Broad Win, Subgroup Miss: The Label-Writing Problem Behind Exelixis's Zanzalintinib Review

A Broad Win and a Subgroup Miss, Mid-Review

On June 23, 2026, Exelixis reported that its STELLAR-303 trial of zanzalintinib plus atezolizumab met its overall-survival endpoint in the intent-to-treat population for previously treated metastatic colorectal cancer — but that the combination did not significantly improve overall survival in the non-liver metastases subgroup. FDA has accepted the NDA for zanzalintinib plus atezolizumab in that previously-treated mCRC setting, with a PDUFA date of December 3, 2026.

The same week, Pfizer's sigvotatug vedotin — a Seagen-derived antibody-drug conjugate from its $43B acquisition — failed to meet its primary overall-survival endpoint in a Phase 3 trial in advanced non-squamous NSCLC, calling the company's oncology strategy into question.

For most readers, these are pipeline-and-strategy stories. For a regulatory writing organisation, the Exelixis situation is the more instructive one — because it is a specific and difficult writing problem: what happens to the submission when the trial succeeds in the population you filed but misses in a subgroup, while the agency is actively reviewing the label. Mixed efficacy is not just a statistical problem. It is a label-scope problem: the dossier has to make the broad-population case while accounting honestly for the subgroup that did not benefit. The data are what they are; the label is written.

Why Mixed Results Are Harder to Write Than Clean Ones

A clean win is, paradoxically, the easiest writing job: the data support the claim, and the submission's task is to present them precisely. A clean failure is hard commercially but, for the writing organisation, relatively unambiguous — the program reframes or stops.

A mixed result is the hard case. When a drug succeeds in the ITT population but misses in a subgroup, the submission has to do several things at once that pull against each other. It has to make the case for the indication the data support. It has to characterize the subgroup result honestly, because a reviewer will find it regardless. It has to address what the subgroup miss means for the benefit-risk in the population being sought. And it has to do all of this while the agency is mid-review, which means every framing choice is being read in real time by the people deciding the label.

The writing organisation cannot make the subgroup result disappear, and should not try. What it controls is whether the submission presents the totality in a way a reviewer reads as candid and coherent, or in a way that reads as either overstated (foregrounding the win, minimizing the miss) or self-defeating (so cautious about the miss that the win is undersold). Both failures are writing failures, not data failures.

What the Mixed-Result Submission Has to Do

State the subgroup result before the reviewer finds it. The fastest way to lose a reviewer's trust on a mixed readout is to make them discover the non-liver-metastases miss buried in a table after the narrative has emphasized the ITT win. The submission that characterizes the subgroup result clearly, early, and in context reads as candid. The one that foregrounds only the positive and lets the negative surface from the data reads as managed — and a reviewer who feels managed scrutinizes everything else harder.

Frame the benefit-risk for the population actually filed. Approval is being sought for previously treated metastatic colorectal cancer on the strength of the ITT result. The benefit-risk argument has to be built for that filed population on its own terms, with the subgroup result addressed as a characterization question rather than allowed to define the whole submission. That requires the writing to separate two questions the data tangle together: "did the combination work in the population we filed" and "what do we understand about where it worked less well." Conflating them is where mixed-result submissions go wrong.

Pre-answer the label-scope question the agency is already weighing. When the trial met its endpoint in the ITT population but a subgroup missed, the central question becomes whether the label should stand broad, carry a subgroup caveat, or be narrowed. The submission that does not address that question leaves the agency to resolve it unaided — and an agency resolving a label-scope question unaided tends toward caution. The submission that proposes a defensible label position, supported by the totality of the data and honest about the subgroup, gives the agency a path it can adopt.

Keep the clinical overview and the statistical narrative telling the same story. Mixed results are where Module 2.5 and the statistical sections most often drift apart — the clinical overview emphasizing benefit, the statistics carrying the subgroup miss, with no reconciling narrative between them. A reviewer who finds the clinical overview and the statistics implying different conclusions about the same data generates questions that consume the review. Internal consistency is always a writing discipline; on a mixed readout, it is the discipline.

Pfizer Is the Cleaner Case

Pfizer is the cleaner case: a failed primary endpoint. The writing task there is trial reporting, program-decision documentation, and portfolio reframing — real work, but unambiguous in direction. When an acquisition's central asset misses outright, the trial has to be reported, the program decision (discontinue, repurpose, or pursue a narrower path) has to be documented, and the portfolio implication has to be written for governance. Exelixis is harder because the trial still supports a filing argument while the subgroup complicates label scope — and that is the writing problem this piece is about.

The Wider Pattern

Mixed and negative oncology readouts are a structural feature of the field, not an exception — the more shots on goal, the more subgroup misses and outright failures. The sponsors that handle them well are the ones whose writing organisations treat a mixed result as a writing problem to be solved with candor and coherence, rather than a data problem to be managed with emphasis. The Exelixis situation — a broad ITT win, a non-liver-metastases miss, an agency mid-review toward a December PDUFA — is the textbook case, and how the submission reads will shape the label as much as the data do.

The Readout Is the News. The Label-Scope Argument Is the Writing Problem.

The clean failure is painful, but clear. The mixed result is harder.

Exelixis has to preserve the all-patient efficacy argument while showing FDA that it understands the non-liver-metastases subgroup result and its implications for label scope.

That is not a matter of hiding the subgroup or over-explaining it.

It is a matter of writing the totality of evidence so the reviewer can see both things at once: the trial met its endpoint, and the subgroup still matters.

The readout is the news.

The label-scope argument is the writing problem.