Bioequivalence and Bioavailability Forum

Hi Simon,

❝ Our standard approach for BE studies is to replace BLQs with zero.

Not ideal. I would define the concentration-column as character instead of numeric (Phoenix/WinNonlin) and keep them coded BQL. If you are using an R-package (bear, PKNCA) keep the column numeric but replace all BQLs by NA. In any case use the linear-up/logarithmic-down trapezoidal method for AUC. The linear trapezoidal should go to the waste bin.

❝ However, one subject has a profile for test product where all values are BLQ; their reference product profile is relatively in line with the rest of the dataset. If this was the other way around (reference profile all BLQ)…

Exactly, bad luck.
The EMA’s logic behind: If the reference product shows low/irregular profiles sometimes (not unusual for highly variable drugs / drug products), it does not matter. There were no safety/efficacy issues in phase III and IV – despite occasional “bad” product performance. Hence, not an issue. On the other hand, there are no safety/efficacy data available for the test product. Therefore, exclusion in BE (which is a surrogate for TE) is considered not acceptable.

❝ There are no other clinical/medical reasons to exclude this subject from the PK analysis set, …

That’s bad. Otherwise, you would have a justification for exclusion.

❝ … however the zero values for Cmax and AUC cannot be log transformed and therefore this subject's data cannot included in the statistical analysis.

Correct in principle. Nevertheless, exclusion “on the basis of statistical analysis or for pharmacokinetic reasons alone” is not acceptable (BE-GL p. 14).

❝ My initial thought would be to replace BLQ with LLOQ/2 for this subject only.

Why only for this subject? BE-GL p. 14: “The data from all treated subjects should be treated equally.” What does your protocol say?

❝ Does that seem like a logical solution?

No – what’s so special about dividing by 2?
Why not \(\hat{C}=LLOQ/\textrm{e}\) or \(\hat{C}=LLOQ/\pi\)?