Bioequivalence and Bioavailability Forum

Dear yicaoting!

❝ It's now midnight in China, tomorrow is Monday, I must sleep now.

Good night!

❝ My results are same to WNL's at the level of 0.0001 for Power, both for untransformed and Ln()-transformed data.

Hhm. Going with Chow’s/Liu’s data set (PHX 6.2, ln-transformed):

Power at 20% = 0.9839865

How did Pharsight derive this value from the given…

Var(Residual)    0.037220643844683
Intrasubject CV  0.194735735490703 almost correct: CV_intra=√ℯ^MSE-1
Ratio %Ref      97.1754486595894

Using famous package PowerTOST 0.8-7 for R (α 0.05, acceptance range 0.80–1.25, CV 0.1947357354907, n 24, 2×2 design, estimation exact or approximative) I got:

PE                power (exact = TRUE or FALSE)
1                 0.9745468083
0.971754486595894 0.9556573312
0.95              0.9096019837

Since power even at PE=1 is lower than PHX’s results, the additional lines in PowerTOST are just for fun. Are they using a larger α of 0.10‽

PE                power (exact = TRUE or FALSE)
1                 0.9918976973
0.971754486595894 0.9831552162
0.95              0.9593213240

Ah – this goes into the right (=wrong!!) direction. I had a quite lengthy discussion with Pharsight 1½ years ago and it is clear that PHX/WNL doesn’t contain an algo for the non-central t-distribution. So let’s try with PowerTOST’s hidden function of the shifted central t-distribution instead (still α 0.10! – .approx2.power.TOST(alpha = 0.1, ltheta1 = log(0.8), ltheta2 = log(1.25), diffm = log(PE), se = CV2se(0.194735735490703), n = 24, df = 22, bk = 2)):

PE                power
1                 0.9864870592
0.971754486595894 0.9774342541
0.95              0.9534457559

❝ I will continue to do the Power story of WNL tomorrow or later with you.

I’m leaving tomorrow for Berlin; eager to meet D. Labes and the co-author of the forum’s scripts Auge. Not sure whether I’ll have free time to work on it.

❝ Do you have any suggestion?

If you understand PHX/WNL’s manual, please let me know. Enlighten me how to derive 0.9839865.
For interested readers: The respective section of WinNonlin’s 4.1 manual here (hasn’t changed in later versions).

---

P.S.: Of course I’m not interested in post-hoc power at all. But would be nice to set up the decision-tree of a Two-Stage design within PHX/WNL.