PROC MIXED - To analyze cluster randomized trial

#1
Hi All,

I have a query regarding how to analyze a cluster randomized trial. in the trial, even though the unit of randomization is the cluster, unit of analysis is the patient. The study had clinics that were randomized as either intervention or no intervention. so patients in intevention clinics received an intervention and did pre and post measures.

trying to find out if intervention had any effect on the measures (pre and post). does the following make sense? can someone help me please?

proc mixed data=work.cluster;
class cluster_;
model measure1 =/Intervention;
random intercept/sub=cluster_;

Your help would be much appreciated!
Thanks
Chami:)
run;
 
#2
Hi chami,

I think your design can be thought of as a split-plot with main plot=cluster and subplot=patient. Main plots received intervention and patients "received" pre/post. SAS shows how to fit the PROC MIXED for split-plot in this example.

One thing to remember is a patint always has Pre first, Post second, so there might be a carry over effect. By fitting the split-plot it would be assumed that pre/post were "assigned" in a random order.