Bioequivalence and Bioavailability Forum

Dear all,

When I am validating WNL's BE (v 5.1.1) results by comparing the results of WNL with those of my manual calculation, I find that WNL's BE's 90% CI of the Ratio is lack of precision.

My dataset is Chow and Liu's famous data:
Design and Analysis of Bioavailability and Bioequivalence Studies, Third Edition, Page 71.

Select Ln()-transformation in WNL's BE wizard.

For comparison, I list some critical results:

First found

---

WNL 5.1.1:
WNL's 90% CI of LSMean(R-T)   -0.066980983221424   0.124285167688542
WNL's DiffSE                   0.055693090418743    
WNL's 90% CI of GeoMean Ratio(%) (T/R)  88.3108418878651   106.9298810918720

Here I find the above 90% CI of GMR is lack of precision.
According to Phoenix WinNonlin 6.0 guide Page 338:
CI_Lower = 100 · exp(lower)
CI_Upper = 100 · exp(upper)

my manual result is:
Manual 90% CI of GeoMean Ratio(%) (T/R) based on WNL‘s 90% CI of LSMean(R-T)
88.3127965410502 106.9275143812610

Precision is only at the lelvel of 0.01(%), Incrediable? Interesting? Strange?

Can anyone give me the result of PNX WNL 6.0 or later version? Dear HS or ElMaestro or others warm-hearted?

Second found

---

Moreover, I find thta WNL‘s 90% CI of LSMean(R-T) -0.066980983221424 0.124285167688542 is also lack of precison.

WNL 5.1.1
WNL's Diff (T-R)              -0.028652092233559    
WNL's DiffSE                   0.055693090418743    
WNL's Tinv(0.1,22)             1.71714434835526
   
WNL's 90%CI of Diff (R-T)     -0.066980983221424     0.124285167688542

my manual result is:
Manual Diff(T-R)              -0.028652092233559    
Manual DiffSE                  0.055693090418743    
Manual Tinv(0.1,22)            1.71714437438148    
Manual 90%CI of Diff (T-R)    -0.124285169138022     0.066980984670905

It can be seen this small bias is caused by the difference of WNL's Tinv(0.1,22) and my Manual Tinv(0.1,22). Here I am not sure that my Manual Tinv(0.1,22)=1.71714437438148 is of enough precison. So I have to try other software to get more Tinvs:

   Software                Tinv(0.1,22)
WNL 5.1.1               1.71714434835526
Excel 2003 and 2007     1.71714433543983
Open Office 3.3.0       1.71714437438025
Gnumeric 1.10.16        1.71714437438148
I know Excel 2003's result is a little poor, but I don't know which is more reliable. So I need Daer HS's help to give out the result of R's Tinv(0.1,22).

Conclusion

---

Even if we accept WNL's Tinv(0.1,22) and thus accept WNL's 90%CI of Diff (R-T). That is to say we accept the low precision of my Second Found.
I really can not accept the poor precision of my First Found.

Or my manual calculation is poor precision?
I need more Dears to validate these results, give my best thanks to you.