Goodness of fit methods for the proportional hazards regression model: a review.

The proportional hazards regression model is now used extensively in survival time studies. Although the model (hazard function, lambda (t;x) = lambda 0 (t)e beta' x) can be considered to be partially distribution-free, the following assumptions regarding the distribution of survival time are made: 1. Hazard functions for any two individuals are proportional. Thus (Formula: see text) is independent of t. 2. Independent variables affect the log hazard in an additive way (Formula: see text) This paper reviews methods available for assessing these assumptions in particular cases.