A phase 3 psychiatry trial threw out data from some of its own sites. Results slip a year.

On 3 December, Bristol Myers Squibb told investors that it had found irregularities in how a small number of sites ran ADEPT-2, its phase 3 trial of Cobenfy in psychosis associated with Alzheimer's disease dementia; excluded those patients' data before database lock; consulted the FDA; and would keep enrolling [s1]. Results from the ADEPT programme — ADEPT-2 along with ADEPT-1 and ADEPT-4 — are now expected by the end of 2026 [s1].

The announcement is unusual in what it admits and unusual in what it withholds.

What the company said happened

BMS described "irregularities due to clinical trial execution at a small number of study sites" [s1]. It did not specify what the irregularities were. It said it excluded patient data from those sites from the primary analysis prior to database lock, commissioned an independent interim efficacy and safety analysis, and proceeded on the Data Monitoring Committee's recommendation to continue enrolment [s1]. The company said it remains blinded to the trial data [s1].

Laura Gault, MD, PhD, senior vice president and head of development for neuroscience at the company, said the decision to exclude data from affected sites "reflects our unwavering commitment to safeguarding the integrity of our studies" [s1].

Why site quality is a live problem in this particular field

Trials in Alzheimer's-related psychosis are hard in ways that create pressure at the site level. Eligible participants have dementia, which complicates consent and symptom reporting. The primary endpoint in ADEPT-2 is change in the Neuropsychiatric Inventory–Clinician score for hallucinations and delusions, with Clinical Global Impression–Severity as a key secondary [s1] — both clinician-rated instruments, which means the measurement itself depends on how carefully a site is run.

Recruitment for a condition this specific is also slow, which raises the value of every enrolled patient and, at badly run sites, the temptation to stretch eligibility. None of that is an accusation about ADEPT-2 specifically; the company has not said what it found. It is the structural reason a disclosure like this one lands harder in neuropsychiatry than it would in, say, oncology.

What excluding sites does to a trial

Removing data from selected sites after the fact is a defensible integrity measure and a statistical hazard at the same time. It shrinks the sample, which costs power. If the excluded sites differed systematically from the rest — in patient severity, in geography, in rater behaviour — it can also shift the population the trial ends up describing.

BMS's stated remedy is to enrol additional patients [s1], which restores sample size but pushes the readout out by roughly a year. The company's prior guidance had ADEPT-2 reading out sooner; the revised expectation is end-2026 for the programme [s1].

The treatment gap this sits inside

Psychosis in Alzheimer's disease — hallucinations and delusions, distinct from the memory loss — has no drug approved for it in the United States. Clinicians manage it off-label, most often with antipsychotics carrying warnings about increased mortality in elderly patients with dementia-related psychosis. That gap is why the ADEPT programme has been watched closely and why a one-year delay is not a neutral event for families dealing with the symptom now.

Cobenfy itself is approved for schizophrenia in adults. ADEPT-2 is a multicentre, randomised, double-blind, placebo-controlled study testing it against placebo in Alzheimer's-related psychosis [s1] — an entirely separate question from the approved indication, and one that remains open.

What is not known

The company has not said how many sites were involved, what the irregularities were, how many patients were excluded, whether any safety signal was implicated, or what the FDA advised. It has said the independent interim analysis was commissioned and that the DMC recommended continuing [s1], which is consistent with no overwhelming efficacy or futility signal, but a blinded sponsor reporting a DMC's continue recommendation tells outsiders very little about the drug.

What to watch

Whether BMS or the FDA discloses the nature of the irregularities. Whether the eventual publication reports results both with and without the excluded sites — the analysis that would let readers judge the exclusion for themselves. And whether the end-2026 timeline holds, given that it now depends on enrolling additional patients into a trial that was already difficult to fill.

Sources

1. Bristol Myers Squibb Announces Continuation of ADEPT-2 Phase 3 Study in Psychosis Associated with Alzheimer's Disease — Bristol Myers Squibb, 3 December 2025