Bioequivalence and Bioavailability Forum

Dear DLabes!

❝ can anybody give me a hint about non-parametric evaluation of BE (average) within replicate cross-over design (2-sequence-4-period and/or 4-sequence-4-period)?

First of all congratulations for being the first colleague posting in my personal favorite category “Nonparametrics”…
… but such a difficult question!
To my knowledge there’s simply no specific method published at all.
Since replicate designs were at least to some degree submitted to the FDA, t_max did not play an important role, because evaluation of early exposure (partial AUC up to median t_max of the reference – rather than a direct comparison of t_max itself) performed by a parametric method is considered the standard over there.

❝ This will be needed for Tmax evaluation, for which most guidances recommend a non-parametric evaluation.

You are right. But if you go beyond the conventiona 2×2 design, nonparametrics will leave you out in the rain. I know of just two publications (for a 2-treatment, 2-period, 4-sequence design¹ and a 6×3 Williams’ design²), although

• both of them don’t deal with confidence intervals,
  
• rely on normal approximations (i.e., are not exact), and
  
• don’t help you in the case of a 2×4 or 4×4 replicate design…

One possibility comes into my mind: Performing a parametric evaluation of the replicate design, but not on the untransformed raw data of t_max, but on the ranks within each period.
But I must confess, I have neither done such an analysis myself nor seen it anywhere…

---

1. Elswick RK, Uthoff VA. A nonparametric approach to the analysis of the two-treatment, two-period, four-sequence crossover model. Biometrics. 1989;45(2):663–7.
   
2. Bellavance F, Tardif S. A nonparametric approach to the analysis of three-treatment three-period crossover designs. Biometrika. 1995;82(4):865–75. doi:10.1093/biomet/82.4.865.