Testing the linearity assumption in a Cox model

[Bill McMillan]

Hello,

I am running a Cox model to examine the effect of a treatment ("treat" - a binary variable) on the risk of death. One of the predictors is a continuous variable (age). However, I do not know how to test if the linearity assumption for age is fulfilled. I have tried doing a search, but without luck. This is my model:
stset t2dth, failure(c4dth==1)
stcox treat age, efron


Thanks a lot in advance.

Bill

[Carlo Lazzaro]

Bill:
welcome to this forum.
You may want to take a look at cox model - Cox PH linearity assumption: reading martingal residual plots - Cross Validated (stackexchange.com).
That said, you may want to check non-linearity assumption via:

stcox treat c.age##c.age, efron

[Bill McMillan]

Thank you so much for your shift response! I have now made a figure where I have plotted the observed cumulative martingale residuals against the values of the explanatory variable.
In addition, I have run the code provided by you. The P-value for the ineraction term between age is highly significant - does that mean that the linearity assumption is not met?

[Carlo Lazzaro]

Bill:
yes, it means that (as it is often the case with -age-).