Bioequivalence and Bioavailability Forum

Dear Pavan!

I never have used the Analyst application in SAS up to now for sample size estimation purposes because it is not documented in full detail what is going on there behind the scenes.

Now after looking at the code produced by this application (in the macro language of SAS, a beasty dragon to struggle with ) I noticed the following:

1. None of the code does exactly met the sample size formulas for a 2x2 cross-over design! (To excuse this part of SAS software: At no place it is mentioned, that the sample size is valid for a 2x2 cross-over.)
   
   
2. The paired equivalence test is the paired t-test with degrees of freedom df=n-1, whereas the correct df is n-2, if you consider period effects in the 2x2 cross-over.
   
   
3. The obtained sample size is only in the order of the correct ones, if you specify the CV as SQRT(2)*CV(usual) = the CV of the log-differences, not the one in the usual sense, i.e. the intra-subject CV.
   (This is only mentioned in the deep jungle of the help files, not in the input window!)
   If CV is used in the usual sense of BE studies, the sample size is much to low, from which your original question (double of sample size?) originates I suspect.
   
   
4. The achieved power of the iterative sample size determination is sometimes below the desired power because of a flaw implementation of the stopping rule of the iterative algorithm.

Example: CV=20%, input in the Analyst-application CV=0.28284
desired power 0.8, alpha=0.05, lower BE limit = 0.8, upper=1.25
Results of sample size:
null          achieved     |
ratio    N  power (df=N-1) | N(Diletti et al.)
---------------------------------------------
0.90    36    0.80028      |  38
0.95    18    0.79867      |  20
1.00    15    0.80074      |  16
1.05    18    0.79845      |  18
1.10    31    0.79942      |  32

Also the differences may be seen not substantial here, the sample sizes are nevertheless not the correct ones.

Lessons to learn:

1. Do not use any software without full documentation of the implemented algorithms.
   
2. Do not use any software without knowing what input is desired.
   
3. Do not use any software without recalculation of some known test data (validation).
   
4. Do not use the "Power to know". Especially if you don't know. Or have the power . (This is not to insult or bother someone, but it is sometimes really hard to guess what SAS syntax really means and to bring SAS code to produce that results you need! I know about what I'm talking about, trust me.)

May be Others have more Donot's.