Dear Statalisters,
I am a novice in multilevel modelling; I am trying to assess the effect
of a preventive intervention on mortality data in a cluster randomised
trial. The randomisation units were clinics in which my patients are
nested. I am trying to assess the efcct of the intervention on
mortality.
I originally did a Cox regression using the following command:
xi: stcox i.randomgroup, cluster(clinic).
After reading the section about survival data in the book Multilevel and
Longitudinal Modelling Using Stata by Sophia Rabe-Hesketh. I am now
confused as to which command or function to use:
Should I stick to my first analytical approach or use one of the
following commands with the cloglog functions?
cloglog event randomgroup yearsfollowup, eform vce(cluster clinic)
xtcloglog event randomgroup yearsfollowup, eform i(clinic)
gllamm event randomgroup yearsfollowup randomnum, eform i(clinic)
link(cll) family(binom) adapt
These commands gave me results that are different from Cox analysis. The
results of xtclolog and gllamm with the cll link are similar. I
understand that these commands are equivalent and allow the introduction
of a random intercept for clinic. I also understand that the
complementary log log link function follows if a proportional hazard
model holds in continuous time and the survival times are interval
censored.
I am tempted to believe that my data is continuous-time data, but it is
not interval censored as I know precisely the time to failure. Can I
still use the command with the complementary log log function? If yes
how to I include the time to failure in the model? I hope I have not
completely mess up my model with years of follow-up to replace censoring
interval. I hope my questions make sense.
I look forward to hearing from you.
Thanks
Justin B.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
