Some Diagnostic Plots and Corrective Adjustments for the Proportional Hazards Regression Model

摘要

In this article, we present a few graphical procedures for the proportional hazards regression model. The focus is on diagnosing the validity of the proportional hazards assumption, irrespective of the validity of other related assumptions (such as additivity/linearity of the covariate effects). If the proportional hazards assumption does not hold, then it may still be possible to use a modified version of this model after suitable adjustment. We present further diagnostic plots for specifically detecting the possibility of a covariate effect being in the form of a scale change of time (rather than a scale factor of the hazard rate). We also propose calibration of the plots through asymptotic confidence limits. Subsequently, we explore simple and intuitive methods of adjusting for the nonproportional effect of a single covariate. The performance of the proposed methods is studied via Monte Carlo simulations. The procedures are then illustrated through the analysis of a dataset. Online supplemental materials for this article include the dataset, its description, SAS programs for data analysis, and an Appendix containing variance calculations needed for calibration of the plots.