1
$\begingroup$

I have some pre-treatment variables which are significantly associated in a cox model with progression-free survival (PFS) in a medical study.

I was wondering if taking a defined cutpoint (for example 3 months or 6 months PFS) and classifying patients according to progression in these time period is possible to do a logistic regression with my pre-treatment variables and see the performance of the model in the prediction of 3 months or 6 months PFS.

Is possible/correct from a statistical point of view?

$\endgroup$
4
$\begingroup$

If your data and model fit the Cox proportional hazards assumption, then you are much better off using the Cox model to predict 3-month or 6-month PFS. The logistic model would throw away all information about actual event times and also ignore cases that were censored (lost to follow up) before 3 or 6 months, respectively. If there was no censoring before your time of interest and the proportional hazards assumption was substantially violated then I suppose you could consider logistic regression, but you would have to recognize that your results might be highly dependent on the particular time that you chose. No harm in trying the logistic regression, but I think most would prefer to see Cox regression results instead.

$\endgroup$

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.