On 28 July, FDA withdrew its May 2021 guidance on ANDAs for certain highly purified synthetic peptide drug products, on the grounds that it "no longer reflects FDA's current scientific thinking." The same notice says the agency plans to revise it this year, per the CDER guidance agenda. If you have a peptide programme that was written against that document, you are now in the odd position of having lost the thing you were citing without yet having the thing that replaces it, and there is no date on which that resolves.
That happened to be the fifth of these I noticed in about five weeks, which is why I started paying attention instead of just filing it.
Earlier in the same stretch, the Peripheral and Central Nervous System Drugs Advisory Committee briefly stopped existing. The Federal Register notice is unusually plain about why: the committee "was terminated on June 4, 2026, because its charter was not renewed on or before that expiration date" (91 FR 46444–46446). The Commissioner reestablished it about seven weeks later, having determined that doing so was in the public interest. Nobody decided that CNS products no longer needed an advisory committee. A renewal date passed. But if you were building a briefing package on the assumption that this committee would be available to convene, then for most of a quarter it was not.
Two papers came out from under people in June. NEJM retracted the pivotal data for Amgen's Tavneos, and on 24 June Nature Medicine retracted the widely circulated time-of-day immunochemotherapy trial in non-small cell lung cancer. We wrote about both at the time, mostly about what a retraction does to a dossier that already cited the paper.
The grounds for the second one are worth reading if you write protocols. The editors described substantial changes to the study registration, with endpoints, eligibility criteria, sample size and study design inconsistently reported and modified over time, and a translated protocol dated 2022 that cited studies published in 2023 and 2024. They concluded that they no longer had confidence in the integrity of the results. Strip away the journal context and that is a document whose own internal provenance didn't hold up: a protocol that couldn't account for its own version history, and a registration that didn't match what was eventually reported. Meanwhile the finding itself had already travelled into other people's protocol rationales and benefit-risk sections, on the strength of being interesting rather than being load-bearing. When it was withdrawn, nothing in an ordinary document control system put a hand up.
The one I keep returning to isn't a retraction or a withdrawal at all. In mid-July, Endpoints reported that Acting Commissioner Kyle Diamantas had written to Rep. Diana DeGette to say that some journal articles authored by former Commissioner Makary and his deputies "didn't reflect FDA regulations or policies." For about a year those commentaries were read as a signal of where the agency was heading. They were quoted in strategy decks. I have seen them used as the reasoning behind a regulatory position, including by people who were otherwise careful. When Makary left I wrote about what an interim leadership window does to an active submission, and I framed it as a timing problem. This is the part I got wrong: the exposure wasn't only the pause in decision-making, it was everything that had already been written on the strength of what the previous commissioner had published under his own name. The agency has now told Congress, in writing, that those articles were not agency policy, which doesn't mean the arguments in them were wrong — only that they never carried the weight people were putting on them.
I think the field is careless about this, and I include myself in that. We tend to sort sources by how persuasive they feel rather than by what kind of authority they actually carry. A commissioner's essay, a final guidance, an advisory committee discussion, a competitor's approval, a peer-reviewed pivotal paper and a published rejection letter all end up doing the same job in a strategy document, which is to make a position feel supported. They are not the same kind of object. Each of them can stop being true in a different way, on a different timetable, and mostly without announcing it.
A small example from this blog, which I offer because this is easy to spot in someone else's work and hard to spot in your own. Two weeks ago I published a post here about FDA resuming publication of its rejection letters, and I wrote that fourteen new letters went up on 10 July, following the Endpoints report. FDA's own weekly update for 13 July says it issued sixteen. I don't know which is right, or whether the two are counting different things. It is a trivial discrepancy and it changes nothing about the argument I was making. But if someone had cited that post internally and built a slide on it, the correction would have to travel from here to there, and there is no mechanism by which it would.
The structural reason this is hard is that regulatory documents cite in one direction only. You can open any CSR, protocol or Investigator's Brochure and read out what it depends on. You cannot go the other way. If you want to know which of your documents rest on a particular guidance, a particular paper, a particular set of meeting minutes, somebody has to go and look, and what they come back with is a list of the ones they remembered to check. This is the same argument I made about the document not being the source of truth, arriving from a different direction: if the file is a render of something else, then the citations in it are pointers into that something else, and pointers are exactly the kind of thing that ought to be resolvable in both directions.
I'd push back a little on how "audit trail" gets used in this context, because most of what is sold under that name records who changed what and when. That is a useful thing to have, and it is not the same as knowing what a document depends on. A complete history of every edit to a section tells you nothing about whether that section rests on a guidance withdrawn last week. Writer-defined provenance gets closer, because it at least ties a claim to the source the writer chose for it, but even that is a record of where a sentence came from rather than a live index you can query when the source moves.
The rejection letters are the case I find hardest to be even-handed about, because I have been enthusiastic about them in print. FDA paused proactive publication in early July, resumed about a week later, added a batch to the openFDA table on 10 July, and there is now rulemaking under way to formalise the practice. The letters already published didn't go anywhere, so this isn't a story about a corpus vanishing. What moved was something more mundane: whether you can count on more of them arriving, and how much of each one you will get to read once a redaction standard is written down. I still think the published letters are among the most useful things a regulatory writer can read. I'd just be careful about building a process on a supply that is currently being defined.
None of this argues for citing less, and I'm not proposing a new document type or another compliance exercise. The change worth making is narrower than that: keep a record of what each document depends on, in a form you can query backwards, so that when a source moves you can answer the only question that matters, which is what else needs looking at. Software has done this unglamorously for decades and calls it dependency tracking. It also depends on something I've argued for separately, which is that the link belongs in the draft rather than being added at the end — a citation you resolve during publishing was never a dependency you could track. In regulatory writing the bibliography still sits at the back of the document as a formality, when it could be the index of everything that might break.