienced samples run by enthusiastic investigators, and the effects fell substantially once trials became multisite, randomized and controlled. That is normal in CNS development, she said, but enthusiasm made sponsors slow to plan for it.
“Psychedelics can fail because the evidence package was built by enthusiasts and for enthusiasts, rather than reviewers.”
— Tanya Ramey
Aticaprant: a failed molecule, or a failed test?
The hardest case for domain-based development is aticaprant, a kappa opioid receptor antagonist that Johnson & Johnson took into Phase 3 for depression with anhedonia and then dropped. Ramey reads the sequence differently from the obvious conclusion.
The record begins with FAST-MAS, a proof-of-mechanism study published by Krystal and colleagues in Nature Medicine in 2020 and funded through a National Institute of Mental Health contract. It randomized 89 people with a score of at least 20 on the Snaith-Hamilton Pleasure Scale (SHAPS), an anhedonia measure, across several diagnoses: depression, bipolar depression, anxiety disorders and PTSD. Antidepressants were excluded. The primary endpoint was a brain measure, ventral striatal activation on fMRI while participants anticipated a monetary reward, and the drug met it. SHAPS scores, a secondary endpoint, also improved. General depression and anxiety ratings did not.
The proof-of-concept study that followed, Schmidt and colleagues in Neuropsychopharmacology in 2024, asked a different question. Aticaprant was added to an existing SSRI or SNRI in people with major depressive disorder who had responded inadequately to one or two antidepressants. The primary endpoint became the clinician-rated MADRS depression scale at week 6. The drug improved MADRS by 2.1 points more than placebo, an effect size of 0.23, with a one-sided p value of 0.044 against a prespecified one-sided threshold of 0.20, a lenient bar set in advance. It would not have met the conventional two-sided 0.05 standard. On SHAPS, the anhedonia measure that had moved in FAST-MAS, there was no signal. The effect looked larger in participants with higher anhedonia, a finding the paper presents descriptively, not as a tested hypothesis.
Johnson & Johnson then ran two Phase 3 trials in depression with moderate-to-severe anhedonia, again as add-on treatment and again with total MADRS score as the primary endpoint. On March 6, 2025, it discontinued the program “due to insufficient efficacy in the target patient population.” Results posted on ClinicalTrials.gov show MADRS differences of 0.9 points in VENTURA-1 (p=0.467) and 0.5 points in VENTURA-2 (p=0.670). No peer-reviewed report of the Phase 3 results appears in PubMed, and nothing public describes what FDA advised at the end of Phase 2.
That is the record. What follows is Ramey’s reading of it.
“In my view, the molecule itself did not fail,” she said. “Rather, the development strategy may have failed.” Her reasoning: the drug produced a signal when its mechanism was tested in antidepressant-free patients selected for anhedonia, then was moved into people already on SSRIs or SNRIs who had responded poorly to them, many of whom, she argues, likely had the emotional blunting those drugs can cause. Sites may have struggled to tell that blunting from anhedonia. And the Phase 2 result, in her reading, did not support expecting an additive effect in Phase 3. If the company wanted to test aticaprant in that population, she says, a Phase 2b designed for it could have answered the question with less risk and at lower cost. She also suggests that non-scientific considerations may have shaped the development path. She offers that as interpretation and names no such consideration; the trial registries, the Phase 2 paper and Johnson & Johnson’s announcement do not say what drove the decision.
The record supports part of her argument. The study that moved the anhedonia measure and the studies that failed tested different populations, different treatment settings and different endpoints, so the Phase 3 failure does not test the original mechanism finding directly. The VENTURA-2 protocol also sized the trial for an effect size of 0.4 at a one-sided alpha of 0.025, above the 0.23 that Phase 2 had shown on its primary endpoint.
Other evidence cuts against a clean exoneration. FAST-MAS was itself a small, eight-week study whose general depression and anxiety ratings did not improve, even as its anhedonia measure did. Neumora’s navacaprant, another kappa opioid antagonist, also failed in Phase 3, and Neumora halted that program in June 2026 after three trials missed their primary endpoints. The Phase 2 authors wrote that the placebo lead-in in their study did not reduce placebo response, and a published commentary asks whether kappa antagonists might act on placebo response itself. Johnson & Johnson has not abandoned aticaprant: a Phase 1b trial in schizophrenia, registered with a March 2026 start date, is recruiting.
Ramey’s narrower claim holds up better. The Phase 3 trials selected patients for anhedonia but measured an add-on effect on total depression scores in people already taking SSRIs or SNRIs, so their failure is weak evidence against a domain-based approach. What the record cannot show is that the molecule would have succeeded in a trial built around the domain.
She sees the larger obstacles to domain-based development as commercial and regulatory, not scientific. Reimbursement, marketing and regulatory pathways are organized around DSM diagnoses, and payers and commercial teams, she said, may see domain-based indications as fragmenting the market.
Measuring the patient, not only the diagnosis
Ramey’s interest in domains comes from addiction research, where diagnosis is unusually loose. DSM-5 substance use disorders are diagnosed by any two or more of 11 criteria, which allows more than 1,000 possible combinations of symptoms, and any of them can qualify a person for a trial. Trial populations defined this way are, in her account, extremely heterogeneous next to those for schizophrenia or major depression.
The NIDA Phenotyping Assessment Battery, on which she is a co-author, was built to characterize participants in substance use disorder trials. The battery’s purpose, in Ramey’s account, is to identify domains that can enrich a trial population and reduce the heterogeneity that buries a signal.
Turning such tools into regulatory evidence is slower. She points to recent work proposing well-being as a psychedelic trial endpoint, described this August in npj Mental Health Research as a candidate co-primary outcome, and calls it a multiyear undertaking. A measure would need evidence of construct validity, content and criterion-related validity, particularly predictive validity, and a sponsor could then choose to seek qualification through FDA’s Clinical Outcome Assessment Qualification Program. A qualified measure, she said, could be used with much greater confidence of regulatory acceptance. No psychedelic well-being measure has been qualified.
“In terms of methodological problems, I see the sensitivity to change as the most important issue,” she said. She also sees EEG measures of brain entropy as a possible future marker, especially if paired with a self-report scale. None is accepted by FDA as a trial endpoint.
Measurement also covers how outcomes are scored. Ramey argues that binary endpoints, such as counting patients who cross a responder threshold, discard information that continuous scores keep, and she co-chairs an ISCTM working group that has submitted a manuscript making that case for substance use disorder endpoints. She points to EPIsoDE, a German trial in treatment-resistant depression published in JAMA Psychiatry this March. It missed its primary endpoint, the share of patients with at least a 50% reduction on the HAMD-17 scale at week 6 (17.0% on psilocybin 25 mg versus 10.6% on nicotinamide, p=0.19). On continuous HAMD-17 change, psilocybin 25 mg improved scores by 4.6 points more than nicotinamide. Ramey calls it the clearest example that binary primary endpoints are a poor fit for CNS trials. The authors describe the continuous result as exploratory, because it was tested only after the primary endpoint failed, and call the trial inconclusive.
What one dose can and cannot carry
Ramey’s sharpest claim concerns durability. “The ‘one dose, lasting benefit’ narrative for psilocybin does not survive contact with the Phase 3 data,” she said. In treatment-resistant depression, she expects psilocybin to be used most often as a chronic-intermittent therapy, “and payers will price, authorize and contract for it on that basis whether or not the label says so.”
Ramey was until mid-2026 chief medical officer of Usona, which is developing its own psilocybin treatment for major depressive disorder. Her argument here concerns Compass’s public Phase 3 results, not Usona’s.
Compass’s own disclosures support the retreatment half of that argument, with a limit on what the figures show. In COMP005, which gave a single 25 mg dose in its blinded primary period, 70% of the 25 mg arm and 53% of the placebo arm received a second administration after week 6, which under the protocol could be another blinded dose or an antidepressant. That redosing was permitted in both arms, so it shows how the trial was run, not how many patients lost benefit. Of the 25 mg participants who entered the open-label year after week 26, about 80% received another dose; the rules allowed one only after relapse or for those who had not remitted. In COMP006, which gave two fixed doses three weeks apart, 58% of the 25 mg arm were retreated after week 9.
On response, the record is more mixed than her formulation. The placebo-adjusted difference in COMP005 narrowed from 7.2 points at week 1 to 3.6 at week 6 but stayed significant at the primary endpoint. In post hoc analyses, 25% of the COMP005 25 mg arm and 39% of the COMP006 arm had reached at least a 25% MADRS reduction by week 6; Compass reports that group stayed improved on average through week 26, using observed, unadjusted data. Compass calls the profile rapid and durable.
Read together, the data say less about whether benefit lasts than about what the product will be. Ramey’s question is what regimen a sponsor is actually developing if benefit fades for many patients and retreatment becomes part of ordinary use. She sketches three possible models: episodic, with one course per depressive episode, which she says may suit the subgroup of stable responders; chronic-intermittent, at two to four doses a year; and chronic scheduled maintenance, which she considers neither supported nor feasible for psilocybin. These are her proposals, not established practice.
For payers, she said, “frequency drives cost.” Each dose carries a monitored session, billed through Category III CPT codes 0820T to 0822T that took effect in January 2024; Medicare lists them as contractor-priced, with no national payment amount. That sits on the medical benefit, alongside the drug on the pharmacy benefit. She expects retreatment criteria, frequency caps and the definition of response to be where coverage is won or lost, and argues that the sponsor’s job, ideally before launch, is to turn retreatment “from an observed behavior into a defined regimen,” with triggers payers can use, survival-style durability data and pricing per course.
She would move that work much earlier. Depression is chronic and relapsing, and regulators reasonably want to know what happens at six months and beyond. Exposure requirements written for daily drugs do not map onto intermittent dosing. Durability, relapse and retreatment criteria, she argues, should be designed in from Phase 2, not left to long open-label extensions that are hard to run.
The timing of any approval is itself unclear. Compass holds one of FDA’s Commissioner’s National Priority Vouchers, which promised review in one to two months. Reuters reported on November 12, citing unnamed sources, that FDA had shelved the voucher program; an HHS spokesperson said work on the pilot is ongoing and that FDA staff continue to evaluate applications granted a voucher. Compass has disclosed no FDA acceptance decision or review date. Ramey expects the risk evaluation and mitigation strategy that follows any approval to show which questions remain open.
The blind can break in the first hour
“Functional unblinding is still the central methodological problem,” Ramey said. In conventional CNS trials the threat to the blind is subtle. With a classic psychedelic, she said, most participants know within an hour which arm they are in.
The literature supports her. A systematic review of 112 randomized psychedelic trials by a University of Toronto group, published in JAMA Psychiatry in April, found that only 33, or 29.5%, assessed whether blinding held. Among those that did, psilocybin, LSD and ayahuasca studies frequently reported that more than 90% of participants and raters guessed assignment correctly, and MDMA trials with inert placebos exceeded 85%. A separate JAMA Psychiatry analysis this year put correct guessing at 90% to 95% in psychedelic-assisted therapy trials against 63% in blinded antidepressant trials, which are not perfectly blinded either. The difference, as Ramey puts it, is one of degree, and it appears after the first dose, not over weeks.
Her more unusual point is about where the bias sits. Expectancy is usually tested by asking whether participants’ hopes before dosing predict their outcome on the drug. But almost everyone who enrolls in a psilocybin trial expects a great deal. The main bias, she argues, is probably in the control arm: disappointment and a nocebo response after participants realize they received placebo, and a placebo response that never appears. “None of these is measured by pre-dose expectancy scales,” she said. That is her hypothesis; current trials are not designed to separate it from other explanations.
She adds that staff are often unblinded too, a point made by Matvey and colleagues in JAMA Psychiatry in 2025, and that this matters more than participant unblinding when the primary endpoint is clinician-rated. Active placebos help less than hoped. Usona’s Phase 2 trial, published in JAMA in 2023, used niacin, which causes flushing, as its comparator and did not assess whether blinding held. In a 120-person healthy-volunteer trial at the University of Zurich published this year, blinding was insufficient overall, and methylphenidate worked better as an active placebo for MDMA than for psilocybin. Asking participants how certain they were of their guess produced lower unblinding estimates than a simple right-or-wrong count, which Ramey calls a useful lesson in how to measure it.
“From a regulatory standpoint, no single fix will satisfy FDA,” she said. Her package layers several measures:
- consent language that discloses the design without revealing which arm is which
- a low but not inert dose of the drug as a pseudoplacebo
- central raters blinded to the dosing-day experience, with scripted interviews
- guesses about assignment, with confidence ratings, from participants and raters at several timepoints, and expectancy measured before and after dosing in both arms
- a prespecified estimand that treats unblinding as an intercurrent event
- durability data, since expectancy effects should fade over months
FDA’s final psychedelic trial guidance, issued in July, already asks sponsors to consider alternatives to inert placebo, central raters, blinding questionnaires with certainty scales and expectancy measures. It does not address estimands. Treating unblinding as an intercurrent event under the ICH E9(R1) framework is a proposal discussed in ISCTM working groups, not an established regulatory standard.
Her strongest single answer is dose response. “A dose-response design is the strongest single answer to ‘this is all expectancy,’” she said, “and one would rather pay for a third arm than defend a two-arm trial at an Advisory Committee.” Her preferred design has three arms, two active doses and a pseudoplacebo, with stratification by prior psychedelic use and a separate study of abuse potential.
The psychological support given around dosing raises a related problem. If support contributes to the effect, she argues, it has to be standardized, manualized and monitored for fidelity, and it becomes part of the label and the REMS. Her lesson from the Lykos outcome is to define the minimum support needed for safety, keep the drug effect and the therapy effect as separable as the design allows, and monitor sites closely. Her second lesson is to pursue next-generation compounds without perceptual disturbances. That is the stated aim of the neuroplastogen class her own company works in: Telesti lists reduced perceptual effects among its research objectives, and its lead program has not been tested in humans.
Ramey would take specific, answerable questions to FDA before the pivotal design is locked, on how blinding will be assessed and analyzed, what role psychological support plays, the durability strategy and the abuse-potential package, and would want the answers in meeting minutes.
Approval does not deliver care
Ramey’s decade at NIDA ends her argument in a field where approval has never been the finish line. Buprenorphine has been approved for opioid use disorder since 2002, yet CDC data show that only about one in four adults with the disorder received medication for it in 2022. “The problem was never only regulation,” she said. Clinicians lacked confidence with the drug, clinics struggled to get it reimbursed, stigma stayed high and payers added friction at each step. Psychedelic therapy, she argues, will face similar problems, made harder by six to eight hours of monitored dosing and a Schedule I history.
She draws four lessons for developers before approval.
Provider training. “A training mandate does not create prescribers,” she said. Under the DEA waiver that governed buprenorphine prescribing until Congress removed it in 2023, about half of waivered clinicians prescribed at all, a 2020 analysis in JAMA Network Open found. Developers, she argues, should build a short, practical training model with the field before launch and keep monitoring staff separate from prescribers, so that scarce psychiatrists do not become the bottleneck.
Patient access. Settings that are too restrictive keep treatment in a few places. Methadone for opioid use disorder can still be dispensed only through licensed opioid treatment programs, even after SAMHSA loosened take-home rules in 2024. Any REMS, she argues, should be sized to the actual risk, which for psilocybin is arguably mostly acute and confined to the session, and Phase 3 should produce the safety data needed to justify lighter monitoring later.
Reimbursement. “Clinics follow the economics,” she said. In her reading of esketamine’s experience, Spravato’s certified-site model works only when monitoring time is paid for; its REMS still requires at least two hours of observation after each dose. The psychedelic monitoring codes exist but do not guarantee payment, so she wants health-economic and durability data collected in Phase 3 and coverage negotiated before approval. Compass has pointed investors to the existing Spravato treatment-center network, which it put at more than 8,500 centers in its November 3 results, as infrastructure for COMP360, though it has not described an agreement to use it. Ramey calls building on that network the simple and cost-effective solution.
Treatment infrastructure. Vermont’s hub-and-spoke model for opioid use disorder, in which specialist hubs support community prescribers, expanded treatment capacity, according to a 2017 account in the Journal of Addiction Medicine. Psychedelic care, she suggests, may need the same structure: regional centers of expertise, protocols for referral and follow-up, and measurement-based tracking of patients after dosing.
Her conclusion ties the four together.
“Delivery is part of development,” she said. “If the care model is not designed and funded before approval, an approved drug can still fail to reach patients.”