Confidence Band for the Differences between Two Direct Adjusted Survival Curves

Abstract

This paper focuses on comparing the survival outcomes of two treatments while adjusting for multiple demographic and clinical factors. We wish to answer the question: "At what times do the survival outcomes of two treatments differ?". Previous work by Zhang and Klein (2001) addressed this issue by considering a stratified Cox model and by constructing the confidence band for the differences between survival functions of treatments, for a given subject. In this study, we utilize direct adjusted survivals of treatments, so that treatment comparison can be performed on a general basis. The confidence band for the differences between direct adjusted survivals was constructed using the simulation method. We developed two SAS macros that compute the band. The first macro was developed for data sets consisting of fixed covariates only. The second macro was developed based on the piecewise proportional hazards Cox model and allows for timedependent variables. We illustrate both macros by analyzing the survival data of a cohort of lymphoma patients.