What FDA-Sponsor Dialogues Reveal About How the Agency Actually Thinks About Estimands
A Strategic Analysis of 700+ Regulatory Precedents and the FDA’s Treatment Policy Preference
An estimand is the precise statement of what a clinical trial is designed to measure: the specific treatment effect, the population it applies to, and how intercurrent events like treatment discontinuation or rescue medication are handled in the analysis. Once a niche corner of statistics, estimands have become a central feature of regulatory dialogue between sponsors and the FDA. But while industry conferences and white papers debate the theory of the ICH E9(R1) framework, something more revealing is hiding in plain sight: the actual regulatory documents where FDA reviewers and sponsors negotiate, disagree, and ultimately align on which estimands matter.
We ran a single search across FDA and EMA approval packages using the Basil Systems platform's Regulatory / Approval Search module. The term: "estimand." The result: 727 hits across 511 unique drugs, spanning both jurisdictions. But the most interesting finding wasn't the volume. It was the documents that captured real-time dialogue between FDA and sponsors about estimand selection, and the patterns that emerged from reading them.
Here's what the data tells us.
1. FDA Has a Clear Preference – And It's Usually Not What Sponsors Lead With
Across these documents, a consistent pattern emerges: FDA pushes sponsors toward the treatment policy estimand, which captures what happens to all randomized patients regardless of whether they stay on treatment. Sponsors, meanwhile, tend to show up with efficacy or hypothetical estimands that estimate what would happen if patients adhered to treatment.
In one GLP-1 receptor agonist submission, the sponsor proposed their primary estimand, and FDA responded with a direct "No, we do not agree" – twice in the same document – and recommended the ITT estimand instead. In a respiratory combination product review, the sponsor led with an efficacy estimand as primary, but FDA noted that the treatment policy estimand was "of primary regulatory interest to the Agency." And in the review of a major GIP/GLP-1 dual agonist, FDA used the treatment policy estimand as the basis for evaluating the primary endpoint across all five pivotal trials.
The takeaway for sponsors: if you're not leading with, or at minimum including, a treatment policy estimand, expect pushback. You might propose a different primary, but FDA will require the treatment policy analysis regardless.
2. FDA Is Actively Coaching Sponsors on Estimand Design
This goes beyond accept or reject. In multiple documents, FDA proactively recommended specific estimand frameworks to sponsors – in some cases before the sponsor even submitted their NDA.
One rare disease NDA stands out. At the pre-NDA meeting, FDA told the sponsor to focus on exactly two estimands: the treatment policy estimand and the while-on-treatment estimand. FDA even pre-computed its own analyses for both and asked the sponsor to confirm whether they could reproduce the results.
In two separate hereditary angioedema and bronchiectasis programs, FDA included unsolicited "Additional Statistical Comments" in pre-submission correspondence, walking sponsors through all five E9(R1) attributes with drug-specific examples. These weren't responses to sponsor questions. They were proactive guidance from FDA biostatisticians telling sponsors exactly how to structure their estimand framework.
Another program received similar treatment, but with a twist: FDA recommended both treatment policy and while-on-treatment estimands specifically for safety analyses, not just efficacy. This signals that FDA's estimand expectations are expanding beyond primary endpoints.
For sponsors paying attention, these documents are essentially a playbook for how to structure your estimand framework before it becomes a review issue.
3. Estimand Alignment Is Becoming a Standing Agenda Item in Pre-Submission Meetings
Multiple drugs show sponsors explicitly asking FDA to sign off on their estimand framework during pre-NDA and Type B meetings. Questions like "Does the Agency agree with the proposed primary estimand?" and "Does the Agency agree with the definition of estimands?" now appear regularly in meeting packages across therapeutic areas, including rare disease, respiratory, and metabolic disorders.
In one respiratory program, the sponsor asked about overall efficacy adequacy, and FDA raised the estimand disagreement on its own initiative. The Agency didn't wait to be asked – it flagged the estimand misalignment unprompted.
The signal is clear: estimand alignment is increasingly something FDA expects to resolve before submission, not during review. If you're not including estimand-specific questions in your pre-submission meeting packages, you're leaving alignment to chance.
4. Post-Hoc Estimand Changes Are a Red Flag – And Regulators Will Call You Out
Several cases involved estimand definitions or changes after trial completion, and each tells a cautionary tale.
In one dermatology program, FDA acknowledged that modifying the estimand after the study was completed is "different to modifications related to endpoint(s), patient population, or multiplicity adjustment." FDA essentially gave it a pass in this case, but the explicit acknowledgment that post-hoc estimand changes sit in a different category is a warning to sponsors who think they can finalize estimands during the analysis phase.
At EMA, one ALS product's post-hoc estimand was flatly rejected. The assessment report states that the estimand "is not agreed upon" and that since the estimand definitions were post hoc, "what the true estimand would have been, if defined in advance, is a hypothetical discussion" that was not further pursued.
The most dramatic example involved a rare disease NDA where FDA told the sponsor at the pre-NDA meeting to focus on two specific estimands. The sponsor submitted the NDA with a hypothetical estimand as primary anyway. The result: a Complete Response Letter, reanalysis by the reviewer, and a resubmission that finally adopted the while-on-treatment estimand.
The lesson: define your estimands prospectively. Post-hoc changes invite scrutiny, and ignoring FDA's pre-submission guidance on estimand selection can cost you a cycle.
5. COVID Forced Real-Time Estimand Innovation
The pandemic created genuinely novel estimand challenges that show up in the regulatory record.
One major obesity/diabetes product's statistical review documents a "hybrid estimand" that categorized missing data into two buckets: Category 1 for missing data due to exceptional circumstances like pandemics and natural disasters, and Category 2 for all other missing data. Each category had different imputation approaches. This wasn't in any textbook – it was a real-time statistical innovation driven by pandemic realities.
Another product took a different approach. Early discontinuation due to COVID-19 was initially included as an intercurrent event but was ultimately removed per FDA feedback, as the Agency did not consider this to be an intercurrent event. This is an important precedent: FDA signaled that pandemic-related discontinuations don't automatically qualify as intercurrent events and treating them as such can overcomplicate your estimand framework.
6. There's a Persistent Gap Between What FDA Asks for and What Sponsors Deliver
Perhaps the most striking finding is how often sponsors don't follow FDA's guidance on estimands, even after receiving explicit direction.
The rare disease case mentioned earlier is the poster child. FDA recommended two specific estimands at the pre-NDA meeting. The sponsor submitted with a hypothetical estimand as primary anyway. That decision contributed to a Complete Response Letter and a resubmission.
In an insulin analog program, the applicant had to reanalyze data using the Agency's preferred estimand of all randomized subjects and resubmit. The reviewer then corroborated the reanalyzed results – meaning the original submission with the sponsor's preferred estimand wasn't sufficient.
A glaucoma product represents the opposite problem: no estimand at all. The reviewer noted that neither the protocols nor the statistical analysis plans for the three Phase 3 studies specified the primary estimand of interest, forcing the reviewer to conduct their own analysis targeting the treatment policy estimand.
These cases suggest that estimand selection isn't just a statistical formality – it's a review risk. Sponsors who don't align with FDA expectations on estimands are creating unnecessary friction in the review process.
What This Means for Regulatory Strategy
The estimand framework is no longer optional, and it's no longer just the biostatistician's problem. These documents show that estimand selection is now a strategic regulatory decision that affects pre-submission planning, review timelines, and approval outcomes.
Three practical recommendations emerge from this analysis:
1. Lead with the treatment policy estimand or be prepared to defend why you didn't. FDA's preference is clear and consistent across therapeutic areas.
2. Include estimand-specific questions in every pre-submission meeting. The agencies expect to align on this before your NDA lands.
3. Define your estimands prospectively and stick to them. Post-hoc changes erode credibility and can trigger additional review cycles.
The regulatory record is the single best source of insight into how agencies actually apply the E9(R1) framework in practice – and it's all searchable if you know where to look.
This analysis was conducted using the Basil Systems platform's Regulatory / Approval Search module, which enables full text search across FDA and EMA approval packages, review documents, letters, and correspondence.
Author: Sam Kay, VP of Pharma, Basil Systems