ACAD Remlifanserin: 0.26 Effect Size, Alzheimer's Psychosis

Acadia Pharmaceuticals just told the market it is betting a registrational Phase 3 program on a drug that missed its Phase 2 primary endpoint by four one-thousandths of a point. The stock fell about 10% to $22.18, and the selloff is roughly right in direction and roughly wrong in magnitude. Here is the actual math investors should care about.

The Phase 2 portion of the RADIANT program tested remlifanserin, a selective 5-HT2A receptor inverse agonist, at 60 mg and 30 mg once daily versus placebo in 318 patients with hallucinations and delusions tied to Alzheimer’s disease psychosis, or ADP. On the primary endpoint, the change from baseline in the SAPS-H+D scale at week 6, the 60 mg dose delivered a -12.6 change versus -10.4 for placebo. That is a 2.2-point treatment difference, a standardized effect size of 0.26, and a p-value of 0.0603. It missed the 0.05 bar. The 30 mg dose did essentially nothing, with effect sizes of 0.05 and 0.11 across endpoints, which is exactly what a dose-response curve should look like and is mildly encouraging that the 60 mg signal is real rather than noise.

The secondary endpoint is where the bull case lives. On the CGI-S-ADP severity scale, the 60 mg dose produced a -1.3 change versus -0.9 for placebo, an effect size of 0.37 with a nominal p-value of 0.0077. The effect widened from week 4 to week 6, meaning the drug was still separating from placebo when the trial stopped measuring. Safety was clean: adverse events, serious adverse events, and discontinuations all tracked placebo, with no QT prolongation signal and no deaths in the treatment arms.

The mechanism matters here because it is the same one that already made Acadia a company. Remlifanserin is a 5-HT2A inverse agonist, meaning it does not just block the receptor but actively suppresses its baseline activity. That is the exact pharmacology of pimavanserin, Acadia’s Nuplazid, the first and only drug approved for Parkinson’s disease psychosis. When Nuplazid won that approval, its registrational trial showed a -5.79 point SAPS-PD change versus -2.73 for placebo, a 3.06-point difference that translated into a clear clinical signal physicians could see. The inverse agonist angle matters for tolerability too: because these drugs do not hit D2 receptors the way older antipsychotics do, they avoid the motor and metabolic side effects that make atypical antipsychotics dangerous in frail elderly Alzheimer’s patients. That is why the clean safety profile in RADIANT is not a footnote. It is the whole reason a 5-HT2A inverse agonist is the right tool for ADP, where the alternative is off-label antipsychotics that carry a black box warning for increased mortality in elderly dementia patients.

The unmet need is total. There is no FDA-approved therapy for ADP. About 7 million Americans live with Alzheimer’s, and roughly 30% of them experience psychosis at some point, which is a population of over 2 million people. The closest precedent is Rexulti, which won approval for agitation in Alzheimer’s in 2023, not psychosis. Nuplazid did $680 million in 2025 sales on about 25% share of roughly 130,000 treated Parkinson’s patients. ADP is a market more than 15 times that size.

That is the bull case, and it is real. The bear case is the effect size. A 0.26 standardized effect size is modest by any standard, and it is smaller than the effect Nuplazid demonstrated in Parkinson’s psychosis before its approval. Remlifanserin’s 0.26 sits in the gray zone where a drug can be statistically significant at a big enough sample size and still be hard for a physician to notice in the clinic. Acadia’s management called the result “encouraging” and cited a “consistent, dose-ordered signal.” A skeptic would call it a coin flip with good optics.

The Phase 3 design does not resolve the tension. The RADIANT program advances into two identically designed confirmatory trials of roughly 378 patients each, with screening and enrollment already underway. A 378-patient trial has materially more statistical power than the Phase 2 portion, so a true 0.26 effect would very likely clear the bar at Phase 3 scale. The problem is that effect sizes shrink when you move from a tightly controlled Phase 2 population to a broader Phase 3 population, and they shrink more in Alzheimer’s than almost anywhere else. The Alzheimer’s drug attrition problem is not just about mechanism; it is about effect sizes that evaporate between Phase 2 and Phase 3. For a psychosis program specifically, the risk is compounded by the challenge of measuring hallucinations in patients whose cognitive decline makes self-report unreliable, which is why the CGI-S-ADP clinician-rated secondary carrying nominal significance is a more durable signal than the patient-reported primary.

So what is Acadia actually worth? The market cap is $3.82 billion, and the company did $1.07 billion in product revenue in 2025 from Nuplazid and Daybue combined. That puts Acadia at roughly 3.6 times trailing revenue, which is not expensive for a profitable, growing biotech but is not the deep-value discount the “free lottery ticket” crowd claims. The honest way to think about it: Nuplazid and Daybue support a floor in the low $20s on their own, and the market is currently assigning remlifanserin roughly zero pipeline value after the near-miss. If ADP Phase 3 works, remlifanserin becomes a multi-billion-dollar peak sales asset in a 2-million-patient indication with zero competition, and Acadia re-rates toward $40 or higher. If Phase 3 misses, the stock gives back the premium it never had and trades on the base business in the high teens.

The readthrough question is whether Acadia can repeat its Parkinson’s psychosis success in Alzheimer’s psychosis, and the Phase 2 answer is a qualified maybe. The competitive picture is thin but not empty. There is no ADP-specific therapy on the market, but the field is not frozen: several companies have tried and failed in this exact indication because the placebo response in agitated, delusional elderly patients is notoriously high, and because measuring hallucination severity reliably in patients who cannot always articulate what they see is a trial-design problem as much as a drug problem. Acadia’s advantage is that it has already solved the regulatory endpoint problem once, in Parkinson’s psychosis, and it is applying the same playbook here with a clinician-rated secondary carrying the weight. In the broader psychiatry and mood disorder catalyst picture, the pattern of the last two years has been near-misses that got over-interpreted as wins, and this readout fits that template uncomfortably well. The 10% haircut is an overreaction, but a defensible one. The market is pricing the p=0.0603 miss as if the signal were fake, when the dose-response and the clean secondary say the signal is probably real but weak. The more important number is the 0.26 effect size, and that number says the Phase 3 is a genuine coin flip, not a slam dunk and not a dead drug.

The position: at $22.18, you are paying a fair price for a base business plus a free option on the largest untapped indication in neuropsychiatry. That is a better setup than most pre-Phase 3 biotechs offer, but it is a position to size at 1% to 2% of a biotech portfolio, not a conviction bet. Watch two things before the Phase 3 readout: whether Acadia’s FDA interactions in the Fast Track designation yield any signal on acceptable endpoints, and whether the full Phase 2 data at a medical meeting shows the 60 mg effect holding in the patients who actually had the most severe hallucinations and delusions at baseline. If it does, this becomes a buy. If the separation was carried by mild cases, the Phase 3 is a coin flip you do not want to own outright. The RADIANT trial is registered on ClinicalTrials.gov if you want to track the Phase 3 design as it matures.