Bioequivalence and Bioavailability Forum

Dear all,

Japan and South Korea have a particular way of dealing with studies failing the common CI inclusion. A study may still pass if the PE is within 90–111% (Japan: if ≥20 subjects or ≥30 subjects in an add-on study, South Korea: >24 subjects; both require comparable dissolution). I set up a simulation:



Another opportunity to validate PowerTOST by a brute-force attack (of course only the left branch; in all 10⁶ sim’s finished so far RSE <±0.2%).

Is this setup correct? Conventional power increases with the sample size (trivial), but power based on the PE inclusion decreases. Example: GMR 1.15, CV_intra 0.3 (don’t hit me for the values; I wanted to start with rather extreme ones):
n    power1    power2   aggregate  power1 (exact)
24  0.238497  0.106900  0.345397   0.238569
30  0.282906  0.056098  0.339004   0.282637
36  0.321658  0.023093  0.344751   0.321168
42  0.358078  0.006565  0.364643   0.357413
48  0.391856  0.001179  0.393035   0.392091
54  0.424837  0.000142  0.424979   0.425363
60  0.457285  0.000009  0.457294   0.457275
With higher sample sizes on the average we will get a PE close to the expected GMR of 1.15 which is outside the PE limits.

Based on some preliminary results I don’t want to speak about α.