MM120 lysergide in generalised anxiety disorder: what is a signal worth when blinding fails?

The context

Generalised anxiety disorder sits uncomfortably in the consulting room. It is common, it lasts, and the pharmacological options have not changed in kind since the serotonergic antidepressants. The authors recall three framing facts: a twelve-month prevalence of around 10% among US adults, an inadequate response to first-line treatments in about half of patients, and no new drug approved for this indication in the United States since 2007. Many patients reach the consulting room already treated and partly improved. The question is then not which first line to choose, but what to offer next.

This is the space into which psychedelic development is moving. The compound assessed here, designated MM120, is lysergide tartrate, that is, a standardised pharmaceutical form of LSD, given as a single dose and not as daily treatment. The authors’ claim to priority is narrower than the coverage that followed would suggest: they write that this trial is the first to assess the dose-dependent efficacy of MM120 without concomitant psychotherapeutic intervention. They do not claim the first randomised controlled trial of lysergide in anxiety, and they point, on the contrary, to an earlier phase 2 trial combining LSD with psychotherapy, with a reduction in anxiety maintained for up to one year.

One practical constraint deserves mention from the outset: 84% of participants had to stop an ongoing treatment before enrolment, only 16% were treatment-naive, and 18% were in psychotherapy that continued during the trial. Among those who had to stop, 43% were taking a serotonin reuptake inhibitor, 34% a benzodiazepine and 19% an antipsychotic or an antiepileptic. The population analysed is therefore not a naive population, it is a population withdrawn from medication, which is not the same thing and does not compare in the same way with what is seen in the consulting room.

The mechanism

What the trial measures, and what it does not

This trial does not test a pathophysiological hypothesis, it looks for a dose. Lysergide is a serotonergic psychedelic, and the authors link the expected effect to agonism at the 5-HT2A receptor and to a sustained increase in neuroplasticity, while stating that the exact mechanism remains unknown. Two things need to be kept apart here. No biological marker, no imaging data and no pharmacokinetic analysis appear in the publication or its supplementary material. On the other hand, it would be wrong to say that no analysis of the link between the acute experience and response was planned: the statistical analysis plan lists among the exploratory outcomes a visual analogue scale of drug effects, the MEQ30 mystical experience questionnaire and the 5D-ASC altered states of consciousness scale, all assessed on the second day, as well as an analysis of blinding integrity and of the association between participants’ answers and the outcomes. The protocol even states that the choice of four dose levels is meant to help interpret the results in case of functional unblinding and to determine whether psychedelic effects are necessary for clinical benefit. These analyses therefore exist in the plan, they are not reported in this publication. This is an absence of reporting, not an absence of intent.

The study at a glance

Quality control

The findings

A word on the analysis method, because it governs how everything that follows should be read. MCP-Mod does not compare arms pairwise. The procedure first tests whether a relationship between dose and response exists, by confronting the data with several curve shapes defined in advance, six here, and then estimates the useful dose from the retained models. For the reader this means two things: the primary result concerns the existence of a slope, not the superiority of a particular dose, and strict control of type I error, ensured by a hierarchical gatekeeping procedure, covers only the primary outcome at week 4 and the key secondary outcome at week 8. The authors state explicitly that the other tests are not adjusted for multiple comparisons.

The primary outcome is met: at week 4, five of the six pre-specified dose-response models exceed the adjusted critical value of 2.032. At week 8, on the key secondary outcome, all six models exceed it. The curve is clear and instructive. The two low doses do not separate from placebo, with −1.2 points (95% CI −6.0 to 3.5) at 25 µg and −1.8 points (95% CI −7.6 to 4.0) at 50 µg. The two high doses do, with −5.0 points at 100 µg and −6.0 points at 200 µg. The minimum effective dose, defined as the one producing a 2.5-point difference from placebo, is estimated at between 56.7 and 85.1 µg, an estimate that remains surrounded by uncertainty since it is modelled and not measured.

The secondary analyses give higher figures than the model-based estimate, because they are based on observed cases without imputation. At week 4, the HAM-A reduction reaches −7.6 points at 100 µg (95% CI −11.8 to −3.4; p < 0.001; Cohen’s d 0.88) and −5.5 points at 200 µg (95% CI −9.7 to −1.3; p = 0.01; d = 0.64). The analysis cannot be more reserved than the data themselves on this point: a d of 0.88 is larger than the overall effect of 0.39 reported by the authors for available treatments for generalised anxiety disorder, an indirect comparison but one worth citing.

The trajectory is not monotonic, and that has to be said. At week 8, the secondary comparison at 100 µg does not reach the threshold (−4.3 points; 95% CI −8.9 to 0.3; p = 0.06), whereas the dose-response test remains significant. At week 12, the difference from placebo is −7.7 points at 100 µg (95% CI −12.6 to −2.8; p = 0.003; d = 0.81) and −7.4 points at 200 µg (95% CI −12.4 to −2.4; p = 0.004; d = 0.77). Response rates at week 12 are 65.0% at 100 µg and 62.5% at 200 µg against 30.8% on placebo, and remission rates 47.5%, 45.0% and 20.5%.

Two reservations follow. The upper bound of the confidence interval for the primary outcome at 100 µg comes close to zero, at −0.4: the imputed estimate is compatible with a difference below the 2.5-point threshold the authors themselves adopt, as much as with a substantial difference. And the direct comparison between the two high doses is not tested here. The authors nonetheless conclude that 200 µg comes with more frequent adverse events without any gain in efficacy, and they retain 100 µg as the optimal dose for phase 3: that is a descriptive reading of their curves, not the result of a test.

The tolerability profile is consistent with what is expected of the drug, and it is reported in detail; in the authors’ words, “The adverse events were consistent with the expected effects of MM120.” At least one adverse event occurs in 97.5% of participants at 100 µg and 100% at 200 µg, against 56.4% on placebo, and 98.5% of all events occur on the day of administration. Two severe events, both a feeling of intoxication, are reported and resolved by the end of the session. A single serious adverse event occurred, a hospitalisation for a panic attack in a participant in the 50 µg arm, 97 days after administration and judged unrelated to the drug. Five participants, that is 2.5%, left the trial because of an adverse event, two in each of the 25 and 50 µg arms and one in the 100 µg arm. Raised blood pressure is reported in 10% of participants at 100 and at 200 µg against none on placebo, in a population from which the protocol excluded any heart disease, any uncontrolled hypertension and any prolongation of the corrected QT interval. No suicidal behaviour was observed; two participants, one on 25 µg and one on placebo, reported suicidal ideation without intent or plan, both with a comparable or more severe history. Acute effects had resolved in more than 97% of participants twelve hours after dosing.

Critical appraisal

The problem with this trial is not its analysis, which is careful, nor its primary outcome, which is met. It is blinding, and the authors deserve credit for not hiding it: they measured recognition of the treatment received on the second day and they publish the pooled result. Around 85% of participants on active drug correctly identified their allocation, and more than 65% of participants on placebo did too. With 92.5% visual perceptual changes at 100 µg against 10.3% on placebo, and 40% nausea against 7.7%, this figure is not surprising. The outcome is an anxiety scale rated from an interview, and therefore dependent on what the participant reports, and the informed consent explicitly described the known effects of the drug. Confidence in the belief of having received an active treatment can, on its own, shift a score of this kind.

What this means, and what it does not mean, needs to be stated precisely. It is a threat to internal validity, not a demonstration that the effect is nil. A signal can be genuine and poorly protected at the same time. The authors put forward an argument to set it aside: since unblinding occurs at every dose and the two low doses show no efficacy, the difference observed at the high doses would not be attributable to unblinding. The argument is admissible, it is not conclusive, because the intensity of the acute experience rises with dose and so does the strength of the participant’s belief about allocation: what separates the high arms from the low arms is not only pharmacology, it is also the participant’s degree of certainty. The analyses that would settle the matter, the association between answers to the blinding questionnaire and the outcomes, and the MEQ30 and 5D-ASC data, are pre-specified in the analysis plan but absent from this publication. The figure on central rater blinding, preserved in more than 80% of assessments, refers for its part to a conference presentation and not to a published article.

This reservation is not specific to this trial: it runs through the whole of psychedelic research, where the drug announces itself (see our analyses of control group outcomes in psilocybin trials and of the EPIsoDE psilocybin trial), and the authors rightly point out that it also exists in other psychotropic classes. It calls for measures that have long been discussed: full publication of recognition rates by arm, an active comparator producing sensory effects, outcomes less dependent on participant report.

The second point is independence. Nine of the twelve named authors are employees of MindMed, the statistician is a paid consultant to the sponsor, the first author receives funding from MindMed among others, and the last author declares research support from a long list of companies including MindMed. The sponsor itself declares that it took part in every stage, up to the decision to submit. That does not make the data false, and the use of independent central raters and of a pre-specified analysis procedure limits the degrees of freedom. It does, however, shift the burden of proof: a result of this kind only becomes solid after replication by a team with no link to the development of the drug.

Then come the ordinary limitations of a phase 2b trial, which the journal itself signals by publishing the work under its preliminary communication heading. The sample is modest, the confidence intervals are wide, recruitment took place in a single country and the population is 83% White. Above all, it is heavily filtered: the protocol excluded any personal or first-degree family history of bipolar disorder or psychotic disorder, any lifetime history of post-traumatic stress disorder, psychiatric comorbidities considered likely to confound the results such as major depressive episode, obsessive-compulsive disorder or panic disorder, any substance use disorder within the previous twelve months, any significant suicide risk, as well as heart disease, uncontrolled hypertension and electrocardiogram abnormalities. On top of this comes the medication withdrawal imposed on 84% of participants. The trial patient is not the patient in the consulting room. Nor is there an active comparator, so nothing can be said about how this compound compares with a serotonergic antidepressant.

A final word on what the authors conclude. They do not present this work as a mere dose-selection exercise: they write that their results “support the dose-dependent efficacy of MM120 and inform the dose selection for phase 3 pivotal trials”, and they retain 100 µg as the optimal dose for the pivotal trials. That is a defensible reading of their data, it is also the reading of a sponsor preparing phase 3, and it deserves to be read as such.

Level of evidence

The formal level of evidence is that of a single randomised controlled trial, CEBM 1b, published by the journal under its preliminary communication heading. Confidence is high on one point: there is a relationship between the dose given and the anxiety score at four weeks, and that relationship was tested according to a plan defined in advance, with strict control of type I error. It is also high for the description of immediate tolerability, which is detailed and consistent with what is expected of a psychedelic. It is low on what matters most for practice: attribution of the improvement to the compound itself rather than to the participant’s knowledge of it, a question the authors document but do not resolve; the true size of the effect, whose estimates range from −5.0 to −7.6 points at week 4 depending on the analysis; and transferability to patients outside the United States, European patients for example, who have not been withdrawn from treatment and often have comorbidities. The sponsor’s full control of the trial does not disqualify the results, it postpones their consolidation to an independent replication.

What you can do with this

Frequently asked questions

Annotated bibliography

Source study. Robison R, Barrow R, Conant C, et al. Single Treatment With MM120 (Lysergide) in Generalized Anxiety Disorder: A Randomized Clinical Trial. JAMA 2025;334(15):1358-1372. Published online on 4 September 2025, under the preliminary communication heading. DOI 10.1001/jama.2025.13481. PMID 40906494. The full byline is not printed in the documents consulted, which carry the mention et al.; the contribution and competing interest sections name twelve authors: Robison, Barrow, Conant, Foster, Freedman, Jacobsen, Jemison, Karas, Karlin, Solomon, Halpern Wernli and Fava. Trial registered as NCT05407064; protocol version 6.0 of 26 July 2023 and statistical analysis plan version 1.0 of 1 August 2023 published with the article. Funding: study funded by Mind Medicine (MindMed) Inc, which also supplied the drug; the sponsor declares that it took part in the design and conduct of the study, in the collection, management, analysis and interpretation of the data, in the preparation, review and approval of the manuscript, and in the decision to submit it. Writing assistance was provided by paid contractors. Competing interests: nine of the twelve named authors declare that they are employees of MindMed, several of them holding shares and patents assigned to the company; one author is a paid consultant to MindMed; the first author, employed by Cedar Clinical Research, declares grants from Janssen Neuroscience, Beckley Psytech, MindMed, Usona Institute and Compass Pathways; the last author declares research support from a very large number of companies, including MindMed, as well as royalties and patents. What the document contributes: the first set of controlled dose-response data on pharmaceutical lysergide in generalised anxiety disorder without concomitant psychotherapy, with a pre-specified dose-selection procedure, twelve weeks of follow-up and a detailed tolerability profile. What it does not allow: conclusions about attribution of the effect, since the blinding data show extensive recognition of the treatment received and the analyses that would link it to the outcomes are not reported, nor placing the compound relative to existing treatments, for want of an active comparator.
No other reference is cited in this analysis: it rests on the source publication alone and on its supplementary material.

Editorial collections

Topics