Definium's LSD tablet beat placebo a second time. The number to read is 68%.

A placebo works because it is a disguise. The sugar pill looks like the drug, so nobody in the room can tell which one they swallowed, and whatever gap opens up between the two groups belongs to the chemistry. Definium Therapeutics' drug is a single 100 microgram tablet of lysergide, which is LSD, and on Monday the company reported that 68% of the patients who took it experienced "illusion" on dosing day. That is the trial's clinical word for seeing things that are not quite there. The disguise came off for two-thirds of the treated group within hours.

The trial still worked. Panorama, the second of two phase 3 studies of DT120 in generalized anxiety disorder, randomised 245 adults and measured the Hamilton Anxiety Rating Scale at week 12. Patients on the 100 microgram dose fell 9.8 points from a baseline of 28.3; patients on placebo fell 4.7 from 28.0. The 5.1-point difference carried a p-value under 0.0001 and a standardised effect size of 0.64. Every secondary endpoint hit too. Shares opened at $43.53, up 12% on the prior close, and finished the day at $40.25, up 3.5%.

Why the illusion rate matters more than the p-value

The p-value tells you the gap between the groups is unlikely to be chance. It says nothing about whether the gap is the drug or the expectation of the drug. If a patient feels the walls move for six hours, they know they got the real thing, and a person who knows they got the real thing tends to report feeling better. The person who felt nothing knows that too. Trialists call this functional unblinding, and for psychedelics the FDA has decided it is the central design problem: its final guidance on psychedelic trials, issued in July, names "functional unblinding of patients, therapists, monitors, or raters" as the source of expectation bias the sponsor has to manage.

The agency has already rejected a psychedelic on this ground. In June 2024 its advisory committee voted 9-2 that Lykos Therapeutics had not shown MDMA-assisted therapy worked for PTSD, with panelists pointing at unblinding: participants could tell whether they had been dosed. Two months later the FDA sent Lykos a complete response letter and asked for another phase 3. Lykos had two positive pivotal trials at the time. Definium now has three.

What the 50 microgram arm is for

Definium's answer to the blindfold problem is written into the trial design. Panorama carried a third arm of 52 patients on a 50 microgram dose, half the target, which the release says was "included to help mitigate functional unblinding." The logic: a half dose still makes the patient feel something, so they cannot use the feeling alone to work out which group they are in.

The half-dose arm improved 3.6 points more than placebo at week 12, against 5.1 for the full dose. Read one way, that is a dose response, and dose response is hard to fake with expectation. Read the other way, a dose that produces illusion in 61% of patients is not a neutral comparator; it is a smaller version of the same unblinding. The company says plainly that the arm "was not designed or powered for statistical comparisons," so neither reading gets a p-value. It is evidence, and it is the kind that persuades a reviewer rather than the kind that settles an argument.

Is 5.1 points a lot?

Translate it into people. At week 12, 32% of treated patients had cut their anxiety score in half, against 14% on placebo. Remission, a score of 7 or below, was 15% against 4%. Discontinuation was almost identical across arms, at 10.4% on the full dose and 10.3% on placebo, so the effect is not an artefact of the sickest patients dropping out of one group.

The first GAD trial, Voyage, reported in August with a 5.4-point placebo-adjusted difference and an effect size of 0.81. Panorama's effect size is smaller on a near-identical point difference, which means the spread of responses was wider, and the two trials agree on the thing that matters for a filing: a single dose still separates from placebo at twelve weeks. In depression, the Emerge study showed an 8.1-point gap on the MADRS scale at week 6, effect size 0.83. Three trials, three wins, one dose each.

What the release leaves out

No confidence intervals on the primary endpoint. No result from a blinding questionnaire, the tool the FDA's guidance points sponsors toward for measuring how many patients guessed their arm correctly. No breakdown of who the 68% were and whether their scores moved differently from the 32% who did not report illusion. Those are the tables the agency will read, and a topline release does not carry them; they arrive at a medical meeting or in the briefing document for an advisory committee.

The release does say how long the dosing session takes. Patients sat under observation for a minimum of eight hours, and the median patient met the discharge checklist at six. The drug is a tablet, but the product is a supervised day in a clinic, and that is what a payer will be asked to price.

What has to happen before anyone can prescribe it

Definium plans a pre-NDA meeting with the FDA in the fourth quarter and an NDA filing in the first half of 2027. An approval would not by itself put the drug in a pharmacy: LSD sits in Schedule I of the Controlled Substances Act, and the DEA would have to reschedule it before a doctor could write the prescription. That step has its own clock and its own politics, and neither is in the company's hands.

Money is not the constraint. The company, which renamed itself from MindMed in January, held about $1.1 billion at the end of June after an $805 million June offering, and guides that it lasts into 2030. That is past the filing, past a standard review, and past most versions of a rescheduling fight.

The market has priced this as a company that keeps winning trials, and it has treated the wins accordingly. On the day Voyage reported, shares opened at $47.51 and closed at $42.79; by early September they were under $38. Monday's open-to-close fade was the same shape. Positive data in a drug with an unblinding question is not a re-rating event, because the unblinding question is what the FDA review is for. Stifel raised its target to $67 from $60 on Monday and the stock closed at $40.25; that gap is a bet on the review, not on the science.

What to watch: the full Panorama dataset with the blinding assessment, the advisory committee the FDA is likely to convene for a Schedule I drug, and the DEA's timeline once a filing exists. The efficacy question has been answered three times. The question that decides the launch has not been asked yet.

This is analysis, not investment advice.