olacy
☆    

Hungary,
2007-02-02 16:10
(6264 d 10:03 ago)

Posting: # 503
Views: 5,641
 

 Number of subjects [Power / Sample Size]

Dear All,

How can I argue for the increased number of subjects in case of a highly variable drug? To set the power to 90% (from 80%) or should I refer to the very high CV% (for example 45%, power=80%)?
Which is less risky? I think the latter one.
I think that the power could not be changed subsequently if the biostudy failed because of 90% power but would pass with 80%.
Please confirm and advise.
When is it needed to increase the power to 90%?
Thanks in advance!

Best regards,
olacy
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2007-02-07 19:23
(6259 d 06:49 ago)

@ olacy
Posting: # 509
Views: 4,602
 

 80% or 90% power?

Dear olacy!

❝ How can I argue for the increased number of subjects in case of a highly variable drug?


Because you will need them to demonstrate BE? :-D

❝ To set the power to 90% (from 80%) or should I refer to the very high CV% (for example 45%, power=80%)?


Now I’m confused :confused:
For a HVD/HVDP it’s difficult enough to show BE even with a power of 80%. Since the CV is not carved out of stone (where did it come from: literature, a pilot study, etc.), it’s not unwise to do some kind of “pre-study sensitivity-analysis”, which is also recommended in ICH-E9: Statistical Principles for Clinical Trials (Section 3.5, 3rd paragraph).
One natural characteristic of HVDs/HVDPs is also the variability in the point estimate between studies; I would never assume a deviation of only ±5% (which is often used in sample size estimation) from the reference.

❝ Which is less risky? I think the latter one.


If you have no financial limits, everything is possible; but since it’s difficult enough to get the job done, increasing power will be the last (and rarely feasible) step.
I think it’s more important to look at a potential change in the point estimate (for CV=30%: PE -5% → -10%, n: 40 → 80), than to increase power for a fixed PE of -5% (for CV=30%: 80% → 90%, n: 44 → 52).
You should also pay attention to the CV itself: CV 35% instead of 30% increases the sample size also dramatically (-5%: 40 → 52, -10%: 80 → 106).

If you have a very high CV% (like the 45% of your example), a standard 2×2 study is almost futile.
I would recommend a replicated design (and go with the reference-scaled ABE approach).

If you are working at a CRO and want to sell “a big study” to the sponsor, it will be shortsighted to argue with “increased power” because the decreased producer’s risk (beta) is built on sand (namely that all your assumptions on CV and PE will hold).

❝ I think that the power could not be changed subsequently if the biostudy failed because of 90% power but would pass with 80%.


A posteriori power does not exist; see:

Hoenig JM, Heisey DM. The Abuse of Power. The Pervasive Fallacy of Power Calculations for Data Analysis. Am Stat. 2001;55(1):19–24. doi:10.1198/000313001300339897. [image] free resource.


Even if all your assumptions were justified in the biostudy, you have a chance of exactly 1–power of failure.
That’s another interpretation of the point estimate and the 90% confidence interval:
If you repeat a study 19 times, in 1 study (1/20=0.05) the point estimate will fall outside the confidence interval of the first study, and the confidence limit will be outside the acceptance range in 4/20 studies (20×[1-0.8]).

❝ When is it needed to increase the power to 90%?


If you want to be sure… :-D
But especially for HVDs/HVDPs you should work rather on the foundations than on the roof of your house first.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
olacy
☆    

Hungary,
2007-02-15 11:41
(6251 d 14:31 ago)

@ Helmut
Posting: # 525
Views: 4,404
 

 80% or 90% power?

Dear Helmut!

Thanks for the very elaborate answer!

Best regards,
olacy
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
72 visitors (0 registered, 72 guests [including 6 identified bots]).
Forum time: 02:13 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