Identifying vaccine-mechanism bias in mathematical models of vaccine impact: the case of tuberculosis

Abstract

AbstractIn the development of vaccines against tuberculosis (TB), a number of factors represent burdensome difficulties for the design and interpretation of randomized control trials (RCTs) of vaccine efficacy. Among them, the complexity of the transmission chain of TB allows the co-existence of several routes to disease that can be observed within the populations from where vaccine efficacy trial participants are sampled. This makes it difficult to link trial-derived readouts of vaccine efficacy to specific vaccine mechanistic descriptions, since, intuitively, the same efficacy readouts may lean on the ability of a vaccine to arrest only some, but not all, the possible routes to disease. This increases uncertainty in evaluations of vaccine impact based on transmission models, since different vaccine descriptions of the same efficacy readout typically lead to different impact forecasts. In this work, we develop a Bayesian framework to evaluate the relative compatibility of different vaccine descriptions with the observations emanating from a randomized clinical trial (RCT) of vaccine efficacy, offering an unbiased framework to estimate vaccine impact even when the specific mechanisms of action of the given vaccine are not explicitly known. The type of RCTs considered here, conducted on IGRA+ individuals, emerged as a promising design architecture after the encouraging results reported for the vaccine M72/AS01Eclinical trial, which we use here as a case study.Authors summaryHere, we focus on a problem that is pervasive in mathematical modeling of vaccines’ impact, consisting of the existence of a multiplicity of vaccine parametrizations that are compatible with the result of a given clinical trial of vaccine efficacy. However, focusing on tuberculosis vaccines, we find that it is possible to use computational simulations and Bayesian statistics to assign these models with posterior probabilities measuring their relative compatibility with the results of a real clinical trial under analysis. The framework presented unlocks the production of unbiased, mechanism-agnostic impact forecasts for vaccines against tuberculosis, and can be extended to the study of vaccines against other communicable diseases with a complex infectious cycle.