Blog

What 445 Published FDA Rejection Letters Actually Say

2026-08-31

Until recently, a Complete Response Letter was a private conversation. FDA told a sponsor why their application was not approved, and unless the company chose to say so, nobody else found out.

That changed in 2025. FDA began publishing CRLs, first a batch of roughly 200, then a further 89, with a stated commitment to release them promptly after issuance. They are now available through the openFDA transparency API.

We pulled the whole corpus. As of FDA's 13 August 2026 update it holds 445 Complete Response Letters, issued between 2002 and 2026. Here is what is actually in there, and one thing that will mislead you if you are not careful.

The trap: you cannot compute an approval rate from this data

Every record carries an approval_status field describing where that application stands today. It is the obvious thing to aggregate, and doing so produces a number that looks like a dramatic finding:

Read naively, first-cycle outcomes collapsed after 2023. They did not.

The two release batches were selected on opposite criteria. The earlier release covered CRLs for products that went on to be approved. The later one covered products that are still unapproved. The split you are looking at is FDA's publication policy, not industry performance.

Any "percentage of CRLs that end in approval" drawn from this dataset is measuring the wrong thing. We have seen the figure quoted. It is not a real number.

What the corpus does support

445 letters, 311 distinct companies. Seventy-nine companies appear more than once. The most frequent is Accord Healthcare with six, followed by Tanvex BioPharma, Eli Lilly, Fresenius Kabi and Regeneron with five each. A CRL is not a mark of an unserious sponsor; it is a normal event in drug development that happens to large and small companies alike.

Mostly NDAs, but a substantial minority of biologics. 328 letters went to NDAs and 103 (23%) to BLAs.

They are long. The median letter runs about 8,700 characters, with the longest over 45,000. These are not one-page rejections; they are itemised lists of deficiencies.

Almost all of them name a person. 438 of 445 letters, 98%, carry a company_rep: the sponsor's named regulatory contact.

Manufacturing, not medicine

The consistent finding across published analyses is that CRLs turn on manufacturing far more often than on whether the drug works. In The FDA Group's review of 89 published letters, 56% contained facility-inspection issues and 41% cited product quality problems such as impurities or failed stability testing.

That reframes what a first-cycle failure usually is. Not "the science did not hold up" but "the evidence that this can be made reliably, at scale, in this facility, was not there yet." Often the facility in question belongs to a contract manufacturer.

Why this is worth reading rather than summarising

The obvious use of a published CRL corpus is benchmarking. The better use is narrower: read the letters issued in your own therapeutic area, by your own review division.

A CRL is the clearest statement you will ever get of what a specific division found insufficient, in its own words, about a product like yours. Reviewers are consistent. The deficiency that sank someone else's stability package is the one your reviewer will look for in yours.

That is a few hours of reading that no summary substitutes for, and it was simply not available two years ago.

What we do with this

Arca reads the published CRL corpus alongside your own dossier and your prior correspondence with FDA, and drafts a point-by-point response with every claim traced to the document it came from. When a number in your response contradicts what is in Module 3, it says so before the agency does.

If you want the background rather than the analysis, our glossary entry on Complete Response Letters covers what a CRL is, the three options that follow one, and the one-year clock.