The advent and subsequent widespread availability of preventive vaccines has altered the course of public health in the twentieth century. In spite of the overall success, vaccines are still lacking for many high-burden diseases, including HIV. An important step in the process of developing effective vaccines is identifying immune responses that are indicative of protective efficacy. In this work, we use a causal inference framework to propose a new approach to studying immune responses in the context of vaccines. We focus on causal quantities defined by stochastic interventions, which may be more relevant than alternative approaches for describing the effects of immune responses on risk of infection or disease. We propose methodology for efficiently estimating these quantities using data generated by preventive vaccine trials with two-phase sampling of immune responses. We propose and evaluate two strategies for estimating these quantities: an inverse probability weighting-based method and an augmented method. The latter method is shown to be nonparametric efficient and multiply robust to misspecification of nuisance estimators. We also provide methods for constructing confidence intervals and hypothesis tests, and provide an open source software implementation of the proposed methodology. We illustrate the methods using data from a recent preventive HIV vaccine trial.

We focus on variable importance analysis in high-dimensional biological data sets with modest sample sizes, using semiparametric statistical models. We present a method that is robust in small samples, but does not rely on arbitrary parametric assumptions, in the context of studies of gene expression and environmental exposures. Such analyses are faced not only with issues of multiple testing, but also the problem of teasing out the associations of biological expression measures with exposure, among numerous confounds such as age, race, and smoking. Specifically, we propose the use of targeted minimum loss-based estimation, coupled with generalizations of moderated empirical Bayes statistics, to obtain estimates of variable importance measures. The result is a data-adaptive approach that can estimate individual associations in high-dimensional data, even in the presence of relatively small samples.