Bioequivalence and Bioavailability Forum

Dear ElMaestro,

❝ But that isn't really between-batch precision, by any reasonable definition of between-batch, in my opinion. It is precision without regard to within and between, because the calculation sd does not take into account where the values came from.

Absolutely. It is rather an overall, Grand Variability.

❝ If we were to truly evaluate between-batch precision (i.e. what is the CV of our estimate between batches) then we could study m batches, extract the m means and do precision on the means.

I'm not sure the information you would get from such calculation would be really meaningful. IMHO, calculating the mean of means does not reflect the variability of the method as applied to subject samples, which are analysed as singlicates.

If I remember correctly there were discussions on how to evaluate these data at the EBF/EUFEPS conference in Brussels in April 2010 while the EMA guideline was in public consultation. At that time, industry was suggesting to have more runs for precision and accuracy, but with less repetitions in each run. So instead of having 3 runs with 5 repetitions each, have one run with 5 repetitions for within-run P&A, and 4 or 5 other runs with just 2 or 3 repetitions. They said that otherwise it was just a repetition of within-run P&A and it was providing less information on between-run. Then, if I remember correctly it was suggested to have an ANOVA to extract the within- and between- components of the variability. I'm writing this from the back of my memory at a time when I should be sleeping in my bed, so I might be totally wrong.

(BTW, the comment was taken into consideration and that's why there is no figure for the number of replicates for between-run P&A in the EMA guideline. A minimum number of replicates is only given for within-run P&A).

❝ The reason it isn't done in practice is apparently that it is too complicated

Yeah, that's basically why the suggestion on the calculation was rejected. I'm not sure the regulators really knew what they would have done with such information and what the acceptance criteria should be .

❝ If we wish to accept companies just doing sd from the m x n values and using this to qualify the validation then we shouldn't call it between-batch precision.

Strictly speaking, yes, you're right. But that's the terminology everybody is used to using, and I wouldn't be surprised to see it again in the ICH guideline.

❝ Or alternatively, if we wish to insist on working on an experiment called between-batch precision then we should calculate it differently.

I think regulators don't really care about the true contribution of the between-run effect on the overall variability of the method. What they are interested in is the overall Grand Variability itself. The question is kind of, is the method good or bad ?