Persuaded, Not Certain: Why I Voted Yes
Reflections on the FDA advisory committee review of RP1 plus nivolumab after progression on PD-1–based therapy
Before I explain why I voted yes, it’s important to explain why I nearly didn’t.
On July 30, 2026, the FDA convened an advisory committee to consider Replimune’s application for accelerated approval of RP1 plus nivolumab for adults with unresectable advanced melanoma after PD-1–based therapy. The meeting followed two complete response letters—in July 2025 and April 2026—after FDA had twice concluded that the application could not be approved in its then-current form. The question before us was whether the efficacy results were “evaluable and clinically meaningful”1.
The FDA review was thoughtful, rigorous, and, in many respects, persuasive. IGNYTE, the single-arm study supporting the application, was not originally designed as a pivotal trial. Its protocol evolved substantially. Response assessment was complicated by the intratumoral route of administration, treatment beyond progression, reinjection of lesions, missing assessments, biopsy and surgery, and the absence of a concurrent control capable of isolating what RP1 added to nivolumab. These were not minor technical objections. They went directly to whether the reported response rate and duration of response could be trusted1.
I still voted yes.
The simplest explanation returns to a basic scientific question: what was observed, and what should have been expected?
In IGNYTE, neither side was straightforward.
Interpreting the Response Signal
For much of the melanoma community, IGNYTE was encountered through its published result: a response rate of approximately one-third, a substantial complete-response rate, and responses that appeared durable2. Once results are published, they acquire a kind of permanence. The estimates enter presentations, reviews, and clinical conversations—and gradually come to stand in for the more complicated underlying evidence. They become “the data.”
The FDA’s patient-level review made clear that the observed result was not nearly so simple.
The applicant reported 47 responders, an objective response rate of 33.6%, and a median duration of response of 24.8 months. FDA’s primary analysis counted 22 responders in whom systemic activity could be assessed through noninjected target lesions, while retaining all 140 patients in the denominator. That produced a response rate of 15.7% and a median duration of response of 14.1 months. FDA also performed a sensitivity analysis excluding all patients without evaluable noninjected target lesions1; among the remaining 89 patients, the response rate was 24.7%a.
So what was observed: 16%, 25%, or 34%?
There is no completely clean answer because these analyses ask different questions. None was simply the “correct” response rate; each embedded a different judgment about what counted as evaluable evidence of benefit. The applicant’s analysis asked whether patients had an overall RECIST response while receiving RP1 plus nivolumab. FDA asked a narrower and, for this application, very important question: in how many patients could a response beyond the direct effect of injecting RP1 be demonstrated?
FDA’s framing made sense. A directly injected lesion receives an exposure that distant disease does not. Its regression may reflect local oncolysis or local immune activation rather than a systemic treatment effect. RECIST was developed to assess systemic therapies, and the combination of a locally administered drug with nivolumab makes attribution particularly difficult1.
Still, a durable response in an injected lesion is not clinically meaningless.
For some patients, local disease is what bleeds, ulcerates, becomes infected, causes pain, compromises function, or leads to morbid surgery or radiation. If an injected lesion disappears and stays gone, that matters. It may not establish systemic disease modification, and it may not support every possible indication, but it can still represent real benefit.
The more precise question is not whether injected-lesion responses count. It is what claim they support.
Responses confined to injected lesions may support local disease control. Responses in noninjected and visceral disease provide more convincing evidence of systemic activity. A broad indication for advanced melanoma reasonably requires confidence that benefit extends beyond whichever lesions happened to be injected.
The application did include evidence of regression in noninjected disease. The sponsor reported responses in both injected and noninjected lesions, including noninjected visceral lesions. But those analyses were exploratory and were not prospectively designed as clean, multiplicity-controlled tests of systemic activity. FDA was right to be cautious about them. They were nevertheless relevant to the overall assessment1,2.
The broader response-assessment concerns also mattered. Some patients were treated beyond progression and subsequently counted as responders. Some had new or enlarging lesions reinjected. Missing assessments made it difficult to determine whether progression had occurred before a later response. In other cases, biopsy, pathology, or surgery affected the best overall response designation, including classifications of complete response1.
The applicant’s 33.6% response rate therefore could not be accepted without qualification.
This is also where the question of labeling becomes important. FDA does not simply decide whether a drug “works.” It determines what the evidence permits the label to claim: for which patients, in what regimen, and for what kind of benefit.
The T-VEC precedent is instructive. Although its sponsor sought a broader melanoma indication, the eventual label was restricted to local treatment of unresectable cutaneous, subcutaneous, and nodal lesions and included a limitation stating that T-VEC had not been shown to improve survival or affect visceral metastases1.
A yes vote on RP1 therefore did not require accepting every possible interpretation of a broad systemic claim. It reflected a judgment that the application contained clinically meaningful activity, while leaving the final indication and limitations of use to be calibrated to what the evidence supported most securely.
Contribution of components was part of the same problem.
IGNYTE did not contain the randomized arms needed to determine precisely what RP1 contributed, what nivolumab contributed, or whether their interaction was additive or synergistic. FDA appropriately emphasized that uncertainty1.
The limited RP1 monotherapy experience nevertheless provided useful context. RP1 alone produced some tumor shrinkage in the small phase I dose-escalation cohort, but the experience was too limited, heterogeneous, and methodologically uncertain to establish meaningful single-agent efficacy. The regulatory record itself described RP1 as unlikely to have substantial single-agent activity in this setting1.
That limitation does not establish contribution of components, but it makes the combination hypothesis more biologically coherent. Patients had just progressed despite PD-1 blockade, while RP1 alone had not demonstrated activity approaching what was observed in IGNYTE. The most plausible interpretation was not that either component independently explained the responses, but that RP1 altered the tumor environment in a way that allowed continued PD-1 blockade to matter again.
This does not prove synergy. It does, however, support the judgment that both components belong in the regimen—and in the label.
Choosing the Counterfactual
But even if the observed response rate was closer to 25%, or perhaps 16%, its meaning still depended on what continued PD-1 therapy would have produced.
That estimate was also uncertain.
Replimune relied heavily on an expected response rate of approximately 6% to 7%. FDA’s position was that no historical control was sufficiently comparable to IGNYTE to make that estimate reliable. The available studies differed in prior treatment, adjuvant versus metastatic exposure, stage, disease burden, duration of prior PD-1 therapy, definitions of resistance, selection for treatment beyond progression, and response-assessment methods1. That criticism was fair.
The frequently cited 6% to 7% benchmark also deserves closer examination.
In a pooled FDA analysis, Beaver and colleagues evaluated patients who continued anti–PD-1 therapy beyond RECIST-defined progression. Among 500 selected patients who continued treatment and had evaluable follow-up imaging, 95 subsequently had at least a 30% reduction in target-lesion burden from the time of progression—a rate of 19%3. But this was not a conventional confirmed RECIST response rate, and the 500 patients represented a selected subgroup of the 2,624 patients in the pooled analysis. Beaver and colleagues themselves concluded that the absolute number of delayed responses was low and that the overall benefit of continuing treatment beyond progression remained uncertain.
Ribas and colleagues then reinterpreted those same data. They argued that the more appropriate denominator was all 1,361 patients who initially progressed, not only the 500 who remained well enough to continue therapy and undergo another scan. That produced an estimated delayed-response rate of approximately 7%. They also noted that some patients were already demonstrating substantial shrinkage of baseline target lesions when progression was declared because of a new lesion. On that basis, they proposed 6% to 7% as a reasonable null hypothesis for future single-arm trials after PD-1 failure4.
Beaver and colleagues responded that the percentages described different populations and different questions, and rejected 6% to 7% as a universal null hypothesis. The appropriate benchmark, they argued, would depend on the treatment strategy, line of therapy, available alternatives, and characteristics of the study population5.
That exchange is important. The Ribas estimate was a thoughtful reinterpretation of the Beaver dataset, not independent confirmation from a second cohort. The debate was fundamentally about the estimand and the denominator. Were we asking about later tumor shrinkage among selected patients who continued treatment and remained evaluable, or delayed response among everyone whose disease initially progressed? Neither denominator perfectly recreates the counterfactual needed for IGNYTE.
Still, uncertainty about the expected rate does not mean every possible expected rate is equally plausible.
If the true response rate to continued PD-1 therapy after genuine refractory progression were 15% or 20%, FDA’s primary estimate would offer little evidence of a meaningful RP1 contribution. If the expected rate were closer to 5%, then even the 15.7% analysis would suggest activity. And if the sensitivity estimate of approximately 25% better describes patients in whom systemic response could actually be evaluated, the absolute difference becomes clinically meaningful.
On balance, the lower expected range seemed more plausible.
One reason was MASTERKEY-115, which was not included in the FDA briefing document’s historical-control table. That phase II study evaluated T-VEC plus pembrolizumab following progression or recurrence on immediate-prior PD-1 therapy and prospectively separated patients with metastatic primary resistance, metastatic acquired resistance, and recurrence after adjuvant treatment6.
Among the 41 patients with established metastatic PD-1 resistance, there was only one conventional RECIST response, including no responses among the 26 patients with primary resistance. The more favorable activity was concentrated in patients who had recurred after adjuvant PD-1 therapy.
MASTERKEY-115 is not a perfect historical control. T-VEC is not RP1, the cohorts were small, and any cross-trial comparison is vulnerable to confounding. But it evaluated a similar strategy—adding an intratumoral oncolytic therapy to PD-1 blockade—and distinguished metastatic resistance from adjuvant recurrence in a way that many retrospective series do not. It therefore offered an imperfect but relevant comparison for what another intratumoral oncolytic–PD-1 combination had achieved in patients with established metastatic resistance.
The timing in IGNYTE was also important. Anti–PD-1 was the immediate prior regimen, confirmed progression was required, and the interval between prior PD-1 treatment and RP1 initiation was short1,2.
Biologically, this looked less like a distant rechallenge and more like progression during essentially continuous PD-1 blockade followed by the addition of RP1.
Delayed responses occur. But nivolumab continuation alone seemed unlikely to explain the full pattern of deep, durable, noninjected, and visceral responses that remained after FDA’s analytical discounting.
The Accelerated Approval Context
The accelerated approval context also mattered.
Accelerated approval does not eliminate the requirement for substantial evidence of effectiveness. It does, however, permit a provisional regulatory pathway based on an endpoint reasonably likely to predict clinical benefit, with confirmation required afterward1,7.
A randomized phase III trial, IGNYTE-3, is underway to answer the question IGNYTE could not. If it fails to confirm benefit, approval can be withdrawn.
The regulatory history cut both ways. After the first complete response letter, FDA suggested that Replimune consider using the ongoing randomized trial to generate an interim comparison of response rate and duration of response that might support accelerated approval. Replimune instead returned largely with the same single-arm package. That was a strong argument on the FDA side1.
At the same time, accelerated approval offers a way forward while definitive evidence is still being generated. It is not consequence-free, but it includes a formal mechanism for withdrawal if benefit is not confirmed.
Testimony, Experience, and Judgment
The testimony affected me, too.
It came not only from patients, but from their family members. Patients described the burden of treatment, the shrinking range of available options, and the uncertainty of what came next. Their family members described what it meant to live with the possibility of losing a spouse, a parent, or a child to melanoma.
One mother invoked the familiar observation that English has words for someone who loses a spouse and for a child who loses a parent, but no corresponding word for a parent who loses a child. The absence of such a word seemed to reflect a loss almost too painful to name.
As a parent, I felt the full weight of what she was describing.
That testimony did not change the response rate. It changed the stakes of the uncertainty. It made the unmet need impossible to reduce to a sentence in a briefing document and clarified what durable disease control could mean not only to patients, but to the people who loved them.
The melanoma experts who testified added something different, including investigators who had directly treated patients on IGNYTE. Their observations were not independent and were plainly vulnerable to bias. Investigators can be influenced by selection, recall, attribution, enthusiasm, and their own commitment to a development program.
But they also conveyed things that were difficult to capture in the application: what patients looked like before treatment, how quickly and coherently their disease changed, whether distant lesions appeared to behave in a biologically connected way, and how unusual the responses seemed compared with what experienced melanoma clinicians ordinarily observe after PD-1 failure.
Not everything clinically knowable fits neatly into a case-report form.
That testimony could not repair the study design or substitute for independent adjudication. But it was not meaningless. Several clinicians were willing to state publicly and unequivocally that these responses appeared real and clinically different from the occasional delayed response seen with continued PD-1 therapy alone.
I gave that judgment some weight.
But the testimony carried another kind of influence as well. Many of those advocating for approval were clinicians whose judgment I deeply respect. Their testimony carried informational value because of what they had observed, but it also carried social weight.
The public nature of the vote made that harder to ignore. Advisory committee members do not deliberate in isolation; they hear respected colleagues interpret the same record and then must state their own conclusions openly. That transparency is valuable, but it is not psychologically neutral.
In a decision this close, the task was to keep separating two possibilities: that their experience had added something important to the interpretation of the evidence, and that disagreeing with them publicly was simply uncomfortable. I cannot know whether an anonymous vote would have produced a different result. Maintaining an independent judgment required returning repeatedly to that distinction.
Pretending that consequential judgments are made in a social vacuum would be less honest than acknowledging it. But expert testimony should matter because of the information it conveys, not because disagreement carries a social cost.
The Decision
Those influences still had to be separated from the evidence itself. After working through the response analyses, the historical controls, the contribution-of-components problem, and the limitations of the study, the question became more concrete:
If RP1 plus nivolumab were available, would I recommend it to an appropriate patient with melanoma progressing on PD-1–based therapy?
Yes.
If I were that patient—fully informed of the uncertainty, the alternatives, and the need for confirmatory evidence—would I consent to receive it?
Yes.
Those questions are not the statutory standard, and they cannot transform an uncertain signal into substantial evidence. A therapy may be reasonable to discuss with a patient without warranting marketing approval. But here they served as a final test of whether the residual signal was clinically consequential rather than merely statistically interesting. The signal remained uncertain, yet strong enough to support both a clinical recommendation and an informed patient’s decision to proceed.
IGNYTE did not provide a precise estimate of treatment effect. Its survival analysis could not support a causal estimate of treatment benefit. It did not definitively establish the contribution of each component. And FDA convincingly showed that the published response rate was less stable than much of the melanoma community had appreciated.
But even after giving substantial weight to the responses most vulnerable to confounding by injection, surgery, biopsy, missing assessments, and treatment beyond progression, a clinically meaningful signal remained.
That signal included deep and durable responses, activity in noninjected disease, and responses among patients with visceral metastases, primary resistance, PD-L1-negative disease, and prior PD-1/CTLA-4 therapy1,2. In the setting of essentially continuous PD-1 blockade, nivolumab continuation alone seemed unlikely to explain all of what was observed.
The randomized trial still needs to answer the question properly. Future studies of intratumoral therapies also need response frameworks that prospectively distinguish local control from systemic activity rather than reconstructing that distinction after the fact.
The members who voted no may ultimately be proved right.
But the question before us was whether the efficacy results were evaluable and clinically meaningful.
Persuaded, but not certain, I concluded that they were.