Bioequivalence and Bioavailability Forum

Hi Niraj,

24 is correct. But, if you assume a CV_WR of 30% the probability of not being allowed to scale is 50%. So it might well be possible that you will be trapped in the situation described in my previous post – therefore, forget the partial replicate! Don’t assume a GMR of 0.95 for HVDs/HVDPs. If you run the study in 24 subjects and the GMR is 0.9, power will be only 59%.

Dear Niraj,

❝ Could you please guide me how to use PowerTOST from the given link? I have never used it before. I have downloaded file from the given link.

❝

❝ We use SAS software and I guess, given link directs to R-language. Correct me if I am wrong.

You are totally right, PowerTOST is an add-on package for the R-language.
An implementation in SAS was not undertaken by me since the scaled ABE power calculations have to be based on simulations and it has been reported that the run times for such an attempt are horrible (around 1 to some days).
R is on the other hand very suitable for simulation purposes and with some clever tricks the run times are below minutes).

To use PowerTOST you must first install R. It is open source software so you don't need something like licenses or many bucks to buy. Obtain it from CRAN.

After downloading and installing R you may install PowerTOST also from CRAN via RGUI (simpler) or R-console (needs some knowledge about R-language).
In the GUI chose the menu 'Packages/Install package(s) ...' and, after choosing a CRAN mirror near to you, select package PowerTOST.

To use it for your problem try in the R-console:
library(PowerTOST) # load the package

# sample size estimation with partial replicate
# assuming intra-subject CV's of test and reference equal 30%
# and assuming a true GMR of 0.9 (pessimistic)
# will give you N=45
sampleN.RSABE(CV=0.3, theta0=0.9, design="2x3x3")

# more optimistic GMR=0.95
# will give you N=27
# (there are some small differences to the paper of the 2 Laszlo's
#  due to higher number of simulations used in sampleN.RSABE)
sampleN.RSABE(CV=0.3, theta0=0.95, design="2x3x3")

# sample size estimation with full 3-period replicate (TRT|RTR)
# according to Helmut's suggestion
# assuming intra-subject CV of test and reference equal
# will give you N=46
sampleN.RSABE(CV=0.3, theta0=0.9, design="2x2x3")

# assuming a more variable test formulation with CV=0.4
# will give you N=62
sampleN.RSABE(CV=c(0.4,0.3), theta0=0.9, design="2x2x3")

# assuming a lesser variable test formulation with CV=0.25
# will give you N=40, i.e. it pays to have a 'better' test
sampleN.RSABE(CV=c(0.25,0.3), theta0=0.9, design="2x2x3")


Hope this helps as a very short intro.
To discover more about sampleN.RSABE() type ?sampleN.RSABE in the R-console and RTFM .

And don't forget: search the forum!

Hi Kumarnaidu,

❝ Cmax-62.57(38.27 - 102.29) CV-83.85

❝ AUCt-46.09(17.47 - 121.55) CV-263.48

❝ As per experts study failed drastically because of its design (2x2 cross over), and replicate design should be used here instead of 2x2 design.

❝ In my opinion there may be design issue but the test product is also not upto the mark. Please provide your suggestions. Is there a need of change in design of pilot study or test product modification?

Your estimate of CV's is conditional on the point estimates and thus GMR. If you believe in the CV's you believe in the GMRs. If you don't believe in either then I guess you should have a reason. Technical error during bioanalysis etc etc.
To me, as long as concerns over technical errors etc have not been raised, this just looks like a clear-cut failed development to me. I would not spend as much as $.02 on further studies on the present formulation.