Design and Analysis of Nested Case-Control Studies in the Presence of a Terminal Event

Abstract

Various methods are available for analyzing data in which multiple outcomes, including a terminal event, are of interest. For example, under the semi-competing risks setting, one may fit an illness-death model to cohort data in order to investigate the association between some covariate of interest and the risk of both non-terminal and terminal events. Unfortunately, such covariates may be expensive or otherwise difficult to obtain, and cannot be ascertained for all members of the cohort. When only a single outcome is of interest, investigators faced with this problem often choose to perform a nested case-control (NCC) study, a cost-efficient design in which covariate information is ascertained for all cases of this outcome as well as a random sample of controls chosen via risk set sampling. However, previously developed methods for analyzing NCC studies have only considered univariate outcomes. This dissertation develops methodology based on inverse probability weighting that enables NCC studies to be used to answer scientific questions about multiple outcomes simultaneously, while additionally investigating novel design possibilities and their operating characteristics. Chapter 1 proposes estimation and inference for the illness-death model for semi-competing risks data arising from an existing NCC study and introduces the supplemented NCC study design. Chapter 2 extends these ideas to the joint frailty model for recurrent events in the presence of a terminal event, while proposing a general framework for the design of NCC studies in this setting. Chapter 3 returns to the semi-competing risks context, generalizing the NCC design to allow subsampling of both non-terminal and terminal events.