## impressive indeed [RSABE / ABEL]

Hi mittyri,

» […] Dr.Patterson cited it in his dissertation.

OK, a Ph.D. thesis (even when supervised by Byron Jones and Stephen Senn) is not a reliable source. Nevertheless, it contains some gems.
1. About PBE/IBE suggested by the FDA (p.69)
Other academic resources (Senn, 2000) held that average bioequivalence should suffice based upon grounds of ‘practicality, plausibility, historical adequacy, and purpose’ and ‘because we have better things to do’. Additionally, Senn (2000) notes that statisticians have ‘a bad track record in bioequivalence’, that ‘the literature is full of ludicrous recommendations from statisticians ’, that ‘regulatory recommendations (of dubious validity) have been hastily implemented’, and that ‘practical realities have been ignored’.

It is known (Patterson et al., 1999; Zariffa et al., 2000) that variance estimates generated in bioequivalence studies powered for average bioequivalence are poorly (i.e. imprecisely) characterized, and estimates in excess of the $$\small{\sigma_\textrm{D}>\textrm{cut-off}}$$ (Hauck et al., 2000) should be expected due to random chance […]. Increasing sample size does appear to provide some benefit in making these estimates quantitatively more precise […] but the larger sample sizes needed to achieve this are not currently recommended for demonstration of average bioequivalence in moderate variability compounds (FDA Guidance, 2001).
Thus, as a practical matter, in studies powered for ABE (the current international standard […]), inference on variance estimates, or resulting metrics like IBE and PBE, should be approached with caution.

Only one REML procedure (UN) was found to yield unbiased estimates in complete data sets and those with missing data. Method-of-moments, as expected, yielded unbiased estimates in complete data sets, but was positively biased in samples with missing data. Bias in method-of-moments (with missing data) and constrained REML procedures increased as drugs become more highly variable and decreased with increasing sample size. Biased method-of-moments estimates in data sets with missing data were greater than those found in CSH REML which were in turn observed to be slightly greater than thos derived using RIS REML. The performance of estimates from FA0(2) REML was questionable. Estimates were positively biased when the true $$\small{\sigma_\textrm{D}^2=0}$$ and estimates were negatively biased when $$\small{\sigma_\textrm{D}^2>0}$$.

#1 is funny but IMHO, true.
#2 is interesting. The same holds true for RSABE. Even worse for the FDA’s RSABE of NTIDs where a test of swT/swR is part of the procedure. Are the sample sizes large enough?
#3 ‘The performance of estimates from FA0(2) REML was questionable’. Oops! Note that the simulations were performed for 4-period 2-sequence full (‼) replicate designs.

Regrettably I failed to find Senn (2000).*

• Senn S. Conference Proceedings: Challenging Statistical Issues in Clinical Trials. 2000. Decisions and Bioequivalence.

Dif-tor heh smusma 🖖
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes