Written by Adigens Health

The FDA’s Complete Response Letter (CRL) is the most courteous way to say try again. Most such letters correct manufacturing lapses or paperwork. But sometimes the problem is conceptual: the science may be solid, yet the story unconvincing.

In an unusual gesture of transparency, the FDA has recently published CRLs for public review, covering the last few years of them being issued. These letters reveal, in bureaucratic prose, the reasons a drug is denied approval. It’s mostly manufacturing lapses or missing data, but occasionally something deeper occurred. Regulatory failures rarely hinge on a single flaw. More often, they reveal a pattern of gaps, like small cracks between what was measured, what mattered and what could be compared. Often the fault is on evidentiary grounds: the data existed, but the argument did not.

When Good Data Go Nowhere

Takeda’s Eohilia provides a case in point. The trial produced clean numbers (histologic remission achieved, symptom scores improved), yet the FDA hesitated despite a clear unmet need and a breakthrough therapy designation. The agency questioned whether the gains on paper reflected real, sustained relief for patients. The endpoints were rigorous, but the narrative of clinical benefit was incomplete. In other words – clinical meaningfulness was not established.  Even though a recommendation was made to run a new trial, a clearly articulated plan for a more structured real-world comparison, designed under a target trial emulation framework, could have first contextualised the trial results and later clarified durability and long-term quality-of-life effects.  These are exactly the dimensions regulators now expect to see corroborated beyond the trial’s experimental setting.

A different tension played out in Amgen’s romosozumab program. Here, the drug’s efficacy was undeniable: patients’ bone density rose, and fractures fell. Yet a shadow of cardiovascular risk hung over the data. The imbalance of cardiac events between groups left regulators unsure whether the benefit outweighed the hazard. A well-defined comparison, drawing on real-world data to benchmark cardiovascular outcomes in comparable populations, could have transformed an ambiguous safety signal into a quantifiable risk. In regulatory terms, uncertainty without context becomes doubt; it’s no surprise then that upon eventual approval, a request for post-marketing evidence on this issue was mandated.

Eli Lilly’s donanemab faced a different type of a safety problem. Amyloid plaques shrank dramatically, but whether this translated into measurable cognitive preservation in a safe way over the long term remained unclear. The open-label extensions that followed offered little clarity as they were too short and ‘uncontrolled.’ What was missing was a structured bridge between biomarkers and lived outcomes. A longitudinal real-world cohort, built as an emulated trial with explicit follow-up and outcome definitions, might have helped regulators see what the brain scans could not: whether slowing amyloid truly safely slows decline.

At the far end of the rarity spectrum, Ipsen’s palovarotene for fibrodysplasia ossificans progressiva stumbled over its control arm. The company compared treated patients to a natural-history registry, but the analysis appeared to have been stitched together after the fact. Differences in disease severity, follow-up and measurement timing left the FDA unconvinced that the comparison held. This, too, was a problem of architecture: a pre-specified comparison with eligibility, flare definitions and outcome assessments laid out clearly for everyone to see could have converted a descriptive dataset into persuasive causal evidence.

Across these four cases, the theme is the same. The drugs were plausible, the evidence, less so. Each faltered not on statistics, but on storytelling by design. Regulators have grown weary of post-hoc justifications and partial narratives. What they now seek are frameworks among which target trial emulation is emerging as the leading one, that force discipline, transparency and pre-specification. These steps do not add decoration; they are the foundation that turns data into belief.

The Architecture of Conviction

The lesson from these letters is not that companies lack rigour, but that rigour without architecture is noise. In modern regulation, success is no longer measured by how much data are collected, but by how well these data mimic the logic of a trial.

Across these cases, the message is less about data quantity than design quality. Each company generated extensive analyses. None presented a framework that made causality (or credibility) inescapable.  Target trial emulation and external control arms built in the emulation framework are the emerging antidotes. The use of frameworks imposes the structure of a randomized trial onto observational data, forcing clarity about who is treated, who is not, and what constitutes success.

It’s a real possibility that the next generation of CRLs will not hinge on manufacturing or statistics but increasingly on inference. Regulators now expect real-world evidence to behave like a trial: transparent, prespecified and reproducible. Sponsors who begin with the data they have, rather than the question they must answer, will find themselves re-reading the same polite refusal.

Well-designed real-world frameworks won’t rescue weak drugs. But they can protect strong ones from weak storytelling. In modern regulation, data alone do not persuade, design does.

http://www.adigenshealth.com/

‍