Assessment of Immune Correlates of Protection via Controlled Vaccine Efficacy and Controlled Risk

Abstract

Immune correlates of protection (CoPs) are immunologic biomarkers accepted as a surrogate for an infectious disease clinical endpoint and thus can be used for traditional or provisional vaccine approval. To study CoPs in randomized, placebo-controlled trials, correlates of risk (CoRs) are first assessed in vaccine recipients. This analysis does not assess causation, as a CoR may fail to be a CoP. We propose a causal CoP analysis that estimates the controlled vaccine efficacy curve across biomarker levels $s$, $CVE(s)$, equal to one minus the ratio of the controlled-risk curve $r_C(s)$ at $s$ and placebo risk, where $r_C(s)$ is causal risk if all participants are assigned vaccine and the biomarker is set to $s$. The criterion for a useful CoP is wide variability of $CVE(s)$ in $s$. Moreover, estimation of $r_C(s)$ is of interest in itself, especially in studies without a placebo arm. For estimation of $r_C(s)$, measured confounders can be adjusted for by any regression method that accommodates missing biomarkers, to which we add sensitivity analysis to quantify robustness of CoP evidence to unmeasured confounding. Application to two harmonized phase 3 trials supports that 50% neutralizing antibody titer has value as a controlled vaccine efficacy CoP for virologically confirmed dengue (VCD): in CYD14 the point estimate (95% confidence interval) for $CVE(s)$ accounting for measured confounders and building in conservative margin for unmeasured confounding increases from 29.6% (95% CI 3.5 to 45.9) at titer 1:36 to 78.5% (95% CI 67.9 to 86.8) at titer 1:1200; these estimates are 17.4% (95% CI -14.4 to 36.5) and 84.5% (95% CI 79.6 to 89.1) for CYD15.