Log in
Register|
Tina ★ India, 2014-05-16 13:38 (4000 d 10:38 ago) Posting: # 12955 Views: 7,173 |
|
|
Dear all, I have a very basic question in statistics.
Tina Edit: Category changed. [Helmut] |
|
ElMaestro ★★★ Denmark, 2014-05-16 15:34 (4000 d 08:42 ago) @ Tina Posting: # 12959 Views: 6,302 |
|
|
Hello Tina, Power is the chance of showing BE if your assumptions are correct. The assumptions relate to the actual GMR and its CV (and distribution of the residual but we can forget that in this context). If the GMR is 0.9 rather than 0.95 then you need more subjects in order to achieve a given (desired) level of certainty that the CI becomes narrow enough to meet BE. If you are extreme un-realistic (let's say you assume GMR=1.0 but in reality some other data -perhaps from a previous study etc- suggest it is 0.5 or whatever) then the true power is exceedingly low and the study may be wasted, unethical. The IEC/IRB should capture that. If you have confidence in 0.95 then use that, no issue at all. But if the true GMR is worse (more different from 1.0) then the true power is lower than you wish for. Note also that the true GMR is never known. You can estimate it in a trial but it is still just an estimate. Secondly, in some cases there is no in vitro method (proper dissolution etc) that gives good info in advance about the prospective GMR. — Pass or fail! ElMaestro |
|
Ohlbe ★★★ France, 2014-05-16 16:27 (4000 d 07:49 ago) @ Tina Posting: # 12961 Views: 6,298 |
|
|
Dear Tina, In addition to ElMaestro's response: ❝ (95 to 105) vs (90 to 110)—Which assumption (for calculating sample size) is preferred by the regulators What the regulators are interested in is the result of the 90 % CI you will calculate at the end. They may take a look at your sample size calculation assumptions in order to check the study is not overpowered and you didn't "force" bioequivalence, but that's probably as far as they will go. The choice is fully yours. ❝ If regulators are fine with both, can we not take the option of (95 to 105) as it helps with lesser costs due to lesser sample size Sure. But if you calculate your sample size based on 95-105, and the study fails because you get a point estimate of 92 in your study, you won't have saved any money. You have to find the right balance between risks, ethical considerations, and costs. — Regards Ohlbe |
|
Tina ★ India, 2014-05-16 19:39 (4000 d 04:38 ago) @ Ohlbe Posting: # 12962 Views: 6,148 |
|
|
Dear ElMaestro and Ohlbe, Thanks for the clear explanation. I wouldn't take the risk of 95-105 considering the possibility of the final value to go outside the assumption. For Potvin B method for calculation, should the expectation also be 90 to 110? Stage II sample size is calculated when power is less than 80%. Why do we say that the study has failed at the completion of stage I if the power is more than or equal to 80%? Kind regards, Tina |
|
ElMaestro ★★★ Denmark, 2014-05-16 20:31 (4000 d 03:46 ago) @ Tina Posting: # 12963 Views: 6,353 |
|
|
Hello Tina, ❝ For Potvin B method for calculation, should the expectation also be 90 to 110? GMR=0.95 for planning of stage 2. ❝ Stage II sample size is calculated when power is less than 80%. Why do we say that the study has failed at the completion of stage I if the power is more than or equal to 80%? First of all, that's just how the authors defined the method. Second, imagine that you have N1=24 at the first stage but you aren't BE yet. Now you derive power and find out it is larger than 80%, and if you want to achieve just above 80% then you need N1=20. It means you should not include any more subjects; from that perspective it would make no sense not to stop. What else could you realistically do? Finally, the whole business with power of the Potvin methods is very funny, if not funky or even freaky. To derive power and find a sample size for stage 2 you must identify the CV or variance. That variance is conditional on and associated with the GMR point estimate. It is the treatment effects (the GMR so to say) that determines the CV, not the other way around. Then that GMR point estimate is immediately disregarded: you just assume it is 0.95 with Potvin B and C, but you keep and use the CV that was conditional on the GMR being something else. Highly weird indeed, but at least it provides some scientist with an opportunity to wonder, speculate and do their own research. There is an enormous need for improved methods in this area. We have not seen the last papers in this area - at least that's why my crystal ball says. — Pass or fail! ElMaestro |
|
d_labes ★★★ Berlin, Germany, 2014-05-18 16:40 (3998 d 07:37 ago) @ ElMaestro Posting: # 12964 Views: 6,197 |
|
|
Dear ElMaestro! ❝ ... That variance is conditional on and associated with the GMR point estimate. It is the treatment effects (the GMR so to say) that determines the CV, not the other way around ... I think you err here. AFAIK the treatment effect (GMR) and the residual variance (CV) are independent. — Regards, Detlew |
Ing. Helmut Schütz