Novel data-adaptive multivariate testing procedures, with applications to HIV research

Abstract

Project 1: Construct a generic data-adaptive framework for multivariate point null testing.Specifically, a data-driven optimality criterion is proposed for selecting among a large collection of candidate test statistics. This framework can be applied in a wide array of problems. It is illustrated on data from HVTN 505, a phase IIB HIV vaccine efficacy trial. Project 2: Extend the framework developed in Aim 1 for testing a functional null hypothesis.The test described in this framework connects to the test in Aim 1 by projecting the function estimator into a finite dimensional vector space using a finite collection of the coefficients from the Fourier transformation. We also provide arguments that the described test can consider more coefficients as sample size grows while still maintaining desirable testing properties. Project 3: Develop novel methodology for estimating open-label effectiveness for trials inwhich many or all study participants have switched off the placebo arm of the trial. The method developed accounts for changes in the population’s distribution of baseline characteristics and adherence behavior. The method developed is used to estimate the open-label effectiveness of a vaginal ring containing dapivirine using data from the MTN-20 (ASPIRE) and MTN-25 (HOPE) clinical trials.