Capricor Therapeutics is now trying to rescue its Duchenne muscular dystrophy drug deramiocel by submitting additional data from an open-label extension of the phase 3 trial the FDA's advisers voted against on July 29. The move has a logic that deserves to be stated before it is dismissed. The families and clinicians in the Duchenne community have watched boys on the drug keep function longer than anyone expected, and the extension contains their outcomes. The submission also has a structural limit that no amount of additional data can fix, because it concerns what an open-label extension can possibly show. That limit, not the company's intentions, is what will decide the application.
Adam Feuerstein, STAT's biotech columnist, calls the submission "a stall tactic," and argues the only remaining path to approval is an entirely new randomized trial. Whether or not that is fair to the company's motives, the reasoning underneath it is sound, and it is worth understanding in its own terms rather than as an attack. An extension study answers a specific question: what happened to the patients who stayed? The approval question is different: what difference did the drug make? Those two questions only converge under conditions an extension cannot provide.
The extension's patients selected themselves
An open-label extension is what it sounds like. After a trial's randomized phase ends, patients are offered continued access to the drug, with everyone knowing what they are taking. Participation is voluntary, and it is therefore a kind of election. The patients who remain are, disproportionately, the ones tolerating the treatment well and the ones who believe, or whose families believe, they are benefiting. The patients who lost function fastest, who suffered side effects, who gave up on the therapy, are absent from the extension's later data. Every extension dataset is filtered by the decision to stay.
That filtering is not misconduct. It is the structure of compassionate continuation, and it exists for humane reasons: patients who were assigned placebo in a fatal disease deserve access, and families should not have to sacrifice treatment for the sake of a clean dataset. But the same feature that makes the extension humane makes it weak as evidence. A survival curve in an open-label extension describes a population that has already been selected for doing well. The drug's effect and the patient's choice to continue cannot be separated in the resulting numbers, because the choice is partly a response to the effect. The FDA's reviewers, weighing exactly this point, called the efficacy data fragile, and fragility of that kind is not fixed by adding more months of follow-up to the same self-selected group.
The comparison the approval question needs does not exist in the extension
The deeper problem is the missing counterfactual. The regulatory question for deramiocel is not whether boys on the drug retain upper limb function. It is whether they retain more function than they would have without it. Answering that requires the thing the extension cannot contain: a contemporaneous group of untreated or placebo-treated patients followed under the same conditions. The randomized phase of HOPE-3 was supposed to provide that group, and the entire dispute now underway is about whether it did.
The two sides' versions of the trial read like two different studies. The FDA's reviewers applied the 2022 statistical plan and found no significant difference between drug and placebo at twelve months, with a between-group gap of 0.66 points on the upper limb scale and a p-value of 0.24. They also flagged that hypersensitivity reactions ran near 42 percent on drug against about 15 percent on placebo, an imbalance that in their view threatened blinding, and that a revised secondary analysis excluded 23 of 106 randomized patients. The company answers that the governing plan is the version finalized before unblinding, not what its chief executive, Linda Marbán, calls "an unsigned incomplete internal draft," and that under the final plan the trial showed a 54 percent slowing of upper limb decline with a p-value of 0.029. The Lancet published the full results under that plan after independent peer review, on the day of the advisory committee meeting. Both accounts are internally coherent. The extension data cannot adjudicate between them, because the extension has no placebo arm at all.
The blinding problem weakens the trial from the inside
One detail in the FDA's briefing documents deserves more attention than it got in the public fight, because it concerns the machinery of the trial rather than its statistics. Deramiocel produced hypersensitivity reactions in more than 40 percent of treated patients, against roughly 15 percent on placebo. That is not merely a side effect to be weighed. It is a signal: when a patient reacts visibly to an infusion, the patient, the family, and often the clinicians know which arm they are in. The FDA's reviewers flagged the imbalance as a threat to the double-blind design, because a trial in which assignment can be guessed is not fully blind even if the labels stayed sealed.
The blinding concern connects directly to the extension question. The outcome measures in Duchenne trials, upper limb performance and assessments of daily function, are scored partly by humans watching videos or administering tasks, and expectation shapes observation. If families and assessors knew who was receiving the drug, a small amount of hopeful scoring could tilt a measure whose clinical meaning hangs on fractions of a point, which is precisely the scale of the dispute: the difference between the agency's 0.66 points and the company's significant result is not a chasm in raw numbers. It is the difference between two readings of a noisy measure, and the blinding imbalance is one reason the agency argued the noise could not be trusted.
None of this proves the drug failed. It explains why the agency kept returning to fragility: when the measure is noisy, the blinding is compromised, and the plan was revised, each problem alone is manageable, and together they make the trial's signal difficult to separate from its expectations. The company's answer is that the final plan was locked before unblinding and the results survived peer review. Both things can be true. The extension data, arriving on top of all of it, adds length without adding the one thing that could settle the argument: a comparison arm that was never treated, never exposed, and never hoping.
The company's strongest argument is the price of the alternative
Capricor's position deserves its best form, and its best form is not about statistics. It is about what a new randomized trial costs in time and in function. A fresh trial in Duchenne would take years to enroll, years to read out, and it would ask families in the placebo arm to watch their sons' arm strength decline on schedule while a therapy their community believes in sits on the shelf. The agency's own process has acknowledged this cost before, and the advisory committee itself split over it. The patient representative who voted yes called the agency's objections procedural; the acting chair who voted no still floated a phase 4 commitment, approval with continued study, as a bridge. Neither is trivial. For a progressive disease that does not pause, delay is not neutral.
The honest counterargument is that the cost of approving on this evidence is also paid by families, just different families. If the standard bends here, every future rare disease sponsor learns that the way to approval is a flexible statistical plan and a persuasive extension dataset, and the patients harmed are the ones who later take drugs whose evidence was manufactured the same way. The strength of the FDA's position is precisely that it is boring: the rules exist so that the agency never has to decide, case by case, whether a family's hope outweighs a sponsor's analysis. Each side's strongest case is strong, which is why the vote split the way it did.
What the extension submission does, in the end, is put the price of the agency's standard on public display. The agency will decide, in days, whether the evidence that the rules require exists. The families will decide nothing, except whether to keep waiting. The asymmetry is the whole story: the agency bears the cost of saying no once, and the boys bear the cost of waiting, in function measured month by month, until the standard is satisfied.
The vote itself showed how close the community is to the question
The advisory committee's nine-to-three vote against effectiveness for the heart indication was nonbinding, and it was narrower than the numbers suggest. Some members who voted yes did so with visible discomfort. The acting chair, Evan Snyder, said that "as a scientist, I would have voted no," while floating a possible phase 4 commitment as a compromise. The committee's patient representative called the FDA's objections technicalities and procedural errors. The Muscular Dystrophy Association called the outcome deeply disappointing for the community and convened a town hall. The split is not between people who care and people who do not. It is between two ways of caring: one that weighs each month of a boy's function as it slips away, and one that guards the standard that decides, for every future family, whether a result is real.
The FDA's decision is due by August 22, days away. Whatever the agency does, the extension submission will have done its work: it will have placed the agency in the position of saying no, explicitly, to data that families experienced as real. That position is excruciating, and it is also unavoidable, because the alternative is to let self-selected continuation stand in for the controlled comparison that the trial was designed to produce. The boys in the extension deserve the dignity of having their outcomes seen for what they are: real experiences, real function, real hope, in a dataset that cannot, by construction, say whether the drug caused them.
Primary sources
- Adam Feuerstein's STAT report of August 20, 2026, for the company's submission of open-label extension data and his characterization of it as a stall tactic.
- BioPharm International's coverage of the FDA briefing documents for the agency's finding that the trial did not meet its pre-specified endpoints under the 2022 statistical plan, the 0.66-point difference and p-value of 0.24, the hypersensitivity imbalance, and the exclusion of 23 of 106 patients from the revised secondary analysis.
- Company statements and GlobeNewswire releases for Linda Marbán's account of the statistical plan versions and the 54 percent slowing of upper limb decline with a p-value of 0.029, and NeurologyLive and the rare disease press for the nine-to-three vote and the reactions of Evan Snyder, the patient representative, and the Muscular Dystrophy Association.
- The Lancet's peer-reviewed publication for the trial results under the final statistical plan.