The p-value was 0.0603. The threshold everyone uses is 0.05. On September 24, Acadia Pharmaceuticals announced that its Phase 2 study of remlifanserin in Alzheimer's disease psychosis had missed its primary endpoint, and in the same release said it was continuing the program into late-stage testing with two Phase 3 trials already enrolling.
A company can do that. Nothing in the rules requires a program to stop when a mid-stage trial comes in just past the line. But the gap between "we missed" and "we are proceeding" is where most of the interesting questions in drug development live, and this readout puts several of them in one place: what a near miss is worth, whether a secondary endpoint can carry a decision the primary one did not, and what it means when the fix for a failed trial is a change to who the trial enrolls.
What the numbers say
The RADIANT program is testing remlifanserin, also called ACP-204, in the hallucinations and delusions that accompany Alzheimer's disease. The drug is a 5HT2A receptor inverse agonist, which is the same target as Nuplazid, Acadia's approved treatment for psychosis in Parkinson's disease. Remlifanserin was designed to be less likely to cause QT prolongation, the cardiac electrical delay that has complicated the earlier drug's profile.
In the trial's primary analysis, the 60 milligram dose produced a 12.6 point reduction on a scale measuring hallucinations and delusions at week six, against a 10.4 point reduction on placebo. That is a standardized effect size of 0.26 and a p-value of 0.0603, which is a miss. Analysts at TD Cowen had told investors in mid-September to expect an effect size between 0.35 and 0.4, so the result landed short of the number the market had been given to work with.
Placebo arms in psychiatric trials move, and a 10.4 point improvement on placebo is a large move. A pronounced placebo response compresses the room available for a drug to separate itself, which is part of why effect sizes in central nervous system studies tend to run smaller than the same molecule might deliver against an objective endpoint like a lab value or a tumor measurement. The 60 milligram dose separated from placebo anyway and kept separating, which is the argument that the drug is active. The size of that separation is the argument that it is modest.
The 30 milligram dose showed minimal improvement across endpoints compared with placebo, and Acadia is removing it from the late-stage program.
Two details inside the primary result matter more than the headline. The first is that the separation between drug and placebo widened over the six weeks of treatment, which is what a drug that works slowly and cumulatively looks like, and it is the kind of pattern that makes a company argue the six-week window was too short rather than that the drug is inert. The second is that the analysis was run on a pre-specified population, not one assembled after the fact. That matters for everything that follows.
The secondary endpoint that carried the decision
The key secondary endpoint, a clinician-rated severity scale, went the other way. The 60 milligram dose reduced symptom severity by 1.3 points against 0.9 on placebo, an effect size of 0.37, with a p-value of 0.0077 that the company described as nominal.
Nominal is the operative word, and the arithmetic behind it matters. A trial designates one primary endpoint and controls the statistical error rate for that endpoint. Secondary endpoints are tested without that protection, which means the threshold for treating them as meaningful is higher than the number printed next to them, not equal to it.
The instrument matters too. A clinician-rated severity scale is coarser than a symptom inventory. It asks a doctor to place a patient on a single seven-point scale rather than to count symptoms, which makes it easier to administer and less sensitive to small movements, and it is closer to the judgment a physician would make about whether a treatment is doing enough to continue. A nominally significant result on that measure is a signal. It is not a substitute for the primary analysis, and the company has not claimed it is.
What the two results do together is more defensible than either alone. A drug that moved one measure and not another looks like noise. A drug that moved both, missed on one and hit on the other, and showed an increasing gap over time looks like a real but modest effect that a small study could not resolve. That is not a reason to skip Phase 3. It is a reason to want Phase 3 to be bigger and better aimed.
Why the company is continuing
The clinical case for pushing on is easy to state. More than seven million Americans are living with Alzheimer's disease, and roughly 30 percent of them develop psychosis at some point. There is no approved treatment for that condition. Physicians treat it off label, often with dopamine-blocking antipsychotics that carry their own risks in this population, including long-term movement problems that the analysts at BMO flagged as a reason to care whether a new drug avoids them.
BMO's read on the data was that the near miss and the significant secondary gave grounds for continued development. Stifel's was blunter. The result, its analysts wrote, was not the worst case, but there was no path to conviction, and the late-stage answer is years away.
The company's own framing was the strongest available. Catherine Owen Adams, Acadia's chief executive, called the data phase 3 enabling, said the efficacy, safety and tolerability profile made the program worth the investment, and pointed to the unmet need as the reason. A chief executive saying that about a missed endpoint is doing the job she was hired to do, and investors priced the difference between her reading and the market's within a day: the stock opened down about 11 percent, from $25.43 to $22.50.
The change that decides whether the do-over works
The amendments to the Phase 3 program are where this gets technically interesting, because one of them is a design change rather than a dose change.
Acadia is refining the enrollment criteria to bring in patients with somewhat higher psychosis scores at baseline. Applying those revised criteria backwards to the Phase 2 dataset raised the primary endpoint effect size from 0.26 to 0.33. That is a substantial improvement produced by changing nothing about the drug.
Enrichment of this kind is legitimate, and regulators accept it. A trial that enrolls people whose symptoms are too mild to measure change will struggle to detect a real effect, and selecting for a population where the drug has room to move is a reasonable response to a study that came close. The charitable version of the argument is that Phase 2 taught the company which patients the drug helps.
That trade-off has a commercial shape. A program that enriches its population buys a better chance of a clean result at the cost of a narrower label, because the evidence it generates will describe the patients it studied. Prescribing tends to spread beyond trial populations once a drug is approved, and a treatment that works in the sicker patients a trial selected for may behave differently in the milder cases a physician meets first. The enrichment question is therefore two questions stacked on each other: whether the effect is real, and who it is real for.
There is a second version of the argument, and it is the reason the near miss did not clear the skepticism. Criteria chosen with a view of the data are not the same as criteria chosen in advance, and the effect size that results is closer to a best case than an expectation. The relevant question for Phase 3 is whether the enriched population reproduces the signal in a trial large enough to be sure, and no amendment can answer that in advance. Acadia has said it is not proposing to change the primary endpoint, which keeps the eventual verdict comparable to the one it just missed.
The safety finding that analysts credited most
The cleanest result in the release is not an efficacy result at all. Across both doses, rates of adverse events, serious adverse events and discontinuations were similar to placebo. There was no signal of QT prolongation, no deaths in the treatment arms, and no indication of harm to motor symptoms or cognition.
In a population of older patients taking multiple medications, that profile is the asset. It is also the part of the readout with consequences beyond this indication. Acadia is running a separate Phase 2 study of remlifanserin in psychosis associated with Lewy body dementia, and the absence of a cardiac signal is the finding analysts said reads positively for that program, where the mechanistic case is considered stronger but clinical trial precedent is thin. That study is expected to report in early 2028.
What to watch from here
The detailed results will be presented at the Clinical Trials on Alzheimer's Disease conference in Boston in November, which is where the full dataset behind the topline becomes available for outside scrutiny. Between now and then, the two Phase 3 studies keep enrolling, and the enrollment criteria are the thing to read when they become public, because the size of the population and the baseline severity threshold determine how much a positive result will mean.
Remlifanserin received Fast Track designation from the FDA in July, which brings more frequent interaction with the agency and eligibility for expedited review, but no change to the evidentiary bar. The drug's path now runs through a study the company has already redesigned once, on the strength of a result that missed. Whether that reads as a program learning from its data or a program being positioned to pass a test it already failed is a question the Phase 3 results will answer, and the near miss means nobody gets to assume which one it is.
Primary sources
- Acadia Pharmaceuticals, topline results announcement, for the primary and secondary endpoint results, the effect sizes and p-values, the safety findings, the removal of the 30 milligram arm and the chief executive's statement.
- Fierce Biotech, for the analyst expectations before the readout, the BMO assessment, the remarks on the conference call, the enrollment enrichment numbers and the share price reaction.
- BioSpace, for the Stifel assessment, the Alzheimer's prevalence and psychosis figures, the Fast Track designation, the Lewy body dementia program and the analyst view on the safety profile.