Statistical methods for identifying surrogate endpoints in vaccine trials

Given a treatment and a clinical outcome, it is often of interest to seek a biomarker which can be measured shortly after the treatment is administered and which can be used to predict the clinical outcome reliably. Such a biomarker is often referred to as a surrogate endpoint. This work develops statistical methods for identifying surrogate endpoints, with a focus on techniques which are useful in the context of randomized vaccine trials where subjects may experience the clinical outcome before their biomarker value is measured. Using the framework of potential outcomes (counterfactuals), I introduce two causal estimands which give complementary information about the surrogate value of the biomarker. The effects of different sets of assumptions on the statistical identifiability of these estimands are described in detail. I propose and compare two estimation methods, and present a sensitivity analysis framework for assessing surrogate value. I also suggest a novel variable selection technique which may be particularly useful for identifying candidate surrogates from high-dimensional pre-clinical data.