Bioequivalence and Bioavailability Forum

Hi Relaxation & all,

❝ From a quick once over interesting topics included .

Since I’m working on a paper on group-effects in BE (still collecting data; see this post) I had a closer look at Section III.D.

The relevant parts:

---

[…] sponsors should minimize the group effect in a PK BE study as recommended below:

1. Dose all groups at the same clinic unless multiple clinics are needed to enroll a sufficient number of subjects.
   
2. Recruit subjects from the same enrollment pool to achieve similar demographics among groups.
   
3. Recruit all subjects, and randomly assign them to group and treatment arm, at study outset.
   
4. Follow the same protocol criteria and procedures for all groups.
   
5. When feasible (e.g., when healthy volunteers are enrolled), assign an equal sample size to each group.

Bioequivalence should be determined based on the overall treatment effect in the whole study population. In general, the assessment of BE in the whole study population should be done without including the treatment and group interaction(s) term in the model, but applicants may also use other pre-specified models, as appropriate. The assessment of interaction between the treatment and group(s) is important, especially if any of the first four study design criteria recommended above are not met and the PK BE data are considered pivotal information for drug approval. If the interaction term of group and treatment is significant, heterogeneity of treatment effect across groups should be carefully examined and interpreted with care. If the observed treatment effect of the products varies greatly among the groups, vigorous attempts should be made to find an explanation for the heterogeneity in terms of other features of trial management or subject characteristics, which may suggest appropriate further analysis and interpretation.

It is important that statistical methods and models for the primary BE analysis are fully pre-specified in the protocol or SAP (e.g., in an ANDA study, the applicant should pre-specify detailed statistical criteria and models to be used if the interaction term of group and treatment is applicable). In addition, the statistical model should reflect the multigroup nature of the study. For example, if subjects are dosed in two groups in a crossover BE study, the model should reflect the fact that the periods for the first group are different from the periods for the second group, i.e., the period effect should be nested within the group effect.

(my emphases)

---

The terrible ‘model (I)’ $$\eqalign{Y&|\;\text{Group},\,\text{Sequence},\,\text{Treatment},\\
&\phantom{|}\;\text{Subject}(\text{Group}\times \text{Sequence}),\,\text{Period}(\text{Group}),\\
&\phantom{|}\;\text{Group}\times \text{Sequence},\,\text{Group}\times \text{Treatment}}\tag{I}$$ as a mandatory pre-test – which was not stated in any of the previous guidances but in deficiency letters – is gone with the wind. Very good, because it inflated the Type I Error and compromised power.
Now the FDA recommends ‘model (II)’ $$\eqalign{Y&|\;\text{F},\,\;\text{Sequence},\,\text{Subject}(\text{F}\times \text{Sequence}),\\
&\phantom{|}\;\text{Period}(\text{F}),\;\text{F}\times \text{Sequence},\,\text{Treatment}\small{\textsf{,}}}\tag{II}$$ where \(\small{\text{F}}\) is the code for \(\small{\text{Group}}\) or \(\small{\text{Site}}\), respectively. Whilst important in a parallel design, I fail to understand why ‘similar demographics’ are of any importance in a crossover design. Equally sized groups are recommended but not necessary.
If any of the first four conditions are not met, ‘model (I)’ should be used to assess the Group-by-Treatment interaction. If \(\small{p(G\times T)<0.1}\), ‘vigorous [sic] attempts’ should be made to find an explanation. It is an open question what to do if treatment effects vary between groups. Base the BE assessment on the largest one or on the one with smallest variability?

See also the updated article and doi:10.1208/s12248-024-00921-x.