The statistical portion, on an active vaccine IND — what the agency is actually asking, and how to close the question rather than open a new one.
An Information Request is a written question from an FDA review division during review of an application or an active IND. It is not a deficiency letter and not a clinical hold. It means a reviewer cannot complete an assessment on what was submitted and needs something specific to proceed.
The right response answers the question actually asked, in the reviewer's frame, with the smallest sufficient addition to the record. Most statistical IRs are asking a sponsor to make an existing decision auditable — to state an assumption, show a derivation, justify a choice already embedded in the protocol. Re-running the analysis a different way is usually the wrong move: it changes the record being assessed and invites a second question.
Statistical IRs cluster into a small number of recurring shapes.
A sample size justification that cannot be traced. The protocol states a target enrollment and a power figure, but the assumptions behind them — the effect size, the control-arm rate, the attrition allowance, the variance estimate and its source — are not all present. A reviewer cannot verify a number they cannot reconstruct.
An endpoint definition with an unresolved edge. For colonization or infection endpoints in particular, the question is usually how a subject is classified when the data are incomplete: a missed sampling visit, an indeterminate assay result, a subject who withdraws mid-window. If the definition does not say, the reviewer will ask.
Multiplicity described but not specified. Naming a correction procedure is not the same as stating the family of hypotheses it applies to, the order in which tests are conducted, and what happens to downstream conclusions when an earlier test fails.
Safety rules without operating characteristics. A stopping or pausing rule needs its false-alarm and detection behavior shown, not asserted. For rare events this often means an explicit Poisson or overdispersion-based assessment rather than a normal approximation.
Documents that disagree. The protocol, the Statistical Analysis Plan and the tables, listings and figures package must describe the same study. Discrepancies between them generate IRs reliably, and they are the easiest category to prevent.
Restate the question before answering it. Reviewers work through many submissions. A response that opens by naming precisely what was asked, in the agency's own words, is faster to assess and harder to misread.
Show the derivation, not the conclusion. If the question is about sample size, give the assumptions, the source of each, the formula or simulation, and the resulting number — so that a reviewer can reproduce it independently. A restated conclusion answers nothing.
Distinguish what was always true from what is changing. If the response documents reasoning that was in place from the start, say so explicitly, and leave the analysis plan alone. If the IR has revealed that something is genuinely underspecified, amend the SAP, state that you have amended it, and describe the change. Both are ordinary. Blurring them is not.
Answer only what was asked. Volunteering adjacent analysis is a common instinct and a costly one. Every additional analysis in the record is something a reviewer may have questions about, and an IR response is a poor venue for material nobody requested.
The measure of a good response is whether it closes the question. A reply that produces a follow-up IR has cost the program a full review cycle.
A statistical IR is usually asking why a choice was made. The person who made that choice can answer in two paragraphs what a substitute analyst reconstructs over two weeks — and still hedges, because they are inferring intent from a document rather than recalling a decision.
This is the practical argument for the same statistician writing the analysis plan and answering the agency's questions about it. It is not a matter of seniority. It is that the reasoning behind a design decision is rarely fully written down, and the IR is precisely the moment it is needed.
MRP Group serves as trial statistician on an NIH-funded Phase 2 whole-cell pneumococcal vaccine trial conducted by a hospital research institute in partnership with an international vaccine manufacturer, under an international Data Safety Monitoring Board. The role has covered authorship of the Statistical Analysis Plan across two protocol revisions, preparation of the statistical responses to FDA Information Requests on the active IND, development of the sample-size, power and safety assessment framework, and specification and review of the TLF package.
Sponsor named on request, subject to client approval.
An Information Request, or IR, is a written question from an FDA review division to a sponsor during review of an application or an active IND. It is not a deficiency letter and not a clinical hold. It signals that a reviewer cannot complete an assessment on the information provided and needs something specific in order to proceed. Responses are typically expected within a short window, often 30 days or less depending on the request.
No. It is a routine part of review and materially different from a deficiency letter or a clinical hold. It means a reviewer is engaged and needs something specific in order to finish. What matters is the response: a precise one closes the item, a diffuse one generates another.
The request states its own timeline, commonly 30 days or less depending on what is asked and where the application sits in review. The binding constraint in practice is rarely the calendar — it is how quickly the sponsor can assemble a defensible derivation for a decision made months earlier.
Only if the answer changes what will actually be done. Amending before database lock is ordinary practice when an analysis turns out to be underspecified. Amending to restate reasoning that was already in place creates a version history someone will later ask about.
Usually not, and usually the reviewer is not asking for that. An IR ordinarily asks the sponsor to justify what was pre-specified, not to replace it. Substituting a new analysis mid-review changes the record being assessed and raises a question about why the pre-specified approach was abandoned.
Whoever wrote the analysis plan, if at all possible. A statistical IR is usually asking why a choice was made, and the person who made the choice can answer in two paragraphs what a substitute analyst reconstructs over two weeks and still hedges. Continuity between the person who specified the design and the person who defends it is the single biggest determinant of how quickly an IR closes.
Have an IR open, or want the analysis plan written so you get fewer of them?
A short call will establish what the agency is actually asking and what closing it requires. Get in touch →
