1
$\begingroup$

Since the analysis is for a rare event, it is possible that some small strata have no event. Hard to predict. Just wondering if there will be any problem. Cox PH model will have problem for sure, but I am not sure about stratified log rank test. The purpose of "stratified" is to adjust for baseline sex and age. We are not interested in the sex and age effect.

$\endgroup$

1 Answer 1

3
$\begingroup$

Neither the stratified Cox PH model nor the stratified logrank test will have a problem. Strata with no event will contribute no information, and the baseline hazard will not be estimable, but $\hat\beta$ and the logrank test are fine.

(This is assuming you mean 'stratified' in the sense of the stratified Cox model, not in the sense of getting separate estimates in each stratum)

$\endgroup$
2
  • $\begingroup$ I see! thanks! I was thinking about including include sex and age as covariate in the Cox model, which will cause problem. Even for stratified Cox PH model, I thought there will be a problem for estimation? You can't fit a Cox PH to no events within stratum. Or would that stratum be dropped automatically? $\endgroup$
    – hehe
    Commented Jan 31, 2022 at 21:56
  • 1
    $\begingroup$ That stratum would not contribute to the partial likelihood/logrank statistic, but that's estimated across all the strata rather than per stratum. Computationally, the stratum might be dropped out in advance or not. It's not for survdiff in R. $\endgroup$ Commented Jan 31, 2022 at 22:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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