A personalized cancer neoantigen vaccine trial tests a selection procedure rather than a compound: each patient's product is assembled from mutant peptides chosen by software from that patient's own tumour. Whether that procedure chooses well is an empirical question, and the phrase "immunogenicity was reported" denotes at least four different quantities which the field applies to all of them interchangeably. We introduce the Immunogenicity Reporting Resolution scale, a four-level taxonomy defined before any result was examined, which classifies the resolution at which a publication reports vaccine-induced immunogenicity: whether any patient responded (L1), how many encoded peptides elicited a response (L2), which specific peptides elicited one (L3), and whether those identities relate to the clinical outcomes of the same patients (L4). We classified twelve individualized neoantigen therapy programmes, described in eighteen publication- and registry-level records, against it. Every programme reports at L1. Six of the eleven programmes with a patient-specific selection step report which individual neoantigens elicited a response; none reports an analysis relating those identities to the clinical outcomes of the same patients. Reporting depth does not improve as programmes mature but narrows, and it narrows independently in the two largest programmes in the set. Five of the eleven do not name the software that chose their contents, and four of the eighteen records state no numeric threshold for what counts as a response. The reported hit rates cannot be pooled across programmes: thresholds differ, denominators are published as caps rather than counts, and whether a response was measured directly from patient cells or after laboratory expansion changes the rate by more than a factor of two within a single matched cohort. One finding is structural rather than editorial. Every patient receives their own set of neoantigens and no patient receives a different set, so no trial in this literature contains a comparison group for the question of whether a different selection would have produced a different outcome. Each patient is, for that question, an arm of one. On 19 August 2026 Merck and Moderna announced that INTerpath-001 (NCT05933577) had met its primary endpoint without releasing effect estimates; full results are scheduled as abstract LBA1 at the ESMO Congress on 24 October 2026. This record, frozen and deposited twelve days before that presentation, states six predictions about it, with a seventh withdrawn before deposit and retained in place, each with a falsification condition fixed in advance — among them, at a predictive probability of 92.3%, that no analysis relating neoantigen identity to outcome will be presented.
No takes yet. Share an insight, caveat, or question.
Quraishi et al. (2026) studied this question.