Bioequivalence and Bioavailability Forum

Dear Balakotu,

❝ [...] but actually using the above anova procedure we can't calculate Least Squares Means,intrasubject cv and 90% conficence interevals and also anova Least Squares Means shows some missing data

First I was very surprised about your message.

But after looking a little bit closer at your used sequences and after playing with some artificial data using that sequences I understood:
No Least square means estimable and no 90% confidence intervals.
Only the differences T₁-T₂ and R₁-R₂ estimable, that ones you are presumably not interested in, but no CIs. Ups!

Ok, this is a consequence of your design, your used sequences.
This study is totally unbalanced with respect to the occurrence of the 4 treatments in the 4 periods:

T₁ and T₂ only in periods 1/3
R₁ and R₂ only in periods 2/4!

The result is that you cannot estimate the period effects independent from the treatment effects.
As way out you have to assume that there are no such period effects in your model.
Just omit the period term in the Proc GLM above and numerically all is Ok.

But this is a major flaw of your design because period effects are rather common (15-20%) in bioequivalence studies.


BTW: The intra-subject CV is estimable, also in the model with period effects.
BTW2: See Helmut's lectures of better common designs for 4-treatment-4-period studies without restricting assumptions. See also this thread for further discussion of 4-period-4-treatment design issues, I just checked it, your design thoroughly discussed as unsuitable.
And think twice before using such an uncommon design.


----------------------------------------------------------
To Helmut: Salzburg is very fine, but very wet at that time.
Now I understand what it means: Salzburger Schnürlregen!
----------------------------------------------------------