Regression Analysis of Fold-Increase Endpoints Using a Distributional Approach for Paired Interval-Censored Antibody Data

Abstract

Biopharmaceutical research often uses a binary “fold-increase” response variable defined by the ratio of a pair of interval-censored measurements taken at baseline and end-of-study. Conventional practice ignores the interval-censoring nature of the data. Moreover, conventional practice dictates that the possible choices for cut-off to define the response must follow a geometric sequence. A novel method based on the “distributional approach” was proposed for the analysis of such paired measurements in a randomized trial context. The degree of fold-increase above which a response is defined can be chosen according to scientific rationale instead of being limited to a geometric sequence. The risk ratio is then estimated for comparison between trial arms. We extend the method to allow for adjustment for baseline covariates in both randomized trial and cohort study settings. The treatment effect is obtained by integrating over the covariate distribution. In the presence of heterogeneity, estimators of the population treatment effect and the average treatment effect are proposed and their performances are evaluated by simulation studies. We apply this method to analyze antibody data measured by the hemagglutination inhibition assay in an influenza study. Supplementary materials for this article are available online.