Bioequivalence and Bioavailability Forum

Hello,
We conducted a full replicated BE study (2T and 2R) with scaling approach for DR formulation of a highly variable class 4 molecule for US
Interestingly we observed zero concentrations throughout all the sampling points for test formulation in at-least one period in ~ 45 % subjects and zero concentrations throughout the all the sampling points for reference formulation in at-least one period in ~ 50 % subjects [Note: LLOQ was sufficiently low to capture the PK profile]. Similar results were observed for this molecule in other literature too.

When statistical analysis was conducted as per Progesterone guidance of USFDA, subjects having zero concentration in at-least one period were automatically deleted in the BE evaluation as log of zero value will lead to not estimable values for Cmax and AUC. This leads to a significant data loss and detrimental effect on sample size and study outcome.

We are concerned about the data loss of subjects having non-zero PK parameters (Cmax and AUC) in at-least 2 or 3 periods, just because mathematically log (0) is not estimable.

1. Is there any statistical methodology to include data of subjects who have non-zero PK parameters (Cmax and AUC) in at least 3 periods to prevent such huge data loss
   
   
2. If a subject has zero Cmax/AUC in one period and non-zero Cmax/AUC in another period for Reference, then how should we treat such data in calculation of reference variability
   
   
3. We have got a literature in which Cmax and AUC with zero value were assigned/imputed with a fixed value = (LLOQ/2) and data of all such subjects were included in statistical analysis. This methodology was accepted by FDA for one of the NDA. Can we adopt the same for our product for ANDA submission