Capricor's Duchenne vote is a verdict on design choices made years before the data

A Food and Drug Administration advisory committee voted against Capricor's Duchenne muscular dystrophy candidate deramiocel last week, and the vote landed on a secondary outcome rather than the trial's primary one ^[1]. In the same week the agency told the sponsor it could submit additional upper limb data and analyses refocused on that primary outcome, which means the panel's negative verdict did not close the file ^[7][12].

The account here comes from Mindy Leffler, writing in an opinion piece for STAT ^[13]. Two disclosures belong up front: she developed the Duchenne Video Assessment used in Capricor's Phase 3 study and consults on its data, and her son has been on deramiocel for five years ^[10][9].

By her description, the committee's problem was the definitiveness of the cardiac effect in a study population that included boys whose heart function was still stable, and the data were much noisier there than in the subgroup that already had signs of heart dysfunction ^[1][2]. Read structurally, that is less a finding about the drug than the downstream cost of an enrollment decision. Narrowing inclusion criteria raises the chance of statistical significance and lowers the chance of filling the study, and a trial that cannot enroll can take the whole program with it ^[4]. Whoever set those criteria was choosing between two failure modes without knowing which would arrive.

The endpoint decision has the same shape. The primary outcome, performance of upper limb, decomposes into shoulder, arm and hand function, and a sponsor can analyze the total or a single domain ^[5]. The total covers more tasks and therefore more chances to catch change in both the weakest and the strongest participants, but it also imports noise, because many participants have already lost shoulder function and many will keep their hands ^[6]. Restricting the analysis to arm tasks buys statistical cleanliness and forfeits signal at the two ends of the distribution ^[6]. Neither answer is wrong when it is made. One of them becomes wrong at the advisory committee.

Leffler's argument is that sponsors make these calls with the best knowledge available at the time and cannot predict which will matter, because they are the ones going first ^[14]. Her description of the field supports it: Duchenne trials are into a second decade and still lack the accumulated data to settle these choices with certainty, and other rare diseases are no further along ^[3]. She frames the regulatory question as a balance between the strictest possible reading of pre-specification and the ability to learn during a program, and calls striking it an art rather than a rule ^[11]. She also says the agency weighed the time a fresh biologics license application would take against function lost in the interim, and sided with patients ^[8]. That is her reading of the motive, not a stated agency rationale.

Watch which upper limb analysis Capricor submits, total score or a single domain, and whether the FDA accepts a refocus on the primary endpoint at this stage of review ^[5][6][7]. The answer sets the terms for the next sponsor writing an analysis plan for a population it cannot yet characterize ^[3][4].