Single-Arm Trial Limitations at an FDA Advisory Committee

Every instinct tells you to lead with your strongest data and never hand the panel a weapon. You have one afternoon, a room of experts who can stall your programme, and a single-arm trial with a hole where a control arm should be. So you bury the weakness, dress up the response rate, and hope nobody goes digging. That instinct is wrong, and the public record shows why.

The sponsors who walk out of an FDA advisory committee with their credibility intact tend to be the ones who name their study's limitations first, before a panellist raises them, and in the FDA's own vocabulary. A panel that discovers an unstated weakness for itself stops weighing your evidence. It starts weighing your candour. That is a far worse afternoon to sit through.

One scope note. This is the live hearing, the public ODAC session, not the written submission or the reply to a follow-up letter. Those are different animals, covered elsewhere: the written back-and-forth after a single-arm trial in You've Conducted a Single Armed Trial and Now Regulators Are Asking Questions, and the submission-stage checklist in FDA Guidance on External Control Arms: A Checklist For Drug Developers. The room has its own rules.

The silence tax: what the sponsor doesn't narrate, the FDA narrates for them

Start with what it costs to say nothing.

When Karyopharm took selinexor (NDA 212306) to ODAC in February 2019, the efficacy story leaned on the single-arm STORM trial (NCT02336815) contextualised against a Flatiron Health external control. Read the FDA's own Multi-Disciplinary Review afterwards and you find the agency did not object to that comparison in passing. It catalogued it. There is a dedicated Figure 8 headed "Immortal Time/Selection Bias," Tables 33 and 34 on "Incomparable Baseline Characteristics," and named sections on selection-criteria, index-date and comparability issues. The agency ran its own corrected sensitivity re-analyses rather than accept the sponsor's comparison, and the approval that eventually came rested on a narrowed STORM subgroup, not on the external control at all. ODAC voted to defer, pending the randomised BOSTON trial. The margin is widely reported as 8 to 5; treat that number as reported, not gospel. [1][2]

Sit with that for a moment. Every one of those figures and tables is the agency framing the sponsor's limitations for it, in the FDA's own words, on the permanent record. That is the silence tax. We have no transcript of Karyopharm's slides, so nobody can say what the sponsor argued live. But the richest account of selinexor's external-control weaknesses was written by the FDA, not by Karyopharm. That is exactly the seat you do not want.

The FDA already published your single-arm trial's list of weaknesses

You are not guessing which weakness a panellist will raise. The FDA already published the list, and you can read it before you ever book the room.

Its March 2023 draft guidance, Clinical Trial Considerations to Support Accelerated Approval of Oncology Therapeutics, itemises five specific limitations of single-arm trials rather than one vague worry about the missing control [3]:

• Small safety databases: without a comparator, rare adverse events cannot be reliably identified.
• Time-to-event endpoints: the guidance treats these as "generally uninterpretable" without a concurrent comparator.
• Low-magnitude response rates: a low response rate, the guidance argues, may not be "reasonably likely to predict clinical benefit."
• Contribution of components: for a combination regimen, you cannot cleanly attribute the effect to your molecule.
• Cross-trial comparability: differences in trial design, patient population and how response was assessed can produce genuinely wrong conclusions from a historical comparison.

That is your pre-emption set, no more and no less. Map your package against those five before you build a slide, work out which genuinely apply, and address each one in your own presentation before the open Q&A. Why these designs draw the scrutiny at all is a longer story, told in The EMA vs Single-Arm Trials.

Watch how precisely that vocabulary walks into the room. When Incyte took retifanlimab (BLA 761209, single-arm POD1UM-202, NCT03597295) to ODAC in June 2021, the committee voted 13 to 4 to defer the decision. The FDA's own later review records that "the major review issue identified was whether the reported ORR was reasonably likely to predict clinical benefit" — limitation number three, almost word for word, not external-control mechanics, because POD1UM-202 had no registered external comparator. A separately reported account notes the briefing document also flagged sparse racial-minority representation: one Black patient, four Hispanic or Latino patients with race and ethnicity reported. [4][5] The list is not hypothetical; it shows up in the vote.

Say what it breaks, and what it doesn't

Naming the limitation is half the move. The other half is bounding it: saying, in one sentence, what the limitation compromises for the question the panel is actually voting on, and what it demonstrably does not.

UGN-102 shows the frame in action. UroGen took the drug (the ENVISION trial) to committee for low-grade intermediate-risk non-muscle-invasive bladder cancer, and the panel voted 5 to 4 that the benefit-risk profile was not favourable. A near-even split, against a real objection: no randomised comparator. What UroGen reportedly did not do was pretend the comparator gap away. Its widely reported argument named a genuine trade-off instead. UGN-102 is an office-based instillation; the standard of care, TURBT, is surgery under general anaesthesia. Against that backdrop it cited a 79.6% complete response rate at three months, with around 82% of those responders still recurrence-free at twelve months. [6]

It did not win the vote, and the dissent landed anyway. Dr Neil Vasan, explaining his no vote, said the benefit was "very hard to ascertain given the wide heterogeneity" and called the trial "a missed opportunity to run a randomized control trial." [6]

Then, three weeks after the negative vote, the FDA approved UGN-102 as Zusduri.

Hold both of those facts together, because that is the honest lesson. Framing the limitation moved a hostile room to within a single vote. It did not manufacture a yes. What disclose-and-frame buys you is narrower than a yes, and it holds: a room that argues about your evidence instead of your candour.

Steal your own thunder, carefully

Let's be honest about what can and cannot be proven here.

No published study measures whether disclosing your external-control limitations first changes how an advisory committee votes; that study simply does not exist. What exists is adjacent evidence, and it points the same way.

Legal-psychology research on a courtroom tactic known as "stealing thunder" (Dolnik, Case & Williams, 2003) has found that when a party volunteers damaging information about itself before its opponent can raise it, audiences rate that party as more credible than when the same fact is first surfaced by the other side. [7] Volunteered bad news reads as confidence; extracted bad news reads as something you were hoping nobody would find.

Peer-reviewed work lets me put the underlying idea more firmly. Inoculation theory (Compton, 2016) shows that forewarning an audience of a coming attack, and pre-emptively refuting it, builds resistance when the attack lands. [8] A panel that has already heard the bounded version of your limitation from you is, in effect, inoculated against the hostile version from a panellist.

One caveat, and it is the important one. The stealing-thunder effect has a documented boundary: it stops working the moment the audience decides your disclosure is a rehearsed tactic rather than genuine candour. Which is the whole problem with the room.

Devil's advocate: you cannot script the room

Here is the strongest version of the case against everything above.

You cannot pre-script a room, because the room does not agree with itself on what matters. Jaksa et al. (2022) compared how regulators and HTA bodies critique real-world and external-control evidence and found that agreement in those critiques was low: selection bias and confounding came up most often, yet no single limitation was the one every reviewer converged on. [9] So the tidy notion that you pre-empt one named weakness, tick the box and call the job done is a fantasy. The room does not work like that.

The sintilimab review makes that vivid. This was a randomised trial, not an external-control case, so it belongs here only for the dynamic. ORIENT-11 ran almost entirely in a single country, and the committee voted 14 to 1 that additional US-inclusive data were needed. The same fact, single-country enrolment, split the room by which value each member brought to it. Richard Pazdur reportedly called it "a step backward in achieving the racial diversity that we need." The lone dissenter, Dr Jorge Nieva, reportedly read the identical fact through access, arguing that "more drugs competing for those patients will have greater impact on health equity than the need for diversity in clinical enrolment." [10] Same limitation, opposite votes, different lenses in the same room on the same day. (Pazdur in his own words is worth reading in our ASCO panel notes.)

So does disclose-first survive that? Yes, with a narrower claim than a marketer would like. You do not control which lens a panellist brings to your data; that is theirs to choose. What you do control is only the framing they start from, bounded by you, or handed to them raw by whichever panellist has the sharpest axe to grind.

That said, to anyone tempted by the older, cheaper logic that single-arm evidence does not really hurt you at committee: that was arguably true once. Tibau et al. (2016), reviewing 82 ODAC transcripts from 2000 to 2014, found that the availability of randomised data did not appear to drive committee recommendations back then. [11] But the more recent Kesselheim-group work (Tibau et al., 2025; Liu et al., 2024) tells a longer story. Low-benefit single-arm evidence strongly predicts later withdrawal from the market, even for drugs approved at the time, with an odds ratio of 4.63 (95% CI 1.50–14.33). [12][13] The bill for thin evidence is deferred, not waived.

A lost advisory committee vote is not a dead drug

One more piece of honesty, because "losing the room" gets treated as a death sentence and the data say otherwise.

Daval et al. (2023) looked at how often the FDA follows its advisory committees. For initial approvals, it went with a positive vote 97% of the time (142 of 147), but with a negative vote only 67% of the time (40 of 60). Read that again: roughly a third of negative committee votes did not stop the drug. ODAC, for context, was the busiest committee they studied, 78 meetings, 19% of the total. [14] Both exhibits above bear it out. UGN-102 lost 5 to 4 and was approved three weeks later. Selinexor was deferred and eventually approved on a narrowed subgroup.

So why does disclose-first matter if the vote is not final? Because the FDA reviewer who signs the decision, not the panel, has read the Multi-Disciplinary Review, and the version of your limitations that lives in that document is the version that decides. A 10-0 rout is a different order of damage: aducanumab drew one, and a documented decline in physician trust followed (Dhruva et al., 2023), even after the FDA approved over its own committee (the full post-mortem is here). [15] The record outlives the vote.

The pre-emption drill

None of this works as instinct. It works as a drill. Here is the one I would run.

1. Map your package against the FDA's five before you build a single slide. Take the March 2023 list, mark which of the five genuinely apply to your study, and stop there. That set, no larger and no smaller, is what you pre-empt.
2. Write the "does / does not affect" sentence for each one. One sentence per limitation: what it compromises for the specific regulatory question, and what it demonstrably does not. If you cannot write the second half honestly, you do not have a framing problem. You have an evidence problem, and it is better to learn that now.
3. Put the concession in your own voice, early. It belongs in your core presentation, ahead of the open Q&A. Wait, and you have handed the framing to whoever raises it first, which is rarely a friendly voice. Bring it up yourself and the room hears someone who already knows their own data cold.
4. Bound it with the analysis the FDA would otherwise run itself. Sensitivity analyses, tipping-point analyses, pre-specified subgroups. Selinexor is the lesson: the agency will run these whether you do or not, so run them first and show your working.
5. Rehearse until it reads as candour, not choreography. The boundary condition is real; a disclosure that lands as a rehearsed tactic backfires. Years ago I sat in on a dress rehearsal for a rare-disease programme where the medical lead had a flawless answer to the response-rate question, delivered it beautifully, and a mock panellist we'd briefed to be difficult just asked why he'd waited to be asked. The room went quiet. That silence was the lesson, and it cost nothing to learn it in a conference room instead of in front of the committee. This is one place an outside mock panel earns its fee.

What we don't have

Which leaves the study nobody has run. Everything above sits over a gap worth naming plainly: no one has done the clean comparison that would settle it. Take a large set of advisory committee reviews of single-arm or external-control evidence, code each sponsor on whether it disclosed and framed its limitations first or waited to be asked, and measure that against the vote and the final decision, controlling for indication, unmet need and effect size. That study would price the framing in vote share, which two contrasting cases and a borrowed courtroom analogy cannot. It doesn't exist, and it's the one I'd most want to read.

References

[1] Karyopharm Therapeutics. STORM trial (single-arm). ClinicalTrials.gov: NCT02336815. https://clinicaltrials.gov/study/NCT02336815

[2] FDA. Multi-Disciplinary Review and Evaluation, selinexor (XPOVIO), NDA 212306. Center for Drug Evaluation and Research (Figure 8 "Immortal Time/Selection Bias"; Tables 33–34 "Incomparable Baseline Characteristics Using the Original/Updated Index Date"; sections "FHAD Selection Criteria Issues," "Index Date Issues," "Comparability Issues"; FDA's own sensitivity re-analyses, Tables 36–37). Selinexor ODAC deferral pending BOSTON, 26 February 2019; vote margin (8–5) widely reported, not stated in the review. https://www.accessdata.fda.gov/drugsatfda_docs/nda/2019/212306Orig1s000MultidisciplineR.pdf

[3] FDA. Clinical Trial Considerations to Support Accelerated Approval of Oncology Therapeutics. DRAFT guidance for industry, March 2023 (p.5 lists the single-arm-trial limitations verbatim: time-to-event endpoints "generally uninterpretable," low-magnitude response rates that "generally may not be reasonably likely to predict clinical benefit"). https://www.fda.gov/media/166431/download — Federal Register: https://www.federalregister.gov/documents/2023/03/27/2023-05910/clinical-trial-considerations-to-support-accelerated-approval-of-oncology-therapeutics-draft

[4] Incyte. POD1UM-202 trial (single-arm). ClinicalTrials.gov: NCT03597295. https://clinicaltrials.gov/study/NCT03597295

[5] FDA. Multi-Disciplinary Review and Evaluation, retifanlimab (Zynyz), BLA 761334, p.25 — recording the 24 June 2021 ODAC held on the earlier SCAC submission (BLA 761209): the majority (13 voted "Yes," 4 voted "No") voted for the regulatory decision to be deferred, and "the major review issue identified was whether the reported ORR was reasonably likely to predict clinical benefit." https://www.accessdata.fda.gov/drugsatfda_docs/nda/2023/761334Orig1s000MultidisciplineR.pdf — Briefing-document racial-representation detail reported by ClinicalLeader.

[6] CancerNetwork. "FDA ODAC Votes 5-to-4 Against UGN-102 Intravesical Solution for Bladder Cancer" (ODAC, 21 May 2025): 5–4 vote that the benefit-risk profile was not favourable; UroGen trade-off framing (79.6% complete response at 3 months, 95% CI 73.9%–84.5%; ~82% of complete responders remaining in complete response at 12 months); Dr Neil Vasan dissent. https://www.cancernetwork.com/view/fda-odac-votes-5-to-4-against-ugn-102-intravesical-solution-for-bladder-cancer — Approval as Zusduri, June 2025 (UroGen; ~3 weeks after the ODAC).

[7] Dolnik L, Case TI, Williams KD. (2003). "Stealing thunder as a courtroom tactic revisited: processes and boundaries." Law and Human Behavior 27(3). PMID: 12794964. https://pubmed.ncbi.nlm.nih.gov/12794964/ (Abstract confirms the stealing-thunder effect and its boundary condition — the tactic fails once opposing counsel reveals it was used; attributed as a finding, not quoted.)

[8] Compton J, Jackson B, Dimmock JA. (2016). "Persuading Others to Avoid Persuasion: Inoculation Theory and Resistant Health Attitudes." Frontiers in Psychology. PMID: 26903925. https://pubmed.ncbi.nlm.nih.gov/26903925/

[9] Jaksa A, et al. (2022). "A Comparison of Seven Oncology External Control Arm Case Studies: Critiques From Regulatory and Health Technology Assessment Agencies." Value in Health. PMID: 35760714. https://pubmed.ncbi.nlm.nih.gov/35760714/

[10] Trade-press coverage of the FDA advisory committee meeting on sintilimab (ORIENT-11): 14–1 vote that additional US-inclusive data were needed; Richard Pazdur and Dr Jorge Nieva remarks (ClinicalLeader, BioSpace, pharmaphorum, The Cancer Letter). Sintilimab is a randomised-trial case, cited here only for the value-lens dynamic, not as an external-control example.

[11] Tibau A, et al. (2016). "Oncologic Drugs Advisory Committee Recommendations and Approval of Cancer Drugs by the US Food and Drug Administration." JAMA Oncology. PMID: 26940233. https://pubmed.ncbi.nlm.nih.gov/26940233/

[12] Tibau A, et al. (2025). "Predictors of withdrawal of anticancer drug indications granted accelerated approval: a retrospective cohort study." EClinicalMedicine. PMID: 40687736. https://pubmed.ncbi.nlm.nih.gov/40687736/

[13] Liu ITT, et al. (2024). "Clinical Benefit and Regulatory Outcomes of Cancer Drugs Receiving Accelerated Approval." JAMA. PMID: 38583175. https://pubmed.ncbi.nlm.nih.gov/38583175/

[14] Daval CJR, et al. (2023). "Association of Advisory Committee Votes With US Food and Drug Administration Decision-Making on Prescription Drugs, 2010-2021." JAMA Health Forum. PMID: 37418270. https://pubmed.ncbi.nlm.nih.gov/37418270/

[15] Dhruva SS, et al. (2023). "Physician Perspectives on the Food and Drug Administration's Decision to Grant Accelerated Approval to Aducanumab for Alzheimer's Disease." Clinical Pharmacology & Therapeutics. PMID: 37218658. https://pubmed.ncbi.nlm.nih.gov/37218658/