Bioequivalence and Bioavailability Forum

Dear key,

nice that you accost me personally, but searching the forum you would have noticed that I don't have the slightest idea about

❝ Have a working SAS model for a replicate design, 2 treatments, 4 periods

❝ (each treatment is delivered twice).

Define 'working model'. For ABE, IBE/PBE?

❝ Now need to do followup analysis on the 4 individual treatment

❝ comparisons

❝ A1vA2

❝ B1vB2

❝ A1vB1

❝ A2vB2

❝

❝ and am having a little bother programming the model.

I guess that your indices denote the order of administrations regardless the period? Let's have a look at FDA's baby:
     P1 P2 P3 P4
S1:  T  R  R  T
S2:  R  T  T  R
So are you thinking about something like:
     P1 P2 P3 P4
S1:  A1 B1 B2 A2
S2:  B1 A1 A2 B2
I don't think that your between occasion (within treatment) comparisons make sense - and I would say they are confounded with the period effect as well. Overall I'm not even sure whether I understand the rationale behind your four questions (especially the last two). In other words - what do you want to know?

❝ Can this even be done using proc mixed?

I don't think so; I would say you are asking the wrong questions. But I'm only a statistical amateur and not gifted with ...

Dear Kev,

❝ Have a working SAS model for a replicate design, 2 treatments, 4 periods

❝ (each treatment is delivered twice).

What is your working model? FDA code?

❝ Now need to do followup analysis on the 4 individual treatment

❝ comparisons

❝ A1vA2

❝ B1vB2

❝ A1vB1

❝ A2vB2

Agree with Helmut. Seems to me not the right question within a replicate design.

If you really like to consider the two replicates of Test (A) and Reference (B) as distinct (but why?) you don't have a replicate design anymore.

It's then a 4-treatment-4-period-x-sequence design.

Have a look into some textbook to find out:
The model for the 4-treatment-4-period-whatever-sequence design is the same as for the usual 2x2 cross-over.

If you prefer Proc Mixed over Proc GLM (but beware of regulatory 'constraints' if your evaluation is deemed as submission to the EMA, see here):

Proc MIXED data=yourdata;
  class tmt sequence period subject;
  model logPK = tmt sequence period /ddfm=KR;
  random subject(sequence);
*The LSMeans statement will give you all comparisons you think you need;
  LSMeans tmt/diff CL alpha=0.1;
run;

But beware that you have coded your replicate treatments as A1,A2 and B1,B2 and your sequences accordingly.

if you know .