In the span of two days this week, two biotech companies brought two very different drugs before two FDA advisory committees, and the pairing produced something close to a natural experiment. Both drugs targeted serious diseases with few or no options. Both rested on efficacy evidence that fell short of the cleanest conventional standard. Both had drawn sharp skepticism from FDA scientists in pre-meeting briefing documents. And both faced the same underlying question: should a drug reach patients when the proof that it works is contested?
The panels answered in opposite directions. Replimune's melanoma drug RP1, whose supporting trial FDA staff had called not interpretable, was endorsed 10 to 3. Capricor's Duchenne muscular dystrophy therapy deramiocel, whose apparent benefit FDA reviewers described as essentially a mirage, was rejected 3 to 9. Same week, same kind of problem, opposite verdicts. The reason the two split apart is more revealing than either vote alone, because it exposes what these committees are actually weighing when they judge imperfect evidence, and it is not simply data quality on a single scale.
Two kinds of doubt
The instinct is to conclude that RP1 had better data than deramiocel. That is not quite what happened, and the more precise account is the interesting one. The two drugs had different kinds of doubt attached to them, and the difference in kind, not degree, is what drove the opposite outcomes.
RP1's problem was a limitation of design, honestly arrived at. Its pivotal trial was single-arm and gave the drug alongside an established immunotherapy, which means it could not cleanly isolate how much of the benefit came from RP1 itself. But nobody accused Replimune of massaging the numbers. The results it reported, 47.8% of treated patients alive at three years and a median survival of nearly 33 months, were honestly generated. The uncertainty was about attribution: the outcomes were real, and the open question was how much credit the drug specifically deserved. One dissenting panelist captured the honest difficulty by saying he had nothing to compare the results to, which is a statement about interpretability, not integrity.
Deramiocel's problem was different in a way that matters enormously. FDA reviewers argued the drug had missed the original goals of its Phase 3 study, and that the benefits Capricor later trumpeted appeared only after the company changed its statistical analysis plan following the trial. The committee's recurring word was fragile; a trial-statistics expert on the panel called the results very fragile. The concern was not merely that the benefit was hard to attribute. It was that the benefit might not be real at all, that it might be an artifact of choosing, after seeing the data, a way of analyzing it that produced a positive result. That is the meaning of "mirage": something that looks like a benefit but may dissolve on closer inspection.
So the two doubts are categorically different. RP1's is "the good result is real, but we cannot fully prove the drug caused it." Deramiocel's is "the good result may not be real, because it seems to depend on how the numbers were cut after the fact." And panels treat those two doubts very differently.
Why panels forgive honest limits and punish apparent manufacturing
The dividing line the week revealed is trust, and the logic behind it is sound rather than arbitrary. A single-arm trial that cannot isolate a drug's contribution is a known, disclosed, tolerated limitation, especially in oncology, where desperate diseases have long justified accepting less-than-perfect designs and confirming the answer later. The data are what they are; everyone can see the limitation and reason about it openly. Nothing about it suggests the sponsor tried to make a failed drug look successful.
A positive result that materializes only after a post-hoc change to the analysis plan is a different animal, because it triggers suspicion about the one thing regulators guard most jealously: whether a claimed benefit can be believed at all. Pre-specifying how you will analyze a trial, before you see the data, is the safeguard that keeps sponsors from fishing through the results for whatever cut happens to look good. When the analysis plan changes after the fact and a previously-missed endpoint turns positive, it looks like the target was moved to fit the arrow after it landed. That does not prove the drug fails, and it may be innocent, but it corrodes trust in the result in a way an honest design limitation does not. Honest uncertainty is survivable at an advisory committee. Apparent manufacturing is close to fatal, because it puts every positive number from the trial under suspicion rather than just one.
That is why the drug whose data was literally labeled "not interpretable" won its vote while the drug whose data was called a "mirage" lost. Uninterpretable-but-honest beat maybe-real-but-manufactured-looking, because the panels were judging credibility, not just completeness.
The forces underneath, and the fair caveats
Several structural factors reinforced the split, and they are worth naming because they generalize. Field norms differ: oncology has a deep tradition of single-arm approvals and accelerated pathways for lethal cancers, so RP1's design read as a familiar, acceptable limitation, while cardiomyopathy sits closer to cardiology's stronger randomized-trial expectations, against which deramiocel's analysis looked weaker. Endpoint hardness differs too: being alive at three years is an unambiguous outcome that analytic choices cannot easily inflate, whereas the functional and surrogate measures in a cardiomyopathy study leave more room for a change in analysis to swing the result, which is precisely what the panel suspected had happened.
Fairness requires the caveats on Capricor's side, because its loss is narrower and more contested than a flat verdict that the drug does not work. The company's CEO forcefully disputed the FDA's briefing document, saying parts of it were hard to reconcile with what actually occurred. The rejected vote addressed a narrower cardiomyopathy indication rather than the drug's overall benefit-risk profile, and on a separate question about upper-limb function, the committee's feedback was directionally supportive of the Phase 3 evidence. Both votes are also non-binding: the FDA makes the final call, with decisions due August 2 for RP1 and August 22 for deramiocel, and the agency can and sometimes does depart from its advisers. Neither this analysis nor the votes themselves settle whether either drug works. The point is narrower and more durable than a verdict on the two drugs: it is about what made the panels lean opposite ways on superficially similar evidence.
There is also a live backdrop worth noting without over-reading it. Both meetings were watched as barometers of how the FDA, under new leadership, weighs benefit and risk for investigational drugs. But two votes are a thin foundation for grand conclusions about the agency's direction, and the cleaner lesson from the week is about the sociology of evidence, not the politics of the FDA.
How to read it
The useful thing the pairing teaches is a sharper question to ask of any contested drug approval. It is not enough to ask whether the evidence is uncertain, because almost all evidence is uncertain to some degree, and uncertainty alone did not sink deramiocel or save RP1. The better question is what kind of uncertainty it is: does the doubt come from an honest limitation that everyone can see and reason about, or does it come from analytic choices that make the headline result itself hard to trust? The first kind is often forgivable, especially for a desperate disease with a hard endpoint and a design whose limits are openly acknowledged. The second kind is much more dangerous, because it does not just leave the benefit unproven, it makes observers wonder whether the benefit was constructed.
Patient voices mattered in both rooms, and deserve respect rather than dismissal; more than thirty people spoke in favor of RP1, some describing their own treatment, and the panel weighed that testimony alongside the data. But testimony tends to be most decisive when it sits atop honest uncertainty, tipping a genuine close call toward access, and least able to rescue a result that the experts suspect may not be real in the first place. That is roughly the pattern the week produced. Two drugs, two desperate diseases, two imperfect evidence packages, and opposite outcomes, separated not by how much proof each had but by whether the proof they had could be believed. The drugs' fates now pass to the FDA, which will decide in the coming weeks whether to follow its advisers. What the advisers already showed is that in the weighing of imperfect evidence, the fault that is hardest to survive is not incompleteness. It is the appearance that the answer was arranged.
Primary sources
- STAT's live coverage for the 10-3 vote that RP1's IGNYTE data were sufficient, evaluable, and clinically meaningful, the FDA's tendency to follow but not be bound by its panels, and the regulatory history including prior rejections under former leaders.
- Benzinga for the 10-3 RP1 vote and 9-3 deramiocel vote, the FDA staff's "not interpretable" characterization of RP1's single-arm study, the 47.8% three-year survival, 32.9-month median survival, and 83.5% responder survival, and the August 2 RP1 action date.
- Quartz for the 10-3 RP1 vote, panelist Paul Chapman's dissent that he had nothing to compare the results to, Hussein Tawbi's support for near-term availability pending the Phase 3 readout, the 30-plus patient speakers, and the note that the FDA retains authority to decide differently.
- BioPharma Dive for the reviewers' argument that deramiocel missed its initial Phase 3 goals and that its trumpeted benefits were essentially a mirage aided by switching up a statistical analysis plan, and the framing of the verdict as a barometer of FDA decision-making.
- BioSpace for the 9-3 vote against deramiocel, CEO Linda Marban's dispute of the FDA briefing document, statistician Janet Wittes's very-fragile characterization, and the chaotic meeting dynamics.
- Capricor's press release for the official 3-for, 9-against tally, the non-binding nature of the vote, the narrower cardiomyopathy indication that excluded overall benefit-risk, the directionally supportive upper-limb-function discussion, and the August 22 PDUFA date.
- Replimune's SEC 8-K for the Class 1 resubmission acceptance and the August 2 goal date.