Log in
Register|
weirddude100 ● 2010-09-21 20:13 (5745 d 00:39 ago) Posting: # 5926 Views: 8,683 |
|
|
Suppose we have 2x2 crossover design with 2 periods, and sample size is given 12 subjects in each seq. (total N=24). How do i calculate power to detect 20% difference between A and B in AUC ? what std. deviation should i use ? please provide me formula or SAS code! Thanks. seq1 seq2 period1(AUC) period2(AUC)Edit: Category changed. See also the Forum's policy. [Helmut] |
|
jdetlor ☆ 2010-09-21 21:14 (5744 d 23:38 ago) @ weirddude100 Posting: # 5927 Views: 7,575 |
|
|
Dear WeirdDude! ❝ Suppose we have 2x2 crossover design with 2 periods, and sample size is given 12 subjects in each seq. (total N=24). How do i calculate power to detect 20% difference between A and B in AUC ? please provide me formula or SAS code! Check out: Section 7.3 - 'Power and sample size for ABE in the 2x2 design' of B. Jones, M.G. Kenward, "Design and Analysis of Cross-Over Trials", Second edition, Chapman & Hall/CRC 2003. This section contains the formulas and SAS code for calculating power. ❝ what std. deviation should i use ? You will have to provide the standard deviation as determined from other sources. Holding the sample size and ratio fixed (@ 24 subjects and a 20% difference as you mentioned), the larger the standard deviation you use, the lower the power. It should probably be mentioned that with a 20% difference it is very likely your power won't be any where near a level that is acceptable. — J. Detlor |
|
Helmut ★★★   Vienna, Austria, 2010-09-28 19:50 (5738 d 01:01 ago) @ jdetlor Posting: # 5944 Views: 7,150 |
|
|
Dear J. Detlor! ❝ It should probably be mentioned that with a 20% difference it is very likely your power won't be any where near a level that is acceptable. That’s an euphemism.  At the borders of the acceptance range, power is exactly alpha (or 5 % in the common setting) – by definition. I guess Weirddude (what a nick!) was copypasting from pre-Schuirmann’s ages (1987) - the infamous “Power Approach”… You can give it a try with D. Labes’ package PowerTOST for R. Don’t try that what standard software – most likely you would get an error at it’s best. — Dif-tor heh smusma 🖖🏼 Довге життя Україна! ![[image]](https://static.bebac.at/pics/Blue_and_yellow_ribbon_UA.png) Helmut Schütz ![[image]](https://static.bebac.at/img/CC by.png) The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
jdetlor ☆ 2010-09-28 20:37 (5738 d 00:14 ago) @ Helmut Posting: # 5947 Views: 7,657 |
|
|
Dear HS! ❝ That’s an euphemism. I find this type of dialog is best suited for converations between statisticians and professionals considering study size  ❝ At the borders of the acceptance range, power is exactly alpha (or 5 % in the common setting) - by definition. I agree with your statement regarding the null hypothesis (specifically the null hypothesis for the lower bound), but we have to be careful here — I believe weirddude specified a 20% difference, which could mean an expected ratio of 120%. With enough subjects, the technical requirements for BE could be demonstated, but I believe this is what is referred to as 'forcing' BE. Some would say a ratio of 120% suggests the formulations are not bioequivalent. To be specific, because we are dealing with TOST (two one-sided tests), to test the type I error we would set delta equal to either the lower bound (ln(0.8), or the upper bound (ln(1.25). Either of these would produce an alpha of at most 5%, which gives us our (100% - 2*alpha) confidence interval for BE. — J. Detlor |
|
Helmut ★★★ Vienna, Austria, 2010-09-28 20:57 (5737 d 23:54 ago) @ jdetlor Posting: # 5948 Views: 7,117 |
|
|
Dear J. Detlor! ❝ I believe weirddude specified a 20% difference, which could mean an expected ratio of 120%. With enough subjects, the technical requirements for BE could be demonstated, but I believe this is what is referred to as 'forcing' BE. Some would say a ratio of 120% suggests the formulations are not bioequivalent. Yes, I’ve heard the term ‘forced BE’ also. If WeirdDude really talked about a ratio of 1.20 – well, let’s see whicht power we would get (24 subjects, usual settings,  ): CV% powerUnless one has to deal with the ‘wonder-drug’ (CV 5.5 %), 24 subjects at a ratio of 1.20 are futile. ❝ To be specific, […] Exactly. — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
jdetlor ☆ 2010-09-28 22:28 (5737 d 22:23 ago) @ Helmut Posting: # 5949 Views: 7,116 |
|
|
Dear HS! ❝ Yes, I’ve heard the term ‘forced BE’ also. If WeirdDude really talked about a ratio of 1.20 – well, let’s see what power we would get (24 subjects, usual settings, ❝ ❝ ❝ ❝ ❝ ❝ Unless one has to deal with the ‘wonder-drug’ (CV 5.5 %), 24 subjects at a ratio of 1.20 are futile. Don't get me wrong — I am not advocating that planning a study with 20% is by any means acceptable. But, what I am saying is that technically it is possible to demonstrate BE with a ratio of 120%. If you were to plot the observed power vs the p-value, we would only need a power of approximately 34%[1] to have the 90% BE confidence interval to fall within the BE limits, which puts us (according to the table) at a CV of about 11%. [1] See Figure 1 in The Abuse of Power: The Pervasive Fallacy of Power Calculations for Data Analysis — J. Detlor |
|
Helmut ★★★ Vienna, Austria, 2010-09-29 00:33 (5737 d 20:18 ago) @ jdetlor Posting: # 5950 Views: 7,073 |
|
|
Dear J. Detlor, … and don’t get me wrong too – I’m with you in all your points stated. BTW, Hoenig’s & Heisey’s paper is one of my favorites. — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
d_labes ★★★ Berlin, Germany, 2010-09-27 10:51 (5739 d 10:01 ago) @ weirddude100 Posting: # 5936 Views: 7,226 |
|
|
Dear Weirddude! Please make your homework first.
BTW: The search field is located in the right upper corner of the forum's page  .— Regards, Detlew |
Ing. Helmut Schütz