Adding a personalized mRNA vaccine to pembrolizumab lowered the risk of melanoma returning after surgery

In Short. Adding a personalized mRNA vaccine to a standard immune drug lowered the risk of melanoma returning after surgery by about two-fifths. Experts disagree about how impressive that is, because results from different trials cannot be compared directly.

MELANOMA IS a cancer of the cells that give skin its color. Caught early, it can often be cut out, but once it has grown deeper into the skin or spread to nearby lymph nodes it tends to come back, sometimes in the lungs, liver or brain. Doctors call a tumor resected when surgery has removed it and scans show no disease left behind. Such patients look healthy, yet stray cancer cells too small to see may remain. Drug treatment given after surgery to kill those cells is called adjuvant therapy. Treatment given before surgery, while the tumor is still in place, is called neoadjuvant therapy. For several years the standard adjuvant drug for many of these patients has been pembrolizumab, sold by Merck as Keytruda, which releases a restraint that tumors place on the immune system.

On October 22nd Merck and Moderna released detailed results from INTerpath-001, a Phase 3 trial that added a vaccine called intismeran autogene to pembrolizumab. The figures appeared on an oncology news website before the planned presentation at the European Society for Medical Oncology (ESMO) congress in Madrid, and an ESMO press representative confirmed to Fierce Biotech that there had been an embargo breach. ESMO then released the abstract, and the companies issued their own statement the same afternoon.

The vaccine is unusual in being made for one patient at a time. Cancer cells carry mutations, and some of these produce altered proteins that the immune system can recognize as foreign. Such proteins are called neoantigens. To make intismeran, a laboratory sequences the DNA of a patient's tumor and of their normal tissue, picks out mutations unique to the cancer and writes instructions for up to 34 neoantigens into a single strand of mRNA. Injected into the arm, the mRNA prompts the body to make those proteins and train immune cells against them. The approach has practical limits. It needs a tumor sample good enough for sequencing (the trial admitted only patients who had one), and each dose must be manufactured for that patient. In the trial, patients started pembrolizumab at randomization and began intismeran or a dummy injection once their vaccine was ready.

The trial enrolled 1,137 adults whose melanoma had been resected. All had stage IIB to IV disease, the stages at higher risk of coming back. Two of every three were assigned at random to the vaccine plus pembrolizumab (757 patients) and the rest to a placebo plus pembrolizumab (380 patients). Neither patients nor doctors knew who got which.

The main measure, or primary endpoint, was recurrence-free survival (RFS), the time until the cancer returns anywhere or the patient dies. A second measure, distant metastasis-free survival (DMFS), counts only a return in distant organs, or death. The third, overall survival (OS), counts deaths from any cause. After a median follow-up of 24.6 months, 82.2% of the vaccine group were alive with no recurrence at two years, against 70.0% of the pembrolizumab-only group. For distant spread the two-year figures were 88.5% and 80.2%.

The headline number is a hazard ratio of 0.58 for recurrence or death. A hazard ratio compares how often the bad event happens in one group with how often it happens in the other at any given moment during follow-up. A value of 1 means no difference, and 0.58 means that at any point the vaccine group was having recurrences or deaths at 58% of the rate of the comparison group, a reduction of 42%. The 95% confidence interval, the range within which the true value plausibly lies, ran from 0.45 to 0.75. The hazard ratio for distant spread or death was 0.62 (0.45 to 0.85). Serious side-effects linked to treatment, those of grade 3 or worse, occurred in 14.2% of the vaccine group (107 of 755 treated) and 13.4% of the comparison group (51 of 380). The most common complaints with the vaccine were fever, pain at the injection site and fatigue.

A hazard ratio describes one trial's patients against that trial's comparison group. With different patients, a different comparison group, longer follow-up or another definition of an event, the same treatment can produce a different number. Doctors and analysts who have commented on the results disagree mainly about this point.

Adam May, who posts as @A_May_MD, wrote that 0.58 matched "EXACTLY what NEO-adjuvant treatment with pembro yielded against adjuvant pembro" in a trial called SWOG S1801. In that trial, published in the New England Journal of Medicine in 2023 by Sapna Patel and colleagues, patients with melanoma that could still be seen and removed received three doses of pembrolizumab before surgery and 15 after (154 patients), or surgery followed by 18 doses (159 patients). At two years, 72% of the first group had had no event, against 49% of the second. His argument is that giving pembrolizumab before surgery may now be the better standard, so a vaccine tested against pembrolizumab after surgery was compared with a weaker option. He urged the companies to test their vaccine "against the REAL standard of care now - NEOadjuvant immunotherapy."

The account @bioinvestor24 drew the opposite conclusion. It called an HR of 0.58 "very impressive" and pointed to KEYNOTE-054, the trial that established adjuvant pembrolizumab. In that study, published by Alexander Eggermont and colleagues in the New England Journal of Medicine in 2018, 1,019 patients with resected stage III melanoma received pembrolizumab (514) or a placebo (505). After a median follow-up of 15 months the hazard ratio for recurrence or death was 0.57. On this view, adding the vaccine to pembrolizumab lowered the risk by about as much as pembrolizumab once lowered it against a placebo.

Cole Smith, a pharmacist who posts as @iamcrsmith, objected to both comparisons. "You cannot directly compare hazard ratios," he wrote, because the trials "have different endpoints, follow up periods, inclusion/exclusion criteria, control-arm risks, and data maturity." S1801 measured event-free survival in patients whose tumors were still in place, with 313 patients in an open-label Phase 2 design, while INTerpath-001 enrolled people whose tumors were already gone. He argued that the two approaches are "complementary, not substitutes," and that "You cannot simply 'give a few doses of pembrolizumab before surgery' for a patient whose tumor has already been removed." His chief concern was practical, that the vaccine "requires adequate tumor tissue for sequencing and has a manufacturing lead time and high cost."

KEYNOTE-054 used stratification, which means randomizing patients separately within groups such as disease stage and region so that each arm gets a similar mix. S1801 stratified by stage and by the level of an enzyme in the blood. Stratification balances the arms within a trial, but it does nothing to make two different trials' patients alike, which is Mr Smith's point.

A further comparison runs between two stages of testing for the vaccine itself. Phase 2 trials are smaller and meant to show whether a drug is worth a full test, while Phase 3 trials are large and designed to settle the question for regulators. The Phase 2 trial, KEYNOTE-942, published in the Lancet in 2024 by Jeffrey Weber and colleagues, assigned 157 patients with resected stage IIIB to IV melanoma to the vaccine plus pembrolizumab (107) or pembrolizumab alone (50), without a placebo. Its hazard ratio for recurrence or death was 0.561, with a confidence interval from 0.309 to 1.017 that crossed 1. The Phase 3 trial added patients with stage II disease, whose cancers are less advanced.

Analysts at Barclays, a bank, compared the Phase 3 figures with the Phase 2 ones in a research note. According to the note, the Phase 3 hazard ratio for distant spread was 0.62, against 0.347 in Phase 2, and management attributes the difference to slower progression among patients with stage II disease. The note also puts the Phase 3 hazard ratio for overall survival at 0.77, against about 0.6 at 24 months in Phase 2, and Barclays regards that Phase 3 figure as probably immature. The companies' press release prints no overall-survival figure. Their statement says only that "Overall survival (OS) data are not yet mature, but a favorable trend in OS with intismeran plus KEYTRUDA has been observed."

Maturity matters because of how trials are analyzed. An interim analysis is a look at the data taken partway through a trial, at a point fixed in advance. The companies' August announcement described the first positive result as coming from "a pre-specified interim analysis," and the study continues in order to measure survival. When few patients have died, a survival hazard ratio is calculated from a small number of events and can change a great deal as more occur.

The companies' statement also reports a second kind of analysis. A piecewise hazard ratio splits follow-up into periods and calculates a separate ratio for each. Someone might report one when a treatment is expected to act late. Intismeran starts only once it has been made, and is given every three weeks for up to nine doses. In the first six months the hazard ratio for recurrence or death was 0.99 (0.66 to 1.49), close to no difference. After six months, in a pre-specified sensitivity analysis, it was 0.42 (0.30 to 0.58), and for distant spread or death 0.50 (0.34 to 0.73). The primary result is still the whole-trial figure of 0.58.

The full presentation is due at ESMO on October 24th, and the companies say the results have been accepted for publication in the New England Journal of Medicine. The paper should give the number of recurrences and deaths in each arm, which will show how many events the survival trend is based on. Later data cuts will show whether the gap in distant spread and death widens, holds or closes as follow-up lengthens. Whether the vaccine does better than pembrolizumab given before surgery is a separate question, because INTerpath-001 did not test that comparison.