Bioequivalence and Bioavailability Forum 10:19 CEST

Main page Policy/Terms of Use Abbreviations Latest Posts

 Log in |  Register |  Search

libaiyi
Junior

China,
2018-07-27 05:34
(edited by libaiyi on 2018-07-27 07:47)

Posting: # 19107
Views: 2,086
 

 the results of sample size based on PowerTOST [Software]

Hi,

I have a naive question about the results of sample size based on SampleN.TOST, sampleN.NTIDFDA, sampleN.HVNTID etc.

For example,
sampleN.RSABE(CV=0.3)

++++++++ Reference scaled ABE crit. +++++++++
           Sample size estimation
---------------------------------------------
Study design:  2x3x3 (partial replicate)
log-transformed data (multiplicative model)
1e+05 studies for each step simulated.

alpha  = 0.05, target power = 0.8
CVw(T) = 0.3; CVw(R) = 0.3
True ratio = 0.9
ABE limits / PE constraints = 0.8 ... 1.25
FDA regulatory settings
- CVswitch            = 0.3
- regulatory constant = 0.8925742
- pe constraint applied


Sample size search
 n     power
39   0.75781
42   0.78122
45   0.80344



I want to know if the target power is 0.8, the sample size is 45 for each group or for whole study.



The other question,
when we apply power.TOST, the sample size we need to fill in should be number in each group or whole study.


Third question,:-):-):-):-)
I am confused about the relationship between power and period. Are the power calculation method applied in 2*2, 3*3 and 2*4 designs the same?:-D For example, if I have the result like this of the study design as T-S-R, could I use the same method to calculate power separately for T-R and S-R? Should I get two power as the result? And, what about parallel design?

[image]

Thanks in advance! :flower:
Helmut
Hero
avatar
Homepage
Vienna, Austria,
2018-07-27 11:24

@ libaiyi
Posting: # 19108
Views: 1,993
 

 sample size based on PowerTOST: total number of subjects

Hi libaiyi,

answering parts (traveling)…

» I have a naive question about the results of sample size based on sampleN.TOST, sampleN.NTIDFDA, sampleN.HVNTID etc.

» For example,
» sampleN.RSABE(CV=0.3)
»
» n     power
»
» 45   0.80344
»
» I want to know if the target power is 0.8, the sample size is 45 for each group or for whole study.

Type help(sampleN.TOST), etc.
It is always the (total) number of subjects in the study.

» when we apply power.TOST, the sample size we need to fill in should be number in each group or whole study.

As above.

Cheers,
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. ☼
Science Quotes
libaiyi
Junior

China,
2018-07-27 11:49

@ Helmut
Posting: # 19110
Views: 1,992
 

 sample size based on PowerTOST: total number of subjects

Hi, Helmut

Thanks for answer my question while you are traveling. Enjoy yourself!:-D:-D:-D
ElMaestro
Hero

Denmark,
2018-07-27 19:57
(edited by Ohlbe on 2018-07-28 11:07)

@ libaiyi
Posting: # 19111
Views: 1,970
 

 the results of sample size based on PowerTOST

» Hi Libaiyi,
»
» » I am confused about the relationship between power and period. Are the power calculation method applied in 2*2, 3*3 and 2*4 designs the same?:-D For example, if I have the result like this of the study design as T-S-R, could I use the same method to calculate power separately for T-R and S-R? Should I get two power as the result? And, what about parallel design?
» »
» » [image]

That table looks strange. You seem to have a pair for which the CI is not estimable?? And what about alpha 0.1? Would you be after a two-sided CI with 80% coverage (yes I am aware of the ridiculous alpha in the usual SAS code for BE, but is this table derived from one such)?

I have no idea what the intention is, it just looks strange to me, and I am sure that if I got such a table in one of my studies then I would not be answering the question I was really asking.


Edit: screwed up post deleted ;-) [Ohlbe]

if (3) 4

Best regards,
ElMaestro

"(...) targeted cancer therapies will benefit fewer than 2 percent of the cancer patients they’re aimed at. That reality is often lost on consumers, who are being fed a steady diet of winning anecdotes about miracle cures." New York Times (ed.), June 9, 2018.
d_labes
Hero

Berlin, Germany,
2018-07-28 12:43

@ ElMaestro
Posting: # 19113
Views: 1,928
 

 Design with not estimable contrast

Dear ElMaestro, dear Libaiyi,

» That table looks strange...
For me too. Seems Libaiyi has used a design, in which one of the contrasts is not estimable, for what reasons ever. Libaiyi, could you give is some more info about the design and the SAS code used?

» And what about alpha 0.1?
That's a peculiarity of the almighty SAS which always calculates a two sided CI, i.e. if you specify alpha=0.05 (the default) it calculates the 95% CI (0.025 to each side) and not the appropriate 90% CI.

Regards,

Detlew
libaiyi
Junior

China,
2018-07-31 04:13

@ ElMaestro
Posting: # 19118
Views: 1,878
 

 the results of sample size based on PowerTOST

Hi, ElMaestro

Sorry for the SAS result last time, here is the modified result.
[image]
The design here is a three-period six-sequence William design. I just wondering when the study design is replicated, dose the power calculation method the same as what used in simple 2 by 2 cross over design? Thank you so much!:lol:
ElMaestro
Hero

Denmark,
2018-07-31 11:07

@ libaiyi
Posting: # 19119
Views: 1,849
 

 the results of sample size based on PowerTOST

Hi libaiyi,

» The design here is a three-period six-sequence William design. I just wondering when the study design is replicated, dose the power calculation method the same as what used in simple 2 by 2 cross over design? Thank you so much!:lol:

I am still somewhat baffled.
You have different SE's for the three comparisons, possibly suggesting that you are employing an EMA-style BE evaluation?

Anyways, if you are going for a 222BE design, then you can look at your MSE from the ANOVAs (plural, right?), convert them to CVs via CV=sqrt(exp(MSE)-1) and you have a decent variability estimate to plug in for any crossover design with or without scaling.

Your best point estimate is exp(-.2088)~0.81 with upper limit ~0.88, for the T-S pair. The others appear worse. I'd personally think twice, but I am widely known as a backward cowardly chicken.


Note, the opinion above implies logarithms and standard BE thinking.


Edit: Congratulations to your post № 1,500! [Helmut]

if (3) 4

Best regards,
ElMaestro

"(...) targeted cancer therapies will benefit fewer than 2 percent of the cancer patients they’re aimed at. That reality is often lost on consumers, who are being fed a steady diet of winning anecdotes about miracle cures." New York Times (ed.), June 9, 2018.
Activity
 Thread view
Bioequivalence and Bioavailability Forum |  Admin contact
18,782 posts in 4,002 threads, 1,256 registered users;
online 17 (0 registered, 17 guests [including 10 identified bots]).

[…] our greatest mistake would be to forget that data
is used for serious decisions in the very real world,
and bad information causes suffering and death.    Ben Goldacre

The BIOEQUIVALENCE / BIOAVAILABILITY FORUM is hosted by
BEBAC Ing. Helmut Schütz
HTML5 RSS Feed