Log in
Register|
beman ☆ 2011-08-22 17:34 (5009 d 00:42 ago) Posting: # 7286 Views: 7,888 |
|
|
Dear All, according to the new EMA-Guideline we want to perform a study in 2-Stage Design. But we want to do it in an 'extreme' way, which is not explicitly forbidden in the Guideline. Stage 1: 12 Subject, 'nearly full penalty for the interims analysis' ⇒ 99.9% CI positive stopping criteria: BE passed (never happens) positive stopping criteria: calculated sample size for stage 2 is more than 60 subjects (due to ethical reasons). Stage 2: amount of subjects to reached 85 % Power based on the results of stage 1, maximum 60 subjects, 'nearly no penalty' ⇒ 90.1% CI Questions: What do you think about such an approach? Do you have any experience with authority concerning this issue? Does the penalty assignment to the stages keeps the 'overall' type 1 error of the study under control? Do you know literature about an alpha-spending function, which i can use to calculate such extreme alpha spending described above? Thanks BEman Edit: Category changed. [Helmut] |
|
ElMaestro ★★★ Denmark, 2011-08-22 18:58 (5008 d 23:18 ago) @ beman Posting: # 7287 Views: 6,823 |
|
|
Hello BEman, ❝ What do you think about such an approach? ❝ Do you have any experience with authority concerning this issue? ❝ Does the penalty assignment to the stages keeps the 'overall' type 1 error of the study under control? ❝ Do you know literature about an alpha-spending function, which i can use to calculate such extreme alpha spending described above? This is not a good approach ethically. Even though you may have control over your alpha the power becomes easily surprisingly low if you implement stopping rules based on stage 2 size. I think such trials are possibly futile and thus not ethical. I am not sure regulators know this (yet?) because it requires simulation software to figure it out. There is no published paper about it as far as I know. — Pass or fail! ElMaestro |
|
Helmut ★★★   Vienna, Austria, 2011-08-22 19:35 (5008 d 22:41 ago) @ beman Posting: # 7288 Views: 6,861 |
|
|
Dear BEman! ❝ […] we want to do it in an 'extreme' way, which is not explicitly forbidden in the Guideline. Right. ❝ What do you think about such an approach ? Personally I’m not happy with it – see also ElMaestro’s comments. ❝ Do you have any experience with authority concerning this issue? According to hearsay (!) similar approaches were accepted in the past – but I have never seen an actual example. ❝ Does the penalty assignment to the stages keeps the 'overall' type 1 error of the study under control? As ElMaestro said: Without (exhaustive!) simulations we simply don’t ‘know’. For an overview of α-inflations see the references given the section IIa of Schwartz & Denne (2003).* ❝ Do you know literature about an alpha-spending function, which i can use to calculate such extreme alpha spending described above? No. A major pitfall is the way the Null-hypothesis is formulated in BE (H0 bioinequivalence, Ha bioequivalence). The abundance of literature on sequential / adaptive designs deals with the much more common superiority test (H0 no difference, Ha difference) – which is the other way ’round. I would be very wary to simply apply any of these methods in an other context. The only papers I’m aware of are Gould (1995), Potvin et al. (2008) and Montague et al. (2011). For the references search the forum.
— 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 |
|
beman ☆ 2011-08-23 14:37 (5008 d 03:39 ago) @ Helmut Posting: # 7289 Views: 6,743 |
|
|
❝ ❝ What do you think about such an approach? ❝ ❝ Personally I’m not happy with it – see also ElMaestro’s comments. First of all: Thank you ElMaestro and HS. I understand your thought about such an approach. But let me argue in an other way: I can do 'nearly' exacty the same divided in a pilot and a pivotal study. First I perform the pilot study with 12 subject. After that i use the internal criteria: calculated sample size for stage 2 is more than 60 subjects (due to ethical reasons) → reformulation of test product, otherwise perform pivotal study with the calculated sample size. Only two difference are between this approach and the 2-Stage-Design approach
Stage 1 is only performed to get CV's for the power and sample size estimation of stage 2 – this is one of the fundamental ideas of a two-stage design. Best regards BEman |
|
d_labes ★★★ Berlin, Germany, 2011-08-23 16:51 (5008 d 01:25 ago) @ beman Posting: # 7294 Views: 6,700 |
|
|
Dear BEman ❝ Because I never reach BE in stage 1 (CV to high, CI will be wider than the acceptance range) I shouldn't have an alpha-inflation. Never is not a valid term to use in statistics  .If you perform a 2-stage study with 12 subjects in the 'pilot' stage 1 and your intra-subject CV comes out between 10%-15% and your point estimate is ok, your sample size estimation for the total sample size will give you N=8...12 (BE acceptance range 0.8 ... 1.25, GMR=0.95, alpha=0.049 Haybittle-Peto, target power=80%). I.e. you have supposedly reached BE in stage 1 with with a probability bordering on certainty  .— Regards, Detlew |
|
d_labes ★★★ Berlin, Germany, 2011-08-23 15:46 (5008 d 02:30 ago) @ beman Posting: # 7292 Views: 6,823 |
|
|
Dear BEman, ❝ according to the new EMA-Guideline we want to perform a study in 2-Stage Design. But we want to do it in an 'extreme' way, which is not explicitly forbidden in the Guideline. Not only not forbidden but explicitly allowed. Just to cite: "For example, using 94.12% confidence intervals for both the analysis of stage 1 and the combined data from stage 1 and stage 2 would be acceptable, but there are many acceptable alternatives and the choice of how much alpha to spend at the interim analysis is at the company’s discretion." ❝ What do you think about such an approach? Many sponsors seem to favourite such extreme settings. Seems the provision of practical "no penalty" is very attractive for them .❝ Do you have any experience with authority concerning this issue? None up to now. ❝ Does the penalty assignment to the stages keeps the 'overall' type 1 error of the study under control? See comments of EM and Helmut. Simulations necessary! Potvin et.al. mention in their introductory text a paper of Hauck et.al. (see below) in which for the O'Brian-Fleming nominal alphas an overall alpha inflation (up to 6%) was supposedly observed. I don't have that paper so I can't check the truth and what they have done exactly. ❝ Do you know literature about an alpha-spending function, which i can use to calculate such extrem alpha spending described above ? Your settings are comparable to the Haybittle-Peto nominal alphas (0.001 stage 1, 0.049 stage 2) or the O'Brian-Fleming nominal alphas (0.005 stage 1, 0.048 in stage 2). But remember these alphas are derived for superiority trials with a parallel-group design and mid-course interim evaluation. Hauck WW, Preston PE, Bois FY. — Regards, Detlew |
Ing. Helmut Schütz