1
$\begingroup$

I'm a statistics newbie (medical student) trying my luck with a Cox regression for a survival analysis on the outcome of a specific type of operation. And I'm trying to determine which variables to control for. And how to handle the age of the patient at operation-time (measure-start)...

Currently I've set the following as covariates in my analysis (PASW/SPSS):

Risk (Low/High)
Gender (Female/Male)
Shunt used in operation (False/True)
Artery stenosis (False/True)
Age at operation (Integer value)

I've set the 4 first variables as categories (dummy coded as 0 and 1), and specified the first value (0) as the reference value.

But what about the patients age at the operation? I assume this must be controlled for? But the age at measure-start (operation) isn't time-dependent (is it?), but are always the same value, right?

Does this mean I have to control for the patients real age aswell? And does this mean I have to use the COX with Time-dependent Covariables function instead of the regular COX Regression?

Thanks for any help!

Regards Alex

$\endgroup$
3
$\begingroup$

You are right that you don't need to use a time-dependent model for this situation. If, however, age would be expected to have a dominant effect (as might be expected), you may be safer using an interval-censored model. With R, that is setup with three arguments to the Surv object, but I don't have any experience with SPSS's facilities for survival analysis.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Can I acheive this interval-censored model by calculating a new variable as the LN(age)? I remember reading about that somewhere... $\endgroup$ – Alex Jan 10 '11 at 10:46
  • $\begingroup$ Interesting suggestion, to which I have not had prior exposure. I am reasonably sure that would not achieve the precise goal I had in mind. It might achieve similar objectives by dampening the effect of age, though. The effect of age on the hazard scale would now only be linear rather than exponential. Whether that would correspond to the "true" age effect on the response under study would still be open to question. $\endgroup$ – DWin Jan 10 '11 at 15:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.