Analysis of time-dependent covariates in failure time data

Abstract

In failure time analyses, time-dependent covariates are only rarely used. In some clinical studies, however, consideration of available covariate information over time could be relevant to understanding complex disease processes. We propose the time-dependent Cox model and the linear model of Aalen as two possible approaches for such time-dependent survival analyses. The approaches are illustrated with the data of the Stanford Heart Transplantation Study and a study of malignant glioma. Differences between these models and the baseline analysis are discussed.