CDER's small business assistance team posted two on-demand webinars on 23 June, then pushed one of them out to subscribers again on 27 July, which is the only reason it crossed my desk twice. There was no guidance attached, no Federal Register notice and no announcement, which is about the level of ceremony you would expect for internal training material. I think they deserve more attention than that.
The first covers the standard safety tables and figures integrated guide, along with the CDER Manual of Policies and Procedures governing premarket safety evaluation of new drugs and biologics. FDA's own description says attendees will come away knowing "the presentation and analyses for tables and figures that the Office of New Drugs considers important understanding drug applications, including adverse events, subgroups, laboratory, and vital signs analyses," plus an overview of what it calls the Targeted Analysis Guides.
The second is the one worth an afternoon. It covers the development and use of Office of New Drugs Custom Medical Queries, described as a way to "standardize groupings of adverse event terms" during premarket safety evaluation. There are 104 of them, governed by MAPP 6025.8, and the stated purpose is to "promote a standardized approach across OND Clinical review divisions." Each has a Narrow component for terms that strongly indicate the condition occurred and a Broad one that trades certainty for sensitivity; only four — muscle injury, hyperglycemia, hypoglycemia and hypersensitivity — carry an Algorithmic component that also reads laboratory values, concomitant medications, medical history or timing. The queries were previously called FDA Medical Queries, and there is a published account of how they were built that is considerably more useful than the webinar blurb.
That last phrase is the part I would underline. It is not describing what the agency wants from a sponsor. It is describing what the agency does with a sponsor's data once it arrives. We have been circling this from the other side for a while — when the reviewer has AI of their own, the useful question stops being how polished your draft is and becomes what the review side will do to it. A standard query set is the same question in a less glamorous form.
My position, and I hold it loosely enough to be argued out of it, is that this changes what a sponsor ought to know before filing without changing anything a sponsor is required to do. If reviewers are likely to look at my adverse event data through a defined set of groupings, I want to know where that view and my integrated summary of safety tell different stories, and I want to know while I can still write about it rather than after somebody has asked me to explain.
The mechanics are worth being concrete about. A sponsor's safety story is built on groupings of adverse event terms. Some come from Standardised MedDRA Queries. A good number are sponsor-defined, often because the standard groupings didn't fit the mechanism, and once one exists it tends to persist across a programme, because changing a grouping mid-development is worse than living with an imperfect one. That grouping is what the integrated summary of safety describes, what the tables count and what the narrative explains. Then the application goes in, and a reviewer applies a different set of groupings to the same underlying events. Most of the time the two views agree closely enough that nothing comes of it. When they don't, the sponsor usually finds out through an information request, months later, about its own data, in a framing it hasn't seen before. The response to that is not purely a re-analysis. Somebody has to write the explanation of why the sponsor's grouping is the better clinical lens and why the reviewer's query produced a different number, and that writing gets done under time pressure by whoever is available.
I want to be careful here, because I don't have a figure for how often those two views diverge in a way that actually matters, and I'd be suspicious of anyone who offered me one. What I'd say instead is that the cost is asymmetric. Running the comparison during drafting is cheap and fairly boring. Finding the divergence in an information request is expensive, and the expense isn't the analysis, it's the schedule. Anyone who has watched a safety analysis move to ITT late knows the shape of it: the recomputation takes an afternoon and the reconciliation across everything that quoted the old numbers takes weeks.
There is a specific reason the two views come apart, and it is more interesting than either being wrong. Standardised MedDRA Queries may exclude preferred terms presumed not to be drug related. The queries FDA uses do the opposite: they attempt to capture every instance of a medical concept regardless of aetiology or whether it could plausibly be drug related. So a reviewer's count of a given event can be higher than yours not because your grouping was careless, but because theirs is deliberately built to include cases yours was designed to filter out. That is a difference in purpose, not in quality, and it is much easier to explain in a document you wrote on your own schedule than in a reply to a question.
None of this makes the queries a requirement, and it would be a mistake to write as though it did. MAPP 6025.8 governs FDA's own review practice rather than instructing industry, and the published account of the queries says nothing about what sponsors should do with them. It does note that the queries don't include every MedDRA preferred term, so reviewing individual reported terms remains necessary regardless. Whatever else this is, it is not a substitute for your own analysis, and I'd distrust anyone who sold it as one.
There is still a reasonable objection here, and I've made it myself. Writing to a review division's internal working method is a form of over-fitting. Follow the reviewer's tooling too closely and you let a list stand in for a clinical judgement about which events belong together, which is precisely the judgement a medical writer and a safety physician are there to exercise. I don't find that objection decisive, because knowing where a second view differs from yours is not the same as adopting it. But anyone who thinks the question is settled hasn't spent long with it.
The tables webinar points somewhere less comfortable, and it's the part I'd think hardest about if I ran a safety writing function. As the reviewer's display layer becomes more standardised, with the tables, the figures, the subgroup cuts and the laboratory and vital signs analyses converging on a common set, the numbers become less and less the place where a sponsor distinguishes itself. Everyone's outputs start to look alike, which is rather the point of standardising them. There is a version of this that is good news, since a table you can audit back to the code that made it is easier to defend than one assembled by hand, and convergence on a common shape makes that easier rather than harder. What's left, though, is the writing around them: the paragraph explaining why an imbalance in a small subgroup is noise rather than a signal, why a laboratory shift resolved without intervention, why the algorithmic tier picks up events the narrow one correctly leaves out. That paragraph has always existed. In most organisations it gets written last, by whoever is left, at the end of a compressed timeline, and it is now carrying more weight than it used to.
Both webinars are on demand and neither is long. I've worked from FDA's published descriptions of them and from the paper describing how the queries were built, rather than from the recordings, so treat this as a reason to go and look rather than a substitute for looking. One footnote that isn't really a footnote: the query definitions are supposed to be available on FDA's own OCMQ page, and while writing this I could not get that page to load. It may well be a transient thing. It is also exactly the sort of source I was complaining about elsewhere this week — internal, revisable, unannounced when it changes, and quietly load-bearing for anyone writing against it. If you run safety writing on a programme heading into a filing, they strike me as a better use of two hours than most of what will compete for that time this week.