Testing for the proportionality assumption in Cox model

[Navid Asgari]

Hi,

I am using Cox proportional hazard model. I want to test the proportional-hazard assumption in the model. I referred to the book "An Introduction to Survival Analysis Using Stata" (Cleves, et al., 2010) which is published by Stata Press. The book gives several examples, mostly with binary covariates.

For example, in the hip fracture data (page 141), the authors compare the hazard functions of two groups (protect==1 & protect==0)

use http://www.stata-press.com/data/cggm3/hip2, clear
stcox protect
stcurve, hazard at1(protect=0) at2(protect=1) kernel(gaussian) width(4) yscal(log)

The authors then suggest that the fact that the two smoothed curves are parallel implies proportionality in the Cox model. This seems pretty straightforward. But, what if all covariates in the model are continuous and there is no binary variable? How would one use this test to show that Cox model is the right model to use for his/her dataset?

Thanks,
Navid

[Carlo Lazzaro]

Navid:
have you taken a look at -Schoenfeld residuals- under -stcox postestimation- entry in Stata .pdf manual?

[Steve Samuels]

In the Help window for stcox, at the upper right is an "Also See" dropdown menu. One of the entries is: "[ST]stcox PH-assumption tests". This is also the title of the corresponding section in the ST Manual (p. 101 for the Version 14 manual.