Date:         Thu, 15 Sep 2005 16:25:09 +0200
Reply-To:     Lukas <mutsikos@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Lukas <mutsikos@YAHOO.COM>
Organization: Ghent University
Subject:      Replicates in Clusters - PROC MIXED?
Comments: To: sas-l@uga.edu

Dear All,

I have a question regarding the use of PROC MIXED.

My data concern the impact of a lifestyle intervention on cholesterol levels
in coronary patients.
It is a multicenter study (8 countries) with cholesterol measurements done
BEFORE and AFTER the intervention program (which was done in each center).
The obvious question that I want to get an answer for: is there a
significant impact of the intervention program on average cholesterol
levels?
In other words, is the change in AFTER-BEFORE differences significant.

This would be a simple exercise if the clustering of patients within
countries was ignored.
However, I would like to take into account the following set of model
properties:

1) the baseline levels (BEFORE cholesterol levels) may randomly differ
between countries (random intercept)
2) the changes after-before may randomly differ between countries (random
slope)

Therefore, I thought of fitting a random slope model that with PROC MIXED
would look like:
(id=unique patient identification number; time: before=1; after=2)

PROC MIXED;
CLASS id time country;
MODEL cholesterol = time / SOLUTION;
REPEATED time / TYPE=cs SUB=id;
RANDOM int time / SUB=centre gcorr;
RUN;

Is this a correct code or do you have any other suggestions?

Thanks,
Luke