Applied Missing Data Analysis in the Health Sciences

Chapter 6Survival Analysis under Ignorable Missingness

6.1 Overview

We introduced the cross-sectional and longitudinal studies in Chapters 4 and 5, where the outcomes are completely observed or missing. In survival analysis, the outcome of interest, the failure time, may be censored such that we may observe either the failure time or the censoring time (see the Pathways example in Section 1.3.6). This makes this chapter different from Chapters 4 and 5. One of the basic tasks of survival analysis is to model the relationship between failure time and covariates. Due to the expensive examination fee or other reasons, some of the covariates might not be observed. In this chapter, we provide methods for survival analysis when the covariates are possibly missing. We mainly focus on the Cox proportional hazards model because it is the most popular model in survival analysis and most of the literature on survival analysis with missing covariates focused on it. We also discuss the semiparametric transformation regression model, a more general class that includes the Cox proportional hazards model as a special case.

This chapter is organized as follows. In Section 6.2, we introduce some commonly used notations and review the Cox regression when covariate data are completely observed. Then several methods are offered. The enhanced complete-case analysis is introduced in Section 6.3, and the weighted methods are given in Section 6.4. Imputation methods are provided in Section 6.5. The ...