0

I'm working on heart-transplant data for which I have to build a Cox mixed model in R. As it is designed for this purpose, I chose to use the coxme package but I'm having much trouble to select my model. I want to use a forward procedure but it is non-applicable to coxme objects. I asked another question here whose answers allowed me to access to step()'s code to modify it so it is applicable to coxme models. However, this is very difficult (for me at least) as the function is quite complicated (it is definitely not just about modifying exctractAIC(), there are many other functions to adapt).

So here are my questions :

1) Is there a simpler way to run a forward on coxme models ?

2) Should I use another package ?

Backlin
  • 14,612
  • 2
  • 49
  • 81
Vincent
  • 955
  • 2
  • 15
  • 32
  • Thanks for editing, I'm new here and i'm not used to the formatting yet. This is much nicer to read now ! :) – Vincent Jun 27 '13 at 14:36

1 Answers1

0

After having tried for quite long to adapt step(), I decided to just create a simplfied -but exact- forward function from the beginning, which was way simpler. So if someone wonder the same question as me, I really recomend him :

1) Not being afraid of coding the whole function from the begining ! (80% of the original function is not usefull for a forward procedure and is not to be recoded)

2) Using the coxme package.

However, if you do so, here is one thing that might prevent confusions : what is called AIC/BIC in coxme models is in fact the improve of the tested model compared to the null model. That means you should maximise it and not minimize it if you decide to use it (you can change the calculation to the real AIC/BIC one if you prefer).

Vincent
  • 955
  • 2
  • 15
  • 32