Log in
Register|
Coja66 ☆ 2011-03-28 16:22 (5562 d 02:10 ago) Posting: # 6823 Views: 11,878 |
|
|
Hi all, Can anyone advice me on how to calculate a sample size for a Bioequivalence study based on PD parameter FFE24 (Fecal fat excretion over 24 hours), in a three-way crossover design (1 test, 2 references) for a HVD (high variable drug) ? It has probably something to do with a dose-scale method (according to EMA documentation)? I am totally not familiar with these kind of analyses. Thanks in advance for your helpful responses ! Edit: Copied from a follow-up post (2011-03-29 12:57) below. [Helmut] Guess my question is either not very easy, or not very interesting  While trying to figure it out myself, I have already come to the conclusion to advice the client to go for a 6x3 design, instead of a 3x3 which was requested. I am however still learning about this dose-scale method. I think the sample size calculation may be quite similar to the 'normal' PK BE calculation, only to be based upon another (PD) parameter. Although bootstrapping did kick in in one of the articles, so my confidence in this matter is still not high  |
|
Helmut ★★★   Vienna, Austria, 2011-03-29 16:34 (5561 d 01:59 ago) @ Coja66 Posting: # 6826 Views: 10,603 |
|
|
Dear Coja! Please don't fork. You can edit your own post within 24 hours. ❝ Guess my question is either not very easy, or not very interesting The former, I would say. ❝ [...] I have already come to the conclusion to advice the client to go for a 6x3 design, instead of a 3x3 which was requested. I like that. Concerning the actual sample size method, I'm afraid just a few people have done that for such a design... We are posting in our freetime, so please be patient.  — 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 |
|
Coja66 ☆ 2011-03-29 17:31 (5561 d 01:01 ago) @ Helmut Posting: # 6828 Views: 10,557 |
|
|
Dear Helmut, Thanks for pointing out the 'forking' - will not do it again. Concerning the question: I thought it was me having a peanut-sized-brain and not being able to come up with the answer myself. It's a relief to hear from the master himself that this is a tricky analysis ! I will wait patiently  , and in the meantime try to come up with a satisfying answer myself.Thanks for the confirmation concerning the 6x3 design. |
|
Dr_Dan ★★ Germany, 2011-03-30 12:13 (5560 d 06:19 ago) @ Coja66 Posting: # 6832 Views: 10,440 |
|
|
Dear Coja66 In order to understand your problem additional information is needed. First HVD: regarding PK or PD? Highly intra-subject variability regarding pharmacokinetic parameters do not necessary mean highly intra-subject variability regarding pharmacodynamic outcome although in general pharmacodynamic parameters are more variable than kinetic. What is the range of the PD parameter? Second: Your study, is it a proof-of-concept study or a pivotal BE study? I personally would not recommend to test two test formulations in a pivotal study. In case of highly variable drugs I recommend to use a full replicate design. In general a 6x3 way design is not possible due to blood loss, but in your case it is different. Increasing the numbers of periods also means to increase the probability of drop outs. Kind regards Dan — Kind regards and have a nice day Dr_Dan |
|
Coja66 ☆ 2011-03-30 13:36 (5560 d 04:56 ago) @ Dr_Dan Posting: # 6833 Views: 10,538 |
|
|
Dear Dan, Thanks for the response - it's really appreciated. The client in this matter has requested me to come up with an idea on the design as well as the sample size for a Phase-I study, so proof-of-concept. The medication concerns Orlistat, which is known as highly variable on PK. I am not sure for PD (fat excretion in this case), but from literature search this seems to have a CV of around 35%. Ofcourse most interest is in T vs R1 and T vs R2. From guideline perspective, FDA recommends a bootstrapping method for the analysis, EMA advices either Schuirmann or the dose-scale method (using a Emax model). I dit not find anything on design nor on sample size or power calculations. For design, I was quite confident that 6x3 would be best, although your point of dropout would be a concern (not to mention the costs for the client  ). Another possibility would be to have the design similar to: Gr1 Gr2 Gr3for testing T vs R1 in crossover for period 1/2, Gr1/2, and for testing T vs R2 in crossover for period 2/3, Gr2/3. But I suppose there are a lot of snags involved in a design like this. My current view would be to calculate the sample size based on the ratio of ED50's for test and reference which should fall within the 80-125% range of bioequivalence, but then I only have it for the 2x2 case of test versus reference. I suppose this is in line with the dose-scale method of EMA, however the study to be performed is going to be FDA-submitted.  Note: in the clinical literature found sofar on this matter, no one bothers with a proper sample size calculation - most are based upon 'empirical considerations'. The used sample sizes vary between 18 and 48... Any alternative views or ideas on the above are highly appreciated. Best regards from sunny Netherlands. |
|
ElMaestro ★★★ Denmark, 2011-03-30 15:48 (5560 d 02:45 ago) @ Coja66 Posting: # 6834 Views: 10,491 |
|
|
Hey Coja66, Gr1 Gr2 Gr3I would do that! ❝ for testing T vs R1 in crossover for period 1/2, Gr1/2, and for testing T vs R2 in crossover for period 2/3, Gr2/3. But I suppose there are a lot of snags involved in a design like this. Sorry, that doesn't compute. Could you explain your thinking? Sample size is pretty unexplored territory as far as I know for this method. I'll try and ask W.H. later today if he knows how to do it. I would imagine that sample size methods would only exist for parametric approaches. ❝ Best regards from sunny Netherlands. Best regards from a cold and grey Washington. — Pass or fail! ElMaestro |
|
Helmut ★★★ Vienna, Austria, 2011-03-30 15:59 (5560 d 02:33 ago) @ ElMaestro Posting: # 6836 Views: 10,333 |
|
|
Hi ElMaestro! ❝ Sample size is pretty unexplored territory as far as I know for this method. I'll try and ask W.H. later today if he knows how to do it. <off topic> Can you ask him also about Potvin's sequential design paper (he is the corresponding author)? Are they exploring the possibility of a full adaptive design (i.e., correct not only for the CV, but also for the PE observed in stage 1)? </off topic>— 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 |
|
ElMaestro ★★★ Denmark, 2011-03-30 16:54 (5560 d 01:38 ago) @ Helmut Posting: # 6838 Views: 10,373 |
|
|
Dear HS, your wish is my command. Anything else, I can do for you (no requests for foot masssage are currently accepted)? |
|
Helmut ★★★ Vienna, Austria, 2011-03-30 17:05 (5560 d 01:27 ago) @ ElMaestro Posting: # 6839 Views: 10,365 |
|
|
My Capt'n! ❝ your wish is my command. Great! ❝ Anything else, I can do for you (no requests for foot masssage are currently accepted)? Yessir. One thing always puzzles me with the sequential design. It would be nice to have the first stage as a Williams’ design with one reference and two candidate test formulations and continue only with one of them in the second stage. α should be preserved, I would say – because the second test goes to the waste bin and will not be marketed. Of course a common variance is assumed. Foot massage in one of Budapest’s famous baths in May? |
|
ElMaestro ★★★ Denmark, 2011-03-31 21:55 (5558 d 20:37 ago) @ Coja66 Posting: # 6846 Views: 10,311 |
|
|
I asked a bit around; talked to a few statisticians in the public system as well as from private companies doing bioassays. Noone had a sample size method for the three-point approach. Only suggestion was to give a statistician a pen and some paper and leave him/her for two weeks undisturbed in the hope she or he at the end will surface with some nifty equations. |
|
jdetlor ☆ 2011-04-01 17:52 (5558 d 00:40 ago) @ ElMaestro Posting: # 6848 Views: 10,298 |
|
|
Hello all! In the mean time, would this be a situation where empirical power simulations would be appropriate? Of course initial parameter estimates would be needed. I could see it becoming quite computationally expensive if bootstraping the final confidence interval is also required (especially the bootstrap BCa CI limits). J. Detlor |
|
ElMaestro ★★★ Denmark, 2011-04-01 18:19 (5558 d 00:13 ago) @ jdetlor Posting: # 6850 Views: 10,252 |
|
|
Hi jdetlor, ❝ In the mean time, would this be a situation where empirical power simulations would be appropriate? Absolutely. How would you specify the model here? — Pass or fail! ElMaestro |
|
jdetlor ☆ 2011-04-01 18:56 (5557 d 23:36 ago) @ ElMaestro Posting: # 6851 Views: 10,282 |
|
|
Hi ElMaestro! ❝ Absolutely. How would you specify the model here? I maybe missing something here, but from my understanding, we are not working with a linear ANOVA model in terms of a cross-over analysis, but with the Emax model and estimating the 'f' parameter or the bioavailability (see the orlistat FDA individual guidance). The cross-over design is used because it would still maintain balance within periods across treatments. Edit: Guidance linked. [Helmut] |
|
ElMaestro ★★★ Denmark, 2011-04-01 19:14 (5557 d 23:18 ago) @ jdetlor Posting: # 6852 Views: 10,273 |
|
|
Hi jdetlor, ❝ I maybe missing something here, but from my understanding, we are not working with a linear ANOVA model in terms of a cross-over analysis, but with the Emax model and estimating the 'f' parameter or the bioavailability ... yes, this isn't an lm. Let's say your have an idea of variability within and between (i guess one could start out with gaussian variability but everything is possibly, especially since the FDA does not like parametrics for this analysis), and some guesses for ED50 and f etc. How do you incorporate the relevant variabilities to simulate the data? — Pass or fail! ElMaestro |
|
jdetlor ☆ 2011-04-01 23:51 (5557 d 18:41 ago) @ ElMaestro Posting: # 6853 Views: 10,228 |
|
|
❝ Let's say your have an idea of variability within and between (i guess one could start out with gaussian variability but everything is possibly, especially since the FDA does not like parametrics for this analysis), and some guesses for ED50 and f etc. How do you incorporate the relevant variabilities to simulate the data? The between and within variabilites are incorporated to the Emax model the same way they are to the linear model. The difference is instead of a response variable summing up effects linearly, the between and within subject variabilty are added to the non-linear Emax model. Categorical index variables for treatment and dose are also coded. It is assumed the errors are normal, which may not be a terrible assumption since they are modeled as such. So for example, when creating the response variable for the test treatment at dose 0, you would sum your Eo parameter, your between-subject variability, and your within-subject variability (as the non-linear portion is 0). Continuing on at non-zero dose levels, the summation for the response variable would be the Eo parameter, your between-subject variability, your non-linear function (using Emax, 'f', ED50 parameters), and your within-subject variability. |
Ing. Helmut Schütz