Bioequivalence and Bioavailability Forum

Dear Marcel!

❝ At what point do you think the increase in sample size becomes unethical?

Good question, next question. Gut feeling:^* n>40 (T/R 0.95, power 0.80, CV_intra just slightly >30%). That’s actually the sample size for HVDs/HVDPs.

❝ Do you have a good reference article that shows that Clast is considered unreliable?

No. I would not say it’s unreliable. Only:

• C_last is likely to be the worst metric when it comes to variability. Variability was the reason why EMA and WHO are asking in single dose studies for AUC_t – not AUC_∞ – as the metric for extent of absorption.
  
• Missing samples are a nightmare. A similar case can be made for truncated AUC₇₂ or AUC_t in general. Any missing last sample(s) will bias the estimated T/R ratio of the particular subject(s) away from 100% – that’s probably the reason why regulators don’t care about this issue. But what if the ‘true’ ratio of the formulations is 105% and in one subject after test the last scheduled sample is missing? This subject will pull the study’s ratio towards 100% (shifting the CI downwards, increasing the chance of BE, and thus increasing patient’s risk). In my studies I pragmatically use AUC_t if the LOQ is low enough so that my residual AUCs are quite small – crossing fingers that everything means out. Haven’t performed a study with AUC₇₂ yet, but most likely I will try to make an argument for an estimated Ĉ₇₂ (since residual AUC will be much higher and the effect on the overall estimate larger).
  
• Given the problems above for AUC, C_last (as a single point metric) is much more difficult to handle. If your drug accumulates, fine – but still the variability is expected to be the highest of all metrics. If not – very bad. The concept paper (Section 3.1) states:
  

  The newly revised Note for Guidance on the Investigation of Bioequivalence (EWP/QWP/1401/98 rev1), the emergence of science as well as applications on new types of formulations require thorough discussions on: […]
  • possibility of widening (and narrowing) of acceptance ranges

  In the presentations given here this statement was mentioned in the introduction but not further explored. No idea whether the PK group is considering it. A multiple dose study in a replicate design is no fun – lots of samples.

❝ By Clast, you mean the sample taken ~5 minutes prior to dosing, right?

Yes. In the actual study we sampled 5 minutes prior to the 2^nd dosing (1^st period: 23:55) and at 24:00 post dose in the last profile (2^nd period) in order to get values as close as possible to τ. We could reliably estimate λ_z. Since the study was balanced we had to estimate C₂₄ from C_23:55 in 50% of cases – we also used the estimated AUC_τ. Another option would be to sample in all profiles at 23:55 (consistent with the IR GL, Section 4.1.4):

In multiple-dose studies, the pre-dose sample should be taken immediately before (within 5 minutes) dosing and the last sample is recommended to be taken within 10 minutes of the nominal time for the dosage interval to ensure an accurate determination of AUC_(0-τ).

• Why ≤-5 min (pre dose) and ≤±10 min at τ?
  
• I prefer to get »an accurate determination of AUC_(0-τ)« by sampling at the intended τ if possible and use an estimate otherwise.

❝ Problems will continue to arise with steady state studies until this new draft guideline is released.

Right. Problems with the current NfG arise from its ambiguities:

4.1.1. Rate and extent of absorption, fluctuation
The pharmacokinetic parameters of interest are AUC, C_max and C_min or other means reflecting fluctuation.
5.1 Prolonged release formulations
Assessment of bioequivalence will be based on AUC_τ, C_max and C_min applying similar statistical procedures as for the immediate release formulations.
(my emphases; note ‘and vs. or’)

Sorry for the lenghty post.

---

• I try not to think with my gut. If I’m serious about understanding the world,
  thinking with anything besides my brain, as tempting as that might be,
  is likely to get me into trouble. Carl Sagan