03-31-2017 09:26 PM
I have a mortality outcome (yes/no) and a categorical predictor (white,hispanic,black, asian) I am running a logistic regression on.
I would like to see if some other factors, like age, type of med, etc. are risk factors for the death outcome. Would it be best to add to the model in a stepwise fashion? Would this answer my question, or would it be an interaction?
04-02-2017 11:04 AM
I could think many ways if you don't have a big table.
1) proc logistic
2) proc ADAPTIVEREG (good for nonlinear relationship)
3) proc genmod + repeat (gee model)
4) proc glimmix (generalize linear mixed model)
I would highly recommend to use proc glimmix.
04-04-2017 03:14 PM
If you fit a logistic model with all of the candidate predictors in PROC GENMOD and include the TYPE3 option, the table of type 3 tests gives you a test of the effect of each predictor after accounting for the others. For example:
class race medtype;
model outcome(event="yes") = race medtype age a b c d e / dist=bin type3;