Statistical Inference for Vaccine Efficacy: A Re-Randomization Procedure to Analyse Poisson Outcomes under Covariate-Adaptive Randomization

Abstract

Re-randomization inference is used as an alternative approach to more traditional statistical methods. Free from parametric assumptions, its use is particularly suited for studies incorporating covariate-adaptive randomization, and can provide additional evaluation and confirmation of inferences drawn from the original analyses, for example, as a sensitivity analysis. We discuss methodological and computational aspects in the context of a Poisson regression and describe an approach to re-randomization inference. This is tested in a simulation study and then illustrated in a case study in which we evaluate vaccine efficacy data from a previously published influenza vaccine study. Our simulations indicate that re-randomization inference corrects for model misspecification. The case study, which accounted for the minimization factors, shows that the p-value and confidence limits from re-randomization inference agree with the original analysis. In conclusion re-randomization inference is a useful method that can be used to support vaccine clinical development.