I'm running a prospective cohort study in which I'm examining depression and dementia outcomes in nursing home residents from three facilities -- one experimental and the other two controls, with residents across facilities matched on demographic and clinical factors at baseline. Their depression and dementia outcomes were measured at baseline and two follow-up points. Many residents at one of the three facilities died before the final follow up point and I don't want to lose their earlier cases. This seems like a good situation for a mixed model rather than repeated measures. How would I structure it? Also, is there a way to treat the data as triplets, or do I need to compute the difference between experimental and control home for each control site? I use SAS and Stata, btw.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
It seems like you want to fit a multilevel model with observation at Level 1 (where time-point can be a predictor), person at Level 2 (where baseline or demographic characteristics can be a predictor), and facility at Level 3 (where control or experimental can be a predictor).
Multilevel models can handle missing data, so you won't lose the earlier cases if residents pass away.
I do not use SAS or Stata, so I'm afraid I cannot help you there. But the lme4 package in R is user-friendly, and many guides exist (I like this one in particular). You could also browse Stack Overflow in the SAS and Stata tags to see how people fit three-level multilevel models in those programs.
answered May 31, 2017 at 2:27