Hello,
Our experiment measures a marker over time (DV=ABS) (4 time points 0, 15, 30, 60 min). Out independent variable is TRT. We used BL as a COVARIATE which is a continuous variable. Values for the covariate BL correspond to the measurement of DV taken at time point 0. We use BL as a covariate since we were under the impression that any differences we could have at baseline could be accounted for by using it as a covariate. This is the code:
proc mixed DATA=TEST;
class ID TRT TIME;
model ABS = TRT|TIME BL /OUTPM=b RESIDUAL VCIRY;
repeated TIME/type=CS sub=ID;
lsmeans TRT|TIME/adjust=tukey;
run;
However, when I run the procedure, the differences didn't make sense. That is, when you graph means for all time points, the stats differences with Tukey do not seem to reflect what you see on the graph. See below.
So I ran an ANOVA on BL (covariate-time point 0) and it was significantly different. Then, we ran proc mixed on change from baseline. Which means that for values for time points 15, 30 and 60, time 0 was subtracted. And we still included BL as a covariate. This is the code:
proc mixed DATA=TEST;
class ID TRT TIME;
model CBL = TRT|TIME BL /OUTPM=b RESIDUAL VCIRY;
repeated TIME/type=CS sub=ID;
lsmeans TRT|TIME/adjust=tukey;
run;
Again, the results didn't make sense when you looked at the graph.
How can a and b be different? BUT when I remove the covariate from the model 2 (change from baseline), the results make sense. Top two lines are different from the third one below.
proc mixed DATA=TEST;
class ID TRT TIME;
model CBL = TRT|TIME /OUTPM=b RESIDUAL VCIRY;
repeated TIME/type=CS sub=ID;
lsmeans TRT|TIME/adjust=tukey;
run;
My question is, is it valid to remove BL as a covariate from model 2 since we have already accounted for the differences at baseline by subtracting it from all the other time points?
Any input is appreciated.
Thank you!