What the press release leaves out: the phase 3 melanoma recurrence trial, read against the original papers

<p>It is worth beginning with a distinction this week’s coverage did not draw, and which orders everything else: on 19 August 2026 no study was published. A press release was published, signed by the two companies that fund, run and analyse the trial. The document announces that INTerpath-001 met its primary endpoint and its key secondary endpoint, describes the result as «statistically significant and clinically meaningful», and offers not a single figure of its own: no hazard ratio, no confidence interval, no p value, no count of recurrences in either arm. It further announces that the data will be presented «at an upcoming international medical meeting», without naming it or dating it. A reader applying the most elementary rule of the discipline —a claim is judged by the evidence attached to it— currently faces a claim with no evidence attached. That does not make it false. It makes it unverifiable, which is a different and provisional category.</p> <p>The figures that did circulate, and that gave the news its air of an event, come from somewhere else, and the provenance matters. The 49% reduction in the risk of recurrence or death and the 59% reduction in the risk of distant metastasis do not come from the 1,137-patient trial just announced, but from the earlier phase 2b trial of 157. The press release attributes them correctly; the coverage, by placing them beside the larger number, invited readers to take them for results of the larger study. They are not. And when those same figures are read with their confidence interval alongside —a hazard ratio of 0.51 with a 95% interval running from 0.294 to 0.887— what the headline suppressed comes into view: the true benefit may be as large as a 71% reduction or as modest as 11%. The midpoint is what gets printed; the range is what is known.</p> <p>The study on which all of this rests is published, can be read, and deserves to be read with respect, because it is serious work. Weber and colleagues signed it in The Lancet in February 2024. They randomised 157 patients with resected stage IIIB to IV melanoma, two to one: 107 received the individualised messenger RNA product plus pembrolizumab, and 50 received pembrolizumab alone. The design was open-label: neither patients nor investigators were blinded to who received what. Recurrence-free survival favoured the combination with a hazard ratio of 0.561. Grade 3 or higher adverse events occurred in 25% of the combination group against 18% of the monotherapy group. Funding is stated with the clarity the journal requires: Moderna, in collaboration with Merck Sharp & Dohme. None of this is hidden; all of it is published; and almost none of it reached the newspaper reader.</p> <p>The most serious objection, however, lies neither in the sample size nor in the absence of blinding, both of which the authors themselves declare. It lies in what was measured. The primary endpoint, in the small trial and the large one alike, is recurrence-free survival: the time elapsed until the disease returns or the patient dies of any cause. And the phase 3 statement adds a precision that should not be passed over, since it includes it in its own definition: recurrence is assessed by the investigator. In an open-label trial, whoever determines that the event being measured has occurred knows what the patient received, and that circumstance is no minor technicality: it is the reason medicine invented blinding.</p> <p>Above all, the one measure no interpretation can bend is missing. Overall survival —how many patients are alive at five or ten years— has not been reported, and the statement says why: «the study will continue in order to evaluate other key secondary endpoints, including overall survival». It is an honest sentence and it is also the most important one in the document. That a treatment delays relapse and that a treatment prolongs life are two distinct claims, and the methodological literature has spent a decade warning that the second does not follow from the first. The systematic review by Prasad and colleagues published in JAMA Internal Medicine in 2015, examining validation studies of surrogate endpoints in oncology, found that most show low correlation with survival. This is not a marginal or ideological objection: it is the state of the question within the discipline itself.</p> <p>There is also a matter of words, and whoever writes these lines cannot let it pass. Everyone called this a vaccine. A vaccine, in the sense the public understands and three centuries of medicine established, is something given to a healthy person so that they do not contract a disease. This is something else: a product manufactured to the measure of each already-excised tumour and administered, in up to nine intramuscular doses, to a patient already diagnosed with melanoma and already operated on, on top of up to eighteen doses of a checkpoint inhibitor that costs what it costs. Calling it a vaccine imports an expectation of prevention that does not belong to it, and the patient who reads the headline understands that something now prevents cancer. It does not. There exists, at best, something that lowers the probability of its returning in someone who has already had it.</p> <p>The other half deserves the same frankness. No one in the primary documents has said this is a cure, and the investigators have been markedly more cautious than those who quote them. What is being tested is a therapy added on top of another therapy, individualised per patient, with a manufacturing chain that by its very architecture can be neither cheap nor one-off, and that is given in cycles. That is a technical description, not an accusation: but it explains why the vocabulary of cure appears in headlines and vanishes in protocols. Trials measure what regulators accept, regulators accept surrogate endpoints because they shorten timelines, and short timelines favour products that are administered continuously over those that would settle the problem once. One need assume no hidden intent to observe that a system rewarding delayed relapse ahead of prolonged life will, over time, yield more treatments than cures. The structure suffices; motives are unnecessary.</p> <p>None of the above licenses contempt, and this analysis contains none. That an individualised product, manufactured from the specific mutations of a specific patient’s tumour, shows a signal in two consecutive trials is a scientifically notable fact, and those who achieved it are working in earnest. If the data presented at the meeting confirm what the release anticipates, and if in a few years the overall survival curve genuinely separates, that will have to be said with the same clarity with which it is said today that we do not yet know. The doubt held here is not incredulity toward science: it is precisely the procedure science applies to itself when it is working well, and which these same authors observe when they write in a journal and suspend when they write in a press release.</p> <p>What must therefore be watched has a date and a shape. First, the conference presentation: whether the hazard ratio, its confidence interval and the distribution of recurrences by arm appear, or whether the adjectives are repeated. Second, the peer-reviewed publication of the phase 3 trial, the only document that will permit scrutiny of the protocol, the criteria for assessing recurrence and the handling of dropouts. Third, and above all, the overall survival curve when it comes. Until then, the exact formulation of what we know is this: in a large trial whose numbers have not been published, a product added to standard treatment delayed the return of melanoma more than standard treatment alone. That is good news. It is not yet the news the headlines gave.</p> <p><em>Legal Notice: This critical analysis is published under Art. 28 (news of general interest) and Art. 10 (right of quotation) of Law 11.723 on Intellectual Property of the Argentine Republic. The original work and its title belong to their respective author and publisher, both cited on this page.</em></p>