Risk Evaluation after Heart Valve Replacement by Sas Proc Phreg

Identification of long-term risk factors after heart valve replacement can be quite challenging due to the nature of the covariates. Two specific factors need to be considered : (1) time-dependent covariates, and (2) the validity of the proportional hazard model. PHREG in SAS® is a very convenient procedure for solving these problems. In addition to those commonly encountered problems such as variable selection and influence assessment, we will use a data set from our St. Jude Medical valve prothesis database to illustrate how to use DATA step statement to incorporate time-dependent covariates into the proportional hazard model and how to plot martingale residuals to check model validity. Results from our analysis indicated that cardiopulmonary bypass time during surgery should be treated as time-dependent because its adverse effect faded away after six months. The proportional hazard model fitted this data set reasonably well once the effect of timedependence was taken into consideration.