Bioequivalence and Bioavailability Forum

Daddy!

❝ From my vacation:

Do I smell some signs and symptoms of forum-addiction?

❝ CVfromCI() gives you the CV for the difference T vs. R.

❝ This is some pooled value of the intra-subject variances of T and R ¹⁾.

Yep.

❝ So don't expect to get a value comparable to s²_wR!

OK, OK. Childish boyish.

❝ s2I=(s2D + s2wT + s2wR/2)

❝ where s²_I is the variance of the difference T-R used for calculating the 90% CI, s²_D is the subject-by-formulation interaction, s²_wT and s²_wR are the intra-subject variabilities of Test or Reference, respectively.

❝

❝ Since you didn't provide me s²_wT try it by your own assuming s2D=0 which in many cases could be reasonably assumed for the partial replicate design.

❝ Remember our discussion on specifying compound symmetry in the FDA Proc Mixed code.

Yessir. PHX gives me a standard error of the difference (FDA’s ABE code) of 0.0468069; s²_WR 0.139466 and s²_WT 0.0153728 [sic]. Not negligible S×F of 0.166035. As we know SAS will spit out different values for s²_WT and the S×F. Stupid enough with John’s data no warning. If I ignore the S×F I get 29.805% for the pooled CV_W – close to PowerTOST.
For completeness results of different parameterizations of the variance structure in PHX:
                                          s²WR      s²WT
Banded No-Diagonal Factor Analytic (f=2)  0.139466  0.0153728 (= FDA’s)
Compound Symmetry                         0.139466  0.0833300
Heterogenous Compound Symmetry            0.139466  0.2073849

Enjoy your vacation!