Bioequivalence and Bioavailability Forum

Hi Darborn,

❝ When design a crossover trial with two test drug (or one ODT given with/without water), I run into confusion about how to give a balanced design for the trial.

If the design is not reference-replicated, easy. See this article.

❝ Furthermore, which make things even worse, I always get confused by "variance-balanced" and "carryover-balance" design.

True unequal carryover will bias the treatment effect. It cannot be “corrected” statistically. It can only be avoided by sufficiently long washouts. See this article. Carryover is not included in the current models for BE. Since sample sizes for a conventional Latin Squares and (variance-balanced) Williams’ designs are similar,* you can use any of them.

❝ I came out with the following design, but I'm not sure if it is suitable to use when analyzed with FDA RSABE approach.

❝     T1  R   T2   R

❝     R   T1  R   T2

❝     T2  R   T1  R

❝     R   T2  R   T1

First of all you have to receive a positive statement from the FDA in a “Controlled Correspondence” because this is not one of the recommended designs for RSABE.
According to ICH M13A you have to exclude treatments (T1 or T2) and perform two separate analyses on the respective others (T1 vs R and T2 vs R). Your design will lead to two partial replicate designs with missing observations (denoted with • below):

1. 
   T1  R   •   R
R   T1  R   •
•   R   T1  R
R   •   R   T1
   
   
2. 
   •   R   T2  R
R   •   R   T2
T2  R   •   R
R   T2  R   •

So far I’ve done that only in pilot studies (deciding which of two candidate formulations should be used in a pivotal study). Drawback: You have to assume CV_wT = CV_wR because in partial replicate designs CV_wT is unknown. If CV_wT < CV_wR (quite often), your study will be overpowered and economically questionable. If CV_wT > CV_wR your study will be underpowered (higher risk of failure). Both cases are ethically questionable.

---

• For 4-treatments the sample size is equal in Latin Squares and one of the six possible Williams’ designs.

Hello,
so the most appropriate design for 2 test treatments and replicated reference is a below one?

❝ T1  R   T2   R

❝ R   T1  R   T2

❝ T2  R   T1  R

❝ R   T2  R   T1

Do you think EMA regulators would have any objections if this is was pivotal study (of course bearing Bonferroni adjustment in mind)?

Probably following designs would be worse?:

a) EMA suggested replicate design with additional test treatment
TRR|T2
RTR|T2
RRT|T2

b) 3x6x3 Williams design with additional reference treatment
ABC|C
ACB|C
BAC|C
BCA|C
CAB|C
CBA|C

Best regards
BEQool