Generating Survival Data in the Simulation Studies of Cox Model

The Cox proportional hazards regression (Cox model) is commonly used to model survival data as a function of covariates. Simulation studies, which use computer intensive procedures to assess the performance of a variety of statistical methods, are increasingly employed in evaluating the properties of the Cox model. The generation of survival data is the most fundamental and important component in the simulation study. However, few published studies provided sufficient details to allow readers to understand fully all the processes to generate the survival data. Furthermore, the generation of the survival data was inappropriate in some situations for the simulation studies, in terms of the choices of distributions and parameters of the models. This paper details the important considerations necessary when generating the survival data, including generating survival time with specified baseline hazard, generating censoring time with specific distribution and fixed censoring proportion. We take a data set with two covariates as an example to illustrate the proposed techniques, and full details of how the study will be performed are given as well. In addition, the SAS Macro program for this example is presented, which is universal for other situations by modifying the values of some parameters. The study might encourage more sound and reliable simulation studies to be performed and reported with credible results.