Dear allstatters,
I would like to get your opinion on the following problem.
I am writing an analysis plan for a randomized clinical trial involving 2
doses of an active compound and a placebo. During the trial there will be
6 post baseline measurements on some rating scale per patient. The change
from baseline is the outcome variable. We expect to see some drop-outs but
are not sure how much. Also we are not sure about the course of the
ratings over the different timepoints for the various groups. We expect to
see some placebo effect, but that may wear off or patient may even worsen
compared to baseline as the trial duration is longer. We have no idea
about the effect of the active compound, nor of the dose effect, it might
be the low dose is effective, it might be the high dose, it might be
neither or both. Also we do not know about the time of onset of action, if
any.
We plan to analyse the data using SAS proc mixed in a repeated measures
mixed model. We will assume an unstructured correlation matrix for the
within-subject errors. The fixed categorical effects are treatment and
time. The SAS code looks as follows:
proc mixed ;
class patient treatment visit ;
model change=treatment visit treatment *visit /solution;
repeated visit /sub=patient type=un;
run;
My question is what would be the appropriate term to use for the F-test
for differences among the three treatment groups in the change from
baseline on the rating scale over the course of treatment duration? I
understand that if I would assume a linear slope model, that the
interaction term should be used to test for differences in 'rate of
change' among the three groups. But when I model the visits as categorical
effects, what does this interaction test mean and what does the F-test for
the main treatment term tell me. Does the treatment term test for a
difference in average changes over the whole duration ? And what if I am
interested in the difference between two treatments at the last visit
only, how would that contrast statement look like ?
Any iinsight is greatly appreciated.
Sylvia Engelen
[log in to unmask]
|