[R] Fitting additional variables to model

[Min-Han Tan]

Good morning,

Sorry to trouble the list. I'm working on Cox models of survival, and
am encountering a problem. I'm trying to group variables into some
kind of new staging system  By grouping, I mean : so-called
'integrated staging systems' for cancer merge categories of variables
such as tumor stage, patient status into a single range of categories
e.g.
System I = Stage I or II, Patient Status 0
System II = Stage I or II, Patient Status 1
     OR
     Stage III, Patient Status 0
So in this example, Stage I + II are grouped together, probably based
on outcome.


So in the scenario where each of the initial 2 variables A and B
involved in the model have 4 categories:

1. Is there any other way to obtain a grouping the variables by
outcome besides examining all possible 16 Kaplan Meier curves
concurrently, and seeing how they group? Would it make sense to run
pairwise survfits - but if so, what happens when more variables are
introduced into the equation? Finally, is it possible to execute this
in R? Thanks!!!

2. If there is a significant interaction between these 2 terms (A*B),
does it even make sense to ask how I can perform "grouping" of the
variables?

Thanks in advance!

Min-Han