I am currently using the coxph function from R on a microarray dataset.
The curve fit is performed using each gene's expression measurements
separately. I then get a p-value for each gene, from the Coxph.summary
method. Can someone please explain how this is computed and the
significance of the p-value?
However, the bigger question I have is how to perform permutation tests
to get a more robust estimate of significance. After permuting the
survival times and running Cox on each gene, I would like to save the
most extreme values (representing the gene most correlated with
survival). After ~1000 runs, I then run Cox on each gene with the
correct survival numbers. My question is, which statistic(s) should I
look at? I was thinking that the coefficient might be appropriate, but
I'm new to survival analysis and am not certain. Can anyone offer any
advice?
Finally, currently, I permute the survival times and maintain the proper
association with the status column. Should I permute the status separately?
Thanks for your time.
