ABSTRACT
The assumption of proportional hazards features prominently in modelling survival data. Although methods for assessing the validity of this assumption have been considered in earlier chapters, some additional methods are described in this chapter. These include stratified proportional hazards models, where the baseline hazard function differs between groups, and a piecewise Cox model to allow for non-proportional hazards between treatments. Restricted mean survival as a summary measure of a treatment effect, and the use of pseudo-values to model the dependence of restricted mean survival on explanatory variables is described. The important application of non-proportional hazards modelling to the comparison of survival rates between institutions is discussed, and techniques that can be used in this context are presented. Use of these methods in comparing the performance of health care providers is illustrated in a comprehensive example.
