BE-proff
●    

2016-08-30 10:35
(2767 d 16:07 ago)

Posting: # 16612
Views: 6,220
 

 Significance of UL/LL in sample size calculation [Power / Sample Size]

Hi All,

How significant is a use of upper/low limits for calculation of sample size?

Will I have a serious error using only CV without limits? :confused:
ElMaestro
★★★

Denmark,
2016-08-30 11:04
(2767 d 15:38 ago)

@ BE-proff
Posting: # 16613
Views: 5,274
 

 Significance of UL/LL in sample size calculation

Hi BE-proff,

❝ How significant is a use of upper/low limits for calculation of sample size?


The use of confidence limits on the CV not in particularly wide practical use, but it is advocated by some. If you go with an upper limit (as compared to the CV itself), that is going to be conservative, sample size will increase, power will go up. If you go with the lower limit, then you may be risking a too low sample size, the risk of failure increases. If you use limit then you need a level like 80% or 90% or whatever. It is really not set in stone and guidelines do not give an answer.

❝ Will I have a serious error using only CV without limits? :confused:


Error is a tricky word, and it has a negative feel doesn't. You will always have some uncertainty on the CV before you plan the trial. Whether you want to translate this uncertainty into a confidence limit is of course up to you.

If I were you and having some uncertainty I would do the following:
a. Try and generate a curve of power versus N for some value of CV and GMR.
b. Now adjust the GMR so that it is a little farther from 1.0 than the GRM under a. and generate the curve.
c. Now try and increase CV a generate the curve.

You will quickly be able to derive a few important, but mostly non-quantitative relationships, from such experiements. It will give you some ideas about the "what if"-scenario that you are alluding to, one way or another.

At the end of the day all this stuff about power, confidence limits on GMR, confidence limits on CV, type I errors etc only works if the residual in the linear model is normal. While it is mandatory to use that model (normal residual) for evaluation and assessment, usually/often/sometimes the residual is really not normal. We never test for it, and that in itself translates directly into additioanal uncertainty on whatever estimate you are working on, regardless of how fancy your formulae otherwise are.

Pass or fail!
ElMaestro
BE-proff
●    

2016-08-30 15:53
(2767 d 10:50 ago)

@ ElMaestro
Posting: # 16614
Views: 5,134
 

 Significance of UL/LL in sample size calculation

Hi ElMaestro,

Thank you for clarification.
Do you have an example of code for PowerTOST to plot a curve for power assessment?
ElMaestro
★★★

Denmark,
2016-08-30 16:07
(2767 d 10:35 ago)

@ BE-proff
Posting: # 16615
Views: 5,169
 

 Significance of UL/LL in sample size calculation

Hi BE-proff,

❝ Do you have an example of code for PowerTOST to plot a curve for power assessment?


Something like this for a 222BE study:

GetMeTheCurve=function(CV, GMR)
{
 N=2*c(5:40)
 Pwr=N
 for (i in 1:length(N))
 {
  Pwr[i]=power.TOST(n=N[i], CV=CV, theta0=GMR)
 }
 ##
 plot(N, Pwr)
 return(0)
}

GetMeTheCurve(CV=0.28, GMR=0.95)

Pass or fail!
ElMaestro
mittyri
★★  

Russia,
2016-08-30 16:25
(2767 d 10:18 ago)

@ ElMaestro
Posting: # 16616
Views: 5,232
 

 a very nice function of powerTOST

Hi ElMaestro and BE-proff,

or even simpler (thanks to Detlew and Helmut!):
library("PowerTOST")
pa.ABE(CV=0.28)


[image]

Kind regards,
Mittyri
d_labes
★★★

Berlin, Germany,
2016-08-30 18:39
(2767 d 08:03 ago)

@ mittyri
Posting: # 16617
Views: 5,112
 

 nice functions of powerTOST

Dear mittyri,

❝ or even simpler (thanks to Detlew and Helmut!):


The Academy award of the best actor goes to ... Helmut :flower::clap:

BTW:
@All
Don't forget the note under 'Details' of the man page of pa.ABE() and similar functions pa.scABE() or pa.NTIDFDA():
"It should be kept in mind that this is not a substitute for the "Sensitivity Analysis" recommended in ICH-E9. In a real study a combination of all effects occurs simultaneously. It is up to you to decide on reasonable combinations and analyze their respective power."

Regards,

Detlew
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
77 visitors (0 registered, 77 guests [including 8 identified bots]).
Forum time: 01:43 CET (Europe/Vienna)

With four parameters I can fit an elephant,
and with five I can make him wiggle his trunk.    John von Neumann

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5