Two mRNA cancer vaccine Phase 3 trials, one win and one loss: the difference is trial design, not the platform

Reading note: Statements marked [Fact] come from the linked public sources; statements marked [Inference] are the author's judgment. Data as of 2026-10-18 (Taipei time). Figures that the companies have not disclosed and that this article could not find are marked "not found"; nothing is estimated.

One-line conclusion

[Fact] INTerpath-001 met its RFS primary endpoint and AHEAD-MERIT's PFS was not statistically significant (Merck PR, BioNTech PR). [Inference] The outcome was decided mainly by disease stage, antigen strategy, and endpoint choice, not by the mRNA platform itself. The former ran in the post-surgical adjuvant setting, when tumor burden is lowest, with personalized neoantigens and a double-blind design; the latter ran in first-line metastatic disease, with fixed HPV16 E6/E7 antigens, open-label, and with PFS as one of two dual primary endpoints. [Inference] For the next vaccine trial, the question is not "should we use mRNA" but "at which disease stage, with which antigens, against what comparator, and with which endpoint."

1. What happened

Both trials added an mRNA vaccine to pembrolizumab (Keytruda), and both control arms were pembrolizumab monotherapy. The similarities end there; population, antigens and endpoints differ almost entirely. The table compares the published designs and results.

Item	AHEAD-MERIT (BNT113, BioNTech)	INTerpath-001 (intismeran / V940 / mRNA-4157)
Trial ID	NCT04534205	NCT05933577
Design	Phase 2/3, open-label, two arms; Part A safety run-in, Part B randomized (BioNTech PR)	Global Phase 3, randomized, double-blind, placebo- and active-controlled (Merck PR)
Population	First-line unresectable recurrent or metastatic HPV16+ head and neck squamous cell carcinoma, PD-L1 CPS ≥1 (22C3 assay) (CT.gov)	Completely resected stage IIB–IV cutaneous melanoma, no prior systemic therapy (Merck PR)
Enrollment	Actual enrollment 358 (including Part A; Part B number not found)	1,137, randomized 2:1; CT.gov still lists an estimated 1,089, not yet updated (CT.gov)
Antigens	Fixed HPV16 E6/E7 oncoprotein antigens (Fierce)	Designed from each tumor's mutations, up to 34 neoantigens
Dose	BNT113 dose and schedule not found	intismeran 1 mg every 3 weeks, up to 9 doses, with Keytruda 400 mg every 6 weeks for about one year
Primary endpoints	Dual primary: PFS (RECIST 1.1, blinded independent central review, BICR) and OS (CT.gov)	RFS (investigator-assessed); DMFS key secondary, OS secondary
Result	PFS not statistically significant; IDMC recommended continuing to follow OS. Under the prespecified statistical design, a significant OS alone would make the trial a success (BioNTech PR)	At a prespecified interim analysis, both RFS and DMFS were statistically significant and clinically meaningful (Merck PR)
HR / 95% CI / p value	Not found: no HR, CI, medians, event counts or p value disclosed, and no statement of interim vs final analysis (MGS summary, secondary source)	Not found: Phase 3 HR scheduled for ESMO 2026 LBA1 (10/24, Presidential Symposium)
Safety	Safety data from the randomized part not disclosed	"Consistent with previous reports, no new safety signals" (no figures given)

The most important cell in this table is the HR: neither trial has yet disclosed a Phase 3 effect size. Any judgment here about "by how much" is inference until the full data are out.

Readers should also watch for a common confusion. The HR 0.51 (95% CI 0.294–0.887) cited in many reports is the five-year RFS from the Phase 2b KEYNOTE-942; the same trial's DMFS was HR 0.411 (0.200–0.843). Neither is an INTerpath-001 result (Merck PR). KEYNOTE-942's primary analysis was n=157, HR 0.561 (0.309–1.017), P=0.0266 (KOL Pulse transcription of ASCO 2026 slides; secondary source, not checked against the original slides).

The market reaction was clearer: on the day of the news, BNT fell premarket from $95.84 to $91.32, about -5%, and Moderna about -3% (Fierce).

2. Why one succeeded and one failed

With no effect sizes yet, this section covers structural reasons. Unless noted otherwise, everything below is [Inference].

The key difference is tumor burden. A vaccine must prime T cells, expand them and get them into the tumor, which takes weeks. With only minimal residual disease after surgery, the immune response can keep pace; in metastatic disease, high tumor burden and a suppressive microenvironment make it hard to separate the curves within a PFS window. BioNTech itself says the immune effect "deepens over time" ([Fact], Fierce); read the other way, that statement says PFS was never a suitable primary endpoint for this class.

Antigen strategy widened the gap. HPV16 E6/E7 are viral antigens with good theoretical immunogenicity, but they are only two fixed targets, and tumors can escape through HLA loss or antigen-presentation defects. A personalized vaccine hitting up to 34 neoantigens at once is harder to escape completely, and melanoma's high tumor mutational burden supplies ample neoantigens.

The control-arm baselines differ. Both are add-on designs, but first-line pembrolizumab monotherapy already performs reasonably in HPV+ oropharyngeal cancer, so an added drug must add a lot to be visible. More importantly, in add-on trials PFS does not predict OS: adding lenvatinib in LEAP-010 improved PFS but not OS (MGS, secondary source). The KEYNOTE-048 pattern BioNTech is counting on, "PFS missed, OS hit," compared immunotherapy with chemotherapy and cannot be applied directly to AHEAD-MERIT, where both arms receive pembrolizumab.

The statistical design reduced AHEAD-MERIT's power. Alpha was split between PFS and OS, but the split was not disclosed (MGS, secondary source). INTerpath-001 used a single RFS primary endpoint, 2:1 randomization and n=1,137, and hit at an interim analysis, implying a sizable effect; that remains inference until ESMO.

Each has a blinding weakness. AHEAD-MERIT was open-label, but PFS was read by BICR, so bias control was acceptable. INTerpath-001 was double-blind, but RFS was investigator-assessed; FDA may ask for BICR or sensitivity analyses.

Development path is the biggest divergence. BNT113's safety run-in had only 15 patients, single-arm, with an unconfirmed ORR of 40% (MGS, secondary source), and that signal did not translate into a randomized PFS benefit. intismeran went through the randomized Phase 2b KEYNOTE-942 before Phase 3.

The remaining factors are either undeterminable now or minor. BNT113 dose-finding data were not found, so a dose problem cannot be excluded. Regionally, AHEAD-MERIT's Asian sites were in South Korea and Taiwan, not Japan (MGS, secondary source); enrollment in China for either trial was not found. Operationally, a personalized vaccine requires sequencing and custom manufacturing, leaving a gap between surgery and first dose; the adjuvant setting buffers that gap, while first-line metastatic patients cannot wait, so the platform was never a good fit there.

3. Regulatory and commercial implications

FDA. [Fact] Merck and Moderna said they will discuss filing with regulators (Merck PR). [Inference] RFS has review precedent in adjuvant melanoma (KEYNOTE-054 and KEYNOTE-716), so immature OS should not block approval. The real new question is CMC: each batch of a personalized product is different, and release criteria, manufacturing turnaround and central sequencing have no review precedent.

NMPA. Whether intismeran enrolled in China or has a filing plan: not found. [Inference] Filing on global data would face practical hurdles in the personalized supply chain and human genetic resources rules (both tumor and normal tissue are sequenced).

Next steps for BNT113. After the PFS miss, only OS remains, and BMO's confidence in OS success has "significantly" declined (Fierce). [Inference] Even if OS succeeds, a negative PFS will weaken label claims and physician uptake.

Competitive landscape. [Inference] INT plus PD-1 may become the new standard of care in high-risk resected melanoma; followers (including BioNTech's own autogene cevumeran) will need randomized data showing superiority over PD-1 alone, and off-the-shelf fixed-antigen vaccines will be expected to complete a randomized Phase 2 before Phase 3.

For the next trial. [Inference] INTerpath-001's review will become the first reference case for personalized-product CMC; sponsors planning cancer vaccine trials should confirm release criteria and manufacturing turnaround expectations with FDA at the pre-IND stage.

Merck's genitourinary (GU) footprint. [Fact] The INTerpath program has 9 Phase 2/3 trials across NSCLC, bladder cancer and renal cell carcinoma (Merck PR). INTerpath-004 is a triple-blind Phase 2 adjuvant RCC trial, planned 272 patients, primary endpoint DFS, primary completion expected 2028-01 (NCT06307431); INTerpath-005 is a Phase 1/2 trial in muscle-invasive urothelial carcinoma (MIUC), including arms with or without enfortumab vedotin (EV), primary completion expected 2027-04 (NCT06305767). These two trials are the benchmarks for the GU recommendations in the next section.

4. Lessons for the next trial design: OncoLattice recommendations

Everything below is [Inference] or recommendation, organized by planning, enrollment and sites.

Planning

First, place vaccine-type agents preferentially in minimal residual disease settings: post-surgical adjuvant, or ctDNA-positive without radiographic disease. Avoid first-line metastatic settings with PFS as the primary endpoint.

Second, always run a randomized Phase 2 first, with the same control as Phase 3 (PD-1 monotherapy plus placebo) and a prespecified go/no-go threshold, e.g. a Phase 2 HR upper bound ≤0.7. Do not go straight to Phase 3 on a single-arm ORR from n=15.

Third, on endpoints, use RFS/DFS in the adjuvant setting, ideally BICR-assessed or with prespecified sensitivity analyses. If dual primary endpoints are unavoidable, disclose the alpha allocation and ensure OS is adequately powered; statistical models should account for delayed effects, e.g. piecewise HR or RMST.

Fourth, prespecify biomarkers: TMB or neoantigen count, HLA LOH, ctDNA clearance, T-cell response (ELISpot, TCR sequencing), as stratification factors or exploratory endpoints.

Enrollment

For personalized products, write the surgery-to-first-dose interval into eligibility and manage central sequencing and manufacturing timelines; for example, randomization ≤12–16 weeks after surgery, as in the INTerpath-005 TiP (Ann Oncol). Selecting by ctDNA or high-risk pathology (ypT2–4/ypN+) raises the event rate, which can shrink sample size and shorten follow-up.

Sites

Prefer sites with mature pathology workflows: reliable FFPE and matched normal tissue, tumor-informed ctDNA, and smooth post-surgical referral. Urology-led sites have a natural advantage because follow-up after cystectomy and nephrectomy is already routine. Taiwan and South Korea have experience with such trials (AHEAD-MERIT had sites in both); entering China requires early assessment of human genetic resources rules and tissue export restrictions.

Direct GU implications

Adjuvant MIUC (ypT2–4/N+) and adjuvant intermediate-high-risk RCC (the KEYNOTE-564 population) are the settings closest to INTerpath-001. If sponsors are developing GU vaccines or immunotherapy add-ons, INTerpath-004 and INTerpath-005 are the benchmarks they will be compared against, and control arms should align with adjuvant pembrolizumab. In BCG-unresponsive NMIBC, locally delivered vaccine strategies (intravesical or intralesional) are worth considering, but randomized data are currently lacking.

5. Unknowns and what to watch

The nearest and most important milestone is ESMO LBA1 on 10/24: INTerpath-001's HR and CI, absolute RFS difference, subgroups (stage IIB/IIC vs III/IV, TMB), grade ≥3 adverse events, and whether OS is mature. AHEAD-MERIT's PFS HR and median, OS analysis timing and alpha allocation are all not found at present.

Three things to watch in the medium term: whether FDA requires a BICR analysis because RFS was investigator-assessed; how manufacturing turnaround and cost of the personalized product affect access; and the primary analyses of INTerpath-005 (2027-04) and INTerpath-004 (2028-01).

Evidence limitations

Two limitations should calibrate the reader's confidence. First, both Phase 3 trials have only press-release-level topline results, with no HR, CI, event counts or safety detail, so any discussion of effect size is inference. Second, some figures come from secondary summaries (MGS, KOL Pulse) not checked item by item against original slides or papers; each is flagged in the text. Sections 1 and 2 will need updating once the ESMO data are released.