Power analysis in a repeated measures desgin


New Member

We are planning a randomised clinical trial treating patients with either an active drug or placebo in 20 weeks. We measure the dependent variable several times during the study and want to evaluate the efficacy of the drug at week 20 in respect to placebo. The analysis will be within a linear mixed model (SAS code with 4 repeated measurements):

CLASS Id Treatment Week; 
MODEL Value = Treatment Week Treatment*Week;   
REPEATED Week / Subject=Id Type=AR(1);
CONTRAST 'Active vs Placebo at week 20' 	Treatment 1 -1
								Treatment*Week 0 0 0 1 0 0 0 -1;
We have a fixed number of participants and want to calculate the gain in power by adding more repeated measurements. We have calculated this both by estimating the noncentrality parameter as in "SAS for Mixed Models" by Littell and by simulation (getting the same results). The difference between groups is modelled to be linear across time.

The problem is that we get the same results with different values of the second parameter in the autoregressive covariance structure. I would have expected different results when the covariance between repeated measurements are different or am I missing something?

Hoping anyone can help us?