Hi,
I'm using stepAIC with a CoxPH model on some data. As
we expect, there are a bunch of predictors, and stepAIC
throws some out and then converges on others.
Anyway, when I run cox.zph on the final model, I get
some evidence of Beta(t) time dependence. Should I
be concerned about the impact of this on the entire
fit procedure, or is it just that I can't trust the final
hazard ratio that the Cox model spits out?
Thanks in advance,
-Ekr