Five years ago, Mindy Leffler moved her son into a college dorm. He was eighteen, had Duchenne muscular dystrophy, could no longer walk, and was losing arm strength, and the family expected his college independence to be brief. This fall she is moving him back into the same dorm for the second year of his master's degree. He has spent five years on deramiocel, the Capricor Therapeutics cell therapy that the FDA's advisory committee voted against in July.
Leffler is not a neutral observer, and her essay this week in STAT says so on its face. She developed the Duchenne Video Assessment, the outcome measure used in Capricor's phase 3 study, and she consults on the trial's data. Her son's story and her professional stake sit in the same essay, and both are worth taking seriously, because she is writing about something larger than one drug: the machinery of decisions that determines what a rare disease trial can prove, and what it cannot.
Every design choice is a bet against an unknown
Leffler's argument is that designing a rare disease study consists of hundreds of tiny decisions, and that sponsors make them with the best knowledge available while knowing they cannot predict what any single decision will do. Rare disease trials are trail-blazing by definition: there is often no predecessor study to copy, no validated endpoint to borrow, no historical enrollment data to project from. Every choice is made once, in the dark, and its consequences arrive years later.
Her essay walks through two examples that show how unglamorous and how consequential these choices are. Consider inclusion criteria. Narrowing them improves the odds of a statistically clean result, because a more homogeneous population produces less noise. But narrowing too far risks enrollment itself, and a trial that cannot fill fails its whole program regardless of how elegant its design was. Or consider the outcome measure. In her son's study the primary outcome is upper limb performance, divided into domains for shoulder, arm, and hand function. The total measure captures more tasks, which means it can register meaningful change in the weakest participants, whose remaining hand function matters enormously, and in the strongest, who still have shoulder function to lose. But the total also carries noise, because many participants have already lost shoulder function entirely while others will keep their hand function throughout the trial. A measure focused only on arm tasks might reach significance more easily, but it risks leaving the most meaningful change for the weakest and strongest participants off the table. There is no right answer visible from the inside; the choice only shows its wisdom or its cost when the data arrive.
A decade in, the field still lacks the map
The most sobering detail in Leffler's essay is that Duchenne trials are now in their second decade, and the field still does not have enough accumulated knowledge to guide these small decisions with confidence. Two decades of trials should have produced a playbook: which endpoints behave, which subpopulations are worth isolating, which imputation rules hold up. Instead each new trial re-argues questions that previous trials only half-answered, because the answers depended on choices that were themselves contested.
This is the structural problem underneath the deramiocel dispute. The FDA's advisory committee voted nine to three on July 29 that the available evidence does not support the treatment's effectiveness for the heart indication Capricor sought. The fight was never really about the raw measurements. It was about which set of decisions should govern the trial after the fact: which version of the statistical plan, which endpoint definitions, which patients get included in the analysis. The agency's reviewers applied the 2022 version of the plan and found no significant difference; the company says the final plan, locked before unblinding, shows a 54 percent slowing of upper limb decline with a p-value of 0.029, results that The Lancet published the day of the vote after independent peer review. Both sides have a coherent story. The trial's fate will be settled by which story the agency accepts as the trial's own.
The decision will teach the next trial how to be designed
There is a dimension of this dispute that matters beyond deramiocel, and it is the shadow the outcome casts forward. Rare disease sponsors are watching, and they will design their next trials in the light of what this one wins. If a flexible statistical plan and a refocused analysis carry the day, the lesson every sponsor learns is that the way through the FDA is to write plans that can be revised later, and the next decade of rare disease applications will arrive with versions attached. If the strict standard holds, the lesson is that the plan locked before the data is the only plan that counts, and sponsors will invest far more in getting the early decisions right, which is to say in exactly the tiny decisions Leffler describes.
Both lessons have costs. The first makes every future approval a little less trustworthy, because the market for favorable reanalysis will price itself in. The second makes trials slower and more expensive, because everything must be right the first time in a field where nobody knows what right is. The regulator's choice, in other words, is not only about this drug. It is about which incentive the field will operate under for the next ten years, and the boys enrolled in the next trial will be living with the answer even if they never hear this drug's name.
That is why the flexibility Leffler hopes for, exercised without sacrificing rigor, is such a delicate thing to ask for. Flexibility and rigor are compatible in principle: a pre-specified sensitivity analysis can answer a question the primary analysis cannot. The danger is that flexibility, once granted, becomes the product, and rigor becomes the branding. The FDA's job in the next few days is to draw the line where the field will look for it for a decade.
The regulator's choice is a patient-by-patient choice
Leffler reports that the FDA last week allowed Capricor to submit additional upper limb data and analyses, refocusing the application on the primary outcome. In her telling, the agency weighed the time a new trial would take against the function that young men would lose during that time, and sided with the patients. Whether or not one accepts that characterization, the choice the agency faces is real and unpleasant: enforce the strictest possible standard of pre-specification, or permit a trial to be read in the light of what it learned.
The strongest version of the strict case is not bureaucratic. A clinical trial is a promise made in advance about what will be measured and how, and the promise is the entire apparatus that separates evidence from wishful thinking. If the rules can be reshaped after the results are known, every negative trial can be reanalyzed into a positive one, and the drug that wins is simply the drug with the most flexible authors. The strongest version of the flexible case is not sentimental. Rare disease patients do not get a second chance at a decade of function, and a standard calibrated for a cancer indication with thousands of patients can operate as a death sentence for a trial population of a hundred boys. Both positions have passionate advocates, and the correct answer is not obvious to anyone honest.
Leffler's essay ends where her stakes are: "They are determining his fate." The decisions in question are not abstract for her. But the sentence cuts both ways. The same FDA process that looks like delay from one side looks like the only protection against false hope from the other, and both perspectives are held by people acting in good faith on behalf of the same patients.
The small decisions are where the fates get settled
What Leffler's essay makes clear, whatever one thinks of deramiocel's merits, is that the public drama of an advisory committee is only the last act of a decision process that has been running for years in choices that never made news: a domain in the outcome measure, an inclusion threshold, an imputation rule. Those choices are where the fate of a drug is largely decided, long before a vote, by people working with incomplete information under pressure they cannot see around. That is not a failure of rare disease research. It is its condition. The honest response is not to pretend the choices can be made with certainty, but to build the institutional memory that makes each successive trial a little less blind, so that the next family at a college dorm move-in has better odds than the last one did.
Primary sources
- Mindy Leffler's STAT First Opinion essay of August 20, 2026, for the trial-design framework, the inclusion-criteria and outcome-measure examples, the account of the FDA's decision to accept additional upper limb data, and her son's story. Leffler disclosed that she developed the Duchenne Video Assessment used in Capricor's phase 3 study and consults on its data; her perspective is that of a trial participant's parent and a consultant to the sponsor.
- STAT's July 29 report on the advisory committee for the negative vote and the committee's stated concerns, and BioPharm International's coverage for the FDA briefing documents' findings, including the hypersensitivity imbalance that raised blinding concerns.
- NeurologyLive for the vote count and committee members' comments, including acting chair Evan Snyder's statement, and The Lancet's peer-reviewed publication of the HOPE-3 results for the 54 percent slowing of upper limb decline and the p-value under the final statistical plan.