0
$\begingroup$

In the context of a clinical trial with time-to-event primary endpoint, if stratified randomization is used for subject enrollment, how should the stratification factors be accounted for in the analysis? In my understanding, it is common practice to use a stratified Cox PH model, with strata being the stratification factors used for randomization. However, in practice, I've also seen stratification factors being included as covariates in the Cox model. Are there any well-established rules or justifications on how stratification factors should be handled during analysis phase?

$\endgroup$

1 Answer 1

0
$\begingroup$

The stratified Cox is nearly always preferred to covariate adjustment. Covariate adjustment assumes that there is a "common" unknown baseline hazard function; but stratified Cox allows such baseline hazards to be different across strata. If interactions between strata and treatment are not included, covariate adjustment will have even more assumptions made.

In trials with sufficient sample size (e.g., a phase III confirmative trial), we should minimize the number of assumptions to improve the robustness of the conclusion. Even if the baseline hazards are nearly identical, with large enough sample size, the efficiency loss by using stratified Cox is ignorable for all practical purposes.

If sample size is not large, e.g., in early phase trials, covariate adjustment might be acceptable to improve efficiency, for the purpose of generating hypotheses for later trials to verify. But model diagnostics is always necessary to check assumptions.

New contributor
Zack Fisher is a new contributor to this site. Take care in asking for clarification, commenting, and answering. Check out our Code of Conduct.
$\endgroup$

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.