Advanced Statistical and Graphical features of SAS® PHREG

ABSTRACT Survival analysis involves the modeling of time-to-event data whereby death or failure is considered an "event". The graphic presentation of Cox proportional hazards model using SAS PHREG is a significant tool which facilitates effective data exploration in survival analysis. The SAS PROC PHREG can generate some of the useful survival analysis plots using the ODS graphics option in version 9.1.3. In this paper, we will demonstrate the advanced features of PHREG for investigating the cumulative martingale residual plots and for selecting best candidate models in model selection. In clinical trials, potential outlier individuals who ‘died far too early’ or ‘lived far too long’ are identified and compared to what the fitted model predicts. The cumulative residuals from PROC PHREG are used to investigate the model specification error of covariate and validate the proportion hazard function. Methods to identify outliers are commonly based on Cox regression residuals such as martingale and deviance residuals. We will use PROC GPLOT in SAS/GRAPH to generate these two residual plots and to detect influential outliers. We will outline a method to perform all possible subset model selection within user-defined subsets using AIC information criterion. Also we will discus the new and improved features of the BPHREG, an experimental upgrade to PHREG procedure that has some user-friendly options such as ‘class’, the ‘hazards ratio’, and ‘strata’ statements which can be used to fit Cox proportional hazards model more efficiently.