Cox Regression survival analysis with clustered data.

[nawar han]

Hello everyone,
I have a database of thousands of patients from many hospitals.
I'm trying to model a cox regression survival analysis. I would like to account for the clustering within hospitals using the hospital_ID variable. I tried using the option shared (hospital_ID ) but the issue was : delayed entries or gaps detected. Then I tried using the forceshared ( ) options. This seemed to exludes everyone from my model.

Does anyone have experience with survival analysis using clustered data?

thank you

[Steve Samuels]

Ignore the shared option and add the following option your stcox statement

vce(cluster hospital_ID)

That will make standard errors a function of between-hospital variation. However if this is a survey data set, then you should svyset the data properly, with whatever the actual PSUs were, then run svy: stcox.