Log in
Register|
Helmut ★★★   Vienna, Austria, 2015-01-27 17:27 (3759 d 22:48 ago) Posting: # 14322 Views: 25,342 |
|
|
Dear all, today (at the XII Congreso de la Sociedad Española de Farmacia Industrial y Galénica in Barcelona) I had a strange encounter with Alfredo García-Arieta. I asked him whether the EMA is aware that by applying their ABEL-method for HVD(P)s the type I error might be substantially inflated.* The answer was no and that the two Lászlós showed inflation of up to 7% only for FDA’s method – which is not correct (already in their paper of 2009). Next I asked whether it would be acceptable to use a method different from the guideline (e.g., iteratively adjusted α). Answer: “No; you have to comply with what is stated in the guideline.” I couldn’t believe it and asked for confirmation whether one has to use a method which might inflate the patient’s risk to 20%. The answer was “Yes. If you want you can change the guideline.” Was too strong stuff for my poor soul. A regulator who values a lousy piece of paper more than the health of patients. I had to leave the lecture hall. Later on he confirmed that Potvin’s methods are not acceptable because the intermediate power estimation is not stated in the guideline. Lesson learned: Whatever might be published by well renowned authors in peer-reviewed journals counts (by far) less than the personal opinion of the almighty oracle speaking ex cathedra.
— 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 |
|
nobody nothing 2015-01-27 18:07 (3759 d 22:08 ago) @ Helmut Posting: # 14323 Views: 23,542 |
|
|
Viva Espania! We, the old Greek are searching for a state of mind called ataraxia, in your language best translated as "Unerschütterlichkeit", not to be confused with nowadays "coolness". I can give you a 3-day introduction course, I might even come to Vienna for that (but not in winter time  ), there will be some substantial reductions in fees, I promise...Sorry for slightly OT posting — Kindest regards, nobody |
|
ElMaestro ★★★ Denmark, 2015-01-27 18:56 (3759 d 21:20 ago) @ Helmut Posting: # 14324 Views: 23,458 |
|
|
Hi Hötzi, it is quite common that regulators have a disclaimer included when they speak somewhere in public. Their talks reflect their own opinion and not the their agency's etc etc. Which is, all things considered, perhaps comforting on the rare occasion. If you still get the regulator-induced mood swings, try and smoke some Schützomycin. — Pass or fail! ElMaestro |
|
d_labes ★★★ Berlin, Germany, 2015-01-27 20:44 (3759 d 19:32 ago) @ Helmut Posting: # 14325 Views: 23,616 |
|
|
Dear Helmut! That's "Starker Toback" (Leo translates: Strong meat)! I can't believe that  .Especially the sentence "If you want you can change the guideline." is a coup de main! What about once again shouting "You are a risk for public health"? But eventually ElMaestro's recommendation with regard to Schuetzomycin is the better choice for your personal well-being  .Remember: Regulators move in mysterious ways. And I'm meanwhile strongly convinced: It's not science what we do all days of our professional part of life. We are sheep, if you remember ... — Regards, Detlew |
|
nobody nothing 2015-01-28 10:04 (3759 d 06:11 ago) @ d_labes Posting: # 14328 Views: 23,611 |
|
|
Remember what kind of organization we are talking about, AMAZON-Juncker beeing just the latest symptom of the disease https://en.wikipedia.org/wiki/Paul_van_Buitenen https://en.wikipedia.org/wiki/Marta_Andreasen Der Fisch stinkt vom Kopf... "Andreasen was fully suspended from her job by the Commission in May 2002 (for "violating Articles 12 and 21 of staff regulations, failure to show sufficient loyalty and respect"). " "failure to show sufficient loyalty and respect"... You might be guilty, Mr. Schütz... — Kindest regards, nobody |
|
ElMaestro ★★★ Denmark, 2015-01-28 14:11 (3759 d 02:05 ago) @ nobody Posting: # 14329 Views: 23,437 |
|
|
Hi Nobody, ❝ "failure to show sufficient loyalty and respect"... You might be guilty, Mr. Schütz... You better sell your BEBAC stocks. They had a really rough day at the Wienerbörse — Pass or fail! ElMaestro |
|
Helmut ★★★ Vienna, Austria, 2015-01-30 11:55 (3757 d 04:21 ago) @ Helmut Posting: # 14331 Views: 24,271 |
|
|
Upon request of Alfredo García-Arieta I post his response below: Dear Helmut, I was informed that you posted a comment on my response to your question during the XII Congress of the Spanish Society of Industrial Pharmacy and Pharmaceutical Technology (SEFIG: Sociedad Española de Farmacia Industrial y Galénica) in Barcelona. When I read it I found your comment to be unfair. Therefore, I would appreciate if you could publish my clarification on the topic in your blog since I think that there was a misunderstanding. I acknowledged that I had not read the paper that you mentioned (Wonnemann et al. 2015). I would like to read all the papers published on bioequivalence or to follow your blog but I do not have the time. After reading it, I think that my opinion is not changed, but confirmed. I explained in Spanish during my presentation, which you could not understand because you do not speak Spanish, that Tothfalusi and Endrenyi had published (J Pharm Pharm Sci. 2009;12(1):138-49) that when the variability is around 30% the consumer risk is around 7% (i.e. we approve 7% of studies when the point estimate is 1.25 instead of 5%). In contrast, with the FDA method the consumer risk is around 14% based on their paper as far as I remember. Therefore, the EMA method is more conservative. I answered to your question that I was not aware of that paper that indicates the EMA method produces a much higher consumer risk than 7%, and that it was 7% for EMA method and for the FDA it was 14%. It seems that you understood me incorrectly when you state: ❝ I asked him whether EMA is aware that by applying their ABEL-method for HVDs/HVDPs the type I error might be substantially inflated.* The answer was no and that the two Lászlós showed inflation of up to 7% only for FDA’s method – which is not correct (already in their paper of 2009). I can understand that you were not able to listen my response properly because I was not able to understand your question initially and I asked you to repeat it. Fortunately, I have all the Spanish audience to confirm what I said. Unfortunately, in my 1-hour presentation slides I have that slide but I only had 15 minutes for my presentation and I had to remove it. I just mentioned it orally, but you do not understand Spanish. Wonnemann et al. investigate if more studies are approved with the present limits compared with the conventional 80–125%. I do not need simulations to know that when widening the limits we approve products that we do not approve with the conventional limits. That is exactly the purpose of scaling or widening. I disagree to consider that as the consumer risk. We simply changed the goalposts. And they acknowledge this because they wrote twice in the paper: “With “scaling” the rejection rate at a GMR of 1.25 cannot be called a type I error, as with scaled limits applied it is not a rejection of the null-hypothesis. However, the empirical rate allows for a direct comparison with the strategy of an unscaled bioequivalence assessment”. In summary, they acknowledge that they are doing something incorrect, but it is interesting as a comparison exercise. However, only those without any knowledge of the real life of bioequivalence assessment can consider that unscaled bioequivalence of Cmax was the regulatory criterion before the current guideline on the investigation. For those that are so naïve I would like to clarify: First, we do not widen AUC, in contrast to the FDA. Second, we use a narrower scaling for Cmax than the FDA does for Cmax and AUC. Third, we widen up to a 30% approx. in contrast to the FDA approach where there is no limit. Fourth, we ask for a clinical justification that the Cmax difference has no clinical relevance, unlike the FDA. Fifth, in the previous guideline the widening for Cmax was accepted with simply a clinical justification even for drugs with low variability and now the widening is only possible in cases of high variability and a replicate design is mandatory for widening, which usually shows that intra-subject variability of the reference product is lower than 30%. Therefore, we are now more conservative than before and than the FDA. Consequently, when we widen the acceptance range we ask for a clinical justification showing that a larger difference in Cmax (30% is the maximum widening) is not clinically relevant. Then, we all were aware in the past, at the time the guideline was discussed, and now that we are accepting larger differences than 20%. So, it is offensive to be accused that we/I are/am not worried about patients by someone that seems to be surprised when a much more permissive widening was done before the present guideline. I cannot believe that you did not know that Cmax could be widened before and that there was no maximum for widening Cmax, but we now have a limit. If you did not know this I am sorry to inform you of it, taking into account that your poor soul is so sensitive (as you wrote ironically), and all these products remain on the market based on the previous guideline that I did not write. We knew that the consumer risk was 7% approx. when CV is 30%. This is confirmed in this paper in 2015, but it is not new. However, the authors claimed: “Surprisingly, when scaled bioequivalence limits were set as bioequivalence limits, the highest rejection rate observed was at the lowest variability investigated. With CVANOVA of 30%, it was 7.05%, and it was still 5.39% with a variability of 40%”. I insist that the widening of Cmax was accepted also in the previous guideline even for low variability drugs, which meant that products with much larger differences in point estimates were accepted because their confidence intervals are narrower thanks to the low variability. In my presentation I also explained that Tothfalusi and Endrenyi had shown in 2012 that when the products are not equal (ratio T/R different from 1) the increase in variability makes the confidence interval narrower (because the scaled difference follows a non-central t-distribution where the non-centrality parameter is reciprocally proportional to variability) and then the sample size has to be lower because the power is higher. This in confirmed in this paper: “Increasing the product difference to a GMR of 1.10 resulted in empirical power values of 53.88% at 30% and 67.42% at a variability of 50%”, but we already knew it. In summary, the paper that you mentioned adds nothing to our knowledge and it should not have caused you so much disturbance. You also asked me if you could use an alternative approach that I was not able to understand. You wrote in your post ❝ (e.g., iteratively adjusted α). Frankly I do not know exactly what you mean with that term and I thought that it was not necessary to explain to you that the applicant can deviate from the guideline if justified. However, a deviation from the guideline would require that we assess the validity of the approach chosen. And it seems that this would provoke a time delay since what you propose is not something standard or something considered in the guideline. So we would have to ask our statistical experts (BSWP). In summary, I recommended you not complicate it more than necessary and to follow the guideline because, at that time, I believed that we are using a conservative approach compared with the previous requirements or the present requirements of other countries like USA or Canada. Now, after reading the paper that you mentioned, my opinion has not changed. If you like another approach is up to you to use it as explained above, but I do not recommend it to you because it is not necessary. Then, when you were excited claiming that the consumer risk can be increased up to 20% and that this evidence had been published by a member of the BSWP, I just reminded you that you can ask for revision of the guideline if you believe in that so vehemently. I cannot ask for a change because I do not believe it is necessary. My English is poor, but I hope I did not say what you wrote ❝ Yes. If you want you can change the guideline. because I do not know what it means. My intention was to say what I explained above. Next to the end I would to insist that ❝ A regulator who values a lousy piece of paper more than the health of patients is something I cannot agree with. First, the guideline is something I have to follow but applicants can deviate if justified. So for me it is like the law. I would not discredit it. I do not remember that you identified it that way during the period for comments when you had the opportunity to change it. Regarding the comparison between the paper and the health of patients I have no words to explain it. I just want to inform you that if you learned a lesson: ❝ Whatever might be published by well renowned authors in peer-reviewed journals counts by far less than the personal opinion of the almighty oracle speaking ex cathedra, I also learned another one. However, I would not like to miss this opportunity to highlight that I would not be a good regulator if I change my criterion in 5 seconds simply because someone says that someone else wrote a paper showing this or that. On the contrary, you could consider that I speak ex cathedra if I have my own criterion out of the blue, but I simply recommended you to follow the common rules that had been agreed. This is not something that I invented. By the way, I thought that you left the lecture hall because you were not going to understand a word after I answered your question since the rest of the questions from the Spanish audience were going to be in Spanish. I was not aware I had inflicted such a pain to your delicate soul. In contrast to what you think, I am very interested in patients. Regarding Potvin’s methods, that is what I have heard from the BSWP. Ask them from now on. I have asked them to make the information publically available. Best regards, Alfredo |
|
nobody nothing 2015-01-30 15:55 (3757 d 00:20 ago) @ Helmut Posting: # 14332 Views: 23,382 |
|
|
Hi all! Ehhm, do I get this completely wrong or does Fig.4, lower half in Wonnemann, 2014 tell a different story? Just asking... btw. Mr. García-Arieta, your English is actually quite good — Kindest regards, nobody |
|
Helmut ★★★ Vienna, Austria, 2015-02-01 02:35 (3755 d 13:41 ago) @ nobody Posting: # 14335 Views: 23,374 |
|
|
To whom it may concern, ❝ Ehhm, do I get this completely wrong or does Fig.4, lower half in Wonnemann, 2014 tell a different story? Just asking... Figure 4 (both parts) is misleading. Let’s concentrate on the lower part. The sample sizes (× Ntotal) are for an expected GMR of 1 and 80% power (see Table II below). These sample sizes were taken from the paper of the two Lászlós. Their table is a little bit unfortunate because sometimes sample sizes are imbalanced (e.g., 25 mod 3 ≠ 0). The filled circles give the chance to pass BE with a GMR of 1.25, which – as Alfredo correctly noted – is irrelevant if scaling is applied. Therefore, the labeling of the y-axis is Example for the partial replicate design GMR 1, ≥80% power (105 simulations for the sample size and 106 sim’s for the type I error): ![[image]](img/uploaded/image287.png) ![[image]](img/uploaded/image288.png) Maximum inflation of the TIE 0.0684 (at CV 30.1%). If one designs the study for higher power, the TIE would further increase. In studies with ≥90% power the maximum TIE would be 0.0697. As Alfredo also noted (and mentioned by the Lászlós as well) we could expect higher inflation of the TIE if the GMR moves away from 1. This time 0.9 (recommended for HVDs): ![[image]](img/uploaded/image289.png) ![[image]](img/uploaded/image290.png) Maximum inflation of the TIE 0.0716 (or 0.0727 if designed for ≥90% power). Generally slightly higher inflation is seen in fully replicated designs. Overview:
In the last case we are already close to the TIE of 1–(1–0.05)²=0.0975 expected for two simultaneous tests at α 0.05. @Alfredo: Since the inflation of the TIE not only depends on the observed CVwR but also – though to a minor degree – on the sample size (dropouts…) it is difficult to counteract the inflation by a pre-specified adjusted α. I’m not sure yet whether even Bonferroni’s 0.025 could maintain the TIE at 5% in all cases.* However, generally less adjustment will be required (especially for a higher CVwR). “Iteratively adjusted α” is a procedure to find a suitable value (given the study’s CVwR and n) which would keep the patient’s risk for the modified Null (i.e., at the scaled limits) at 5%. Only the procedure could be stated in the protocol. Hence, since the CI will be wider than the GL’s 90% CI I think that – as a conservative approach – it should be acceptable for regulators.
— 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 |
|
nobody nothing 2015-02-01 12:18 (3755 d 03:57 ago) @ Helmut Posting: # 14336 Views: 22,981 |
|
|
Hi Mr. Schütz (still have your stocks )Thank you, learned somefink... eieiei. What do you mean when you write: "...Therefore, the labeling of the y-axis is not unfortunate." — Kindest regards, nobody |
|
Helmut ★★★ Vienna, Austria, 2015-02-01 15:49 (3755 d 00:27 ago) @ nobody Posting: # 14337 Views: 23,102 |
|
|
Hi nobody, ❝ Hi Mr. Schütz (still have your stocks Wow. Didn’t know that I issued some. ❝ […] What do you mean when you write: ❝ "...Therefore, the labeling of the y-axis is not unfortunate." Dammit, at leftover from editing. Should read “… labeling of the y-axis is I updated the plots. The solid black line is at 0.05 and the dotted red line at the significance limit for 106 simulations (0.05036). Blue circles are ≤0.05036 and red circles >0.05036.
— 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, 2015-02-01 19:43 (3754 d 20:32 ago) @ Helmut Posting: # 14339 Views: 22,978 |
|
|
Hi all, a very interesting thread, but also one that has the potential go be blown very quickly out of proportions when things go personal. It is noteworthy that 7% type I error is acceptable when doing scaling in EU but not when doing other types of BE. As my brain is only walnut-sized I may have missed something but are the present explanations more or less that
Consider the requirement about as a mandatory term for stage in two-stage trials. Why? For the known methods at the time this guideline was published there was no known (published) design for which that requirement made any difference. One might think the regulators would perhaps have reviewed some designs and found them useful and where the requirement would have made sense. We'll never know. We can only watch and be gobsmacked that they in that case never made the designs visible to us. I am not in this game to hurt someone's feelings. At the end of the day I just hope the patient's risk is being controlled. It just doesn't appear to be controlled for the present regulatory version of scaling, so I naturally hope it will be fixed. Likewise I am not asking for anyone to stand up and admit a mistake was made, I am just asking for it to be fixed. Compare to individual BE and parametric population BE, two concepts that were also extremely interesting from a theoretical perspective and which mercifully were put in the bottom drawer as knowledge grew. This wasn't someone's mistake, the concepts were abandoned because they turned out not to be too useful for the purposes they were invested for even though the basic thinking was right. At the end noone was to blame. And perhaps the time has now come to take a fresh view on things. Fresh, I wrote. The fuel that drives regulatory evolution is science and -I hope- not personal pride or hurt feelings. I wish you all a good day. — Pass or fail! ElMaestro |
|
nobody nothing 2015-02-01 23:04 (3754 d 17:12 ago) @ ElMaestro Posting: # 14340 Views: 22,879 |
|
|
Hi to Danmark, wish I was there Yeah, sound scientific discussion, that would be it, regarding regulatory requirements. And that's why this consultation process was invented for EMA guidelines, I guess (politics is not my business, you can imagine  ). But did it work out well yet? We had the discussion on the scientific basis of Ctrough in the MR guideline recently. And I think there was at least one large-scale workshop/conference on this guideline, but apparently not everything was on the table (at that time?) or discussed to the very end. In my opinion the structure of a guideline should clearly state the problems the current draft is going to address, sound scientific analysis of this problem and conclusions to be drawn based on this. That's what EMA or any other agency expect you to present at an Sci. Adv. meeting, so why not doing it the same way when formulating new regulatory requirements? I think what we see currently is not state-of-the-art scientific discussion of the real problems. Why does not EMA together with external experts set up an online discussion board for new draft guidelines instead of a deadline for comments (which end nowhere?)? Everybody could participate and in the end there might be a more rational (but at least a much more transparent) evaluation of drafts. Let's have open source guidelines the sourceforge.org-style. We are in 2015... PS: I love lively discussions, as you can imagine — Kindest regards, nobody |
|
Helmut ★★★ Vienna, Austria, 2015-02-02 01:48 (3754 d 14:28 ago) @ ElMaestro Posting: # 14343 Views: 22,921 |
|
|
Hi ElMaestro, ❝ It is noteworthy that 7% type I error is acceptable when doing scaling in EU but not when doing other types of BE. Yep. Double moral standards? As Alfredo suggested I will ask the BSWP. Amazingly one of the co-authors of Wonnemann’s paper (Armin Koch) is a current member of the BSWP. ❝ … are the present explanations more or less that ❝ a. 7% is better than what the FDA came up with. At CVwR ~30% yes. In a fully replicated 4-period design FDA’s inflation of the TIE might reach 22.4% [sic]. But: With FDA’s method there is little inflation (if any) at CVwR >30% and the method becomes quickly very conservative. At ≤30% we face inflation in both methods. Below downscaled contours (EMA top, FDA bottom). Black lines enclose areas of significant inflation. ![[image]](img/uploaded/image291.png) ![[image]](img/uploaded/image292.png) ❝ b. BSWP is to blame. Not being aware of the 2009 paper of the two Lászlós? I don’t know to which degree the BSWP was involved. At least two members of the PKWP at that time published together on the topic. Maybe they didn’t consider it relevant? Can’t read tea-leaves. ❝ Consider the requirement about as a mandatory term for stage in two-stage trials. Why? For the known methods at the time this guideline was published there was no known (published) design for which that requirement made any difference. One might think the regulators would perhaps have reviewed some designs and found them useful and where the requirement would have made sense. We'll never know. We can only watch and be gobsmacked that they in that case never made the designs visible to us. Yep. On the other hand Alfredo stated above ❝ Regarding Potvin’s methods, that is what I have heard from the BSWP. […] I have asked them to make the information publically available. My emphasis. There is hope. ❝ I am not in this game to hurt someone's feelings. So am I. ❝ At the end of the day I just hope the patient's risk is being controlled. […] I am just asking for it to be fixed. Exactly. Could easily be done in the Q&A. — 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 |
|
nobody nothing 2015-02-02 09:09 (3754 d 07:07 ago) @ Helmut Posting: # 14346 Views: 22,722 |
|
|
So: make your 25-28% something formulation go with 31% in the study and you are on the lucky side of life, hu? — Kindest regards, nobody |
|
Helmut ★★★ Vienna, Austria, 2015-02-02 15:06 (3754 d 01:10 ago) @ nobody Posting: # 14353 Views: 22,603 |
|
|
Hi nobody, ❝ So: make your 25-28% something formulation… Scaling depends on the CVw of the reference. If you are able to manufacture a new product with smaller variability you get a reward in sample size (same widening of the AR but tighter CI). Example (T/R 0.9, 80% power, 4-period full replicate):
❝ …go with 31% in the study and you are on the lucky side of life, hu? If you mean that the true CVwR is 25–28% and you observe 31%, yes. You scale the AR to 79.44–125.89% although the reference is not highly variable. Lucky punch for the producer, slightly increased risk for the patient. — 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 |
|
mittyri ★★ Russia, 2015-02-03 13:44 (3753 d 02:31 ago) @ Helmut Posting: # 14361 Views: 22,789 |
|
|
Dear Helmut, Could you please help with the following questions: 1. Could you explain: ❝ At 30% the chance for such a wrong decision is 50%. Therefore, we can expect the highest inflation of the TIE at 30%. But: ❝ Maximum inflation of the TIE 0.0684 (at CV 30.1%). Why the maximum inflation is not observed exactly at 30%? 2. Where did you get the image with inflation of the TIE? (is it possible to find it in high resolution?) As I understand the TIE in this image can be changed to CI for adjusted α. (So the black line is the line of 90% CI) 3. I read this thread one more time, but didn't catch one thing For example before the study (TRT|RTR) I have estimated number of subjects 42 proceeded from expected CV (40%), required power (80%), TIE (5%) and expected Ratio (90% for HVD). OK, I'm on a lucky side, the CV and GMR were as expected. After that I should adjust the TIE (and wide the CI). And new CI could be outside the limits! Could I expect it and plan the study in accordance with coming procedure of TIE adjusting? Yes, it is not possible to counteract the inflation by a prespecified adjusted α, but as well as for power is it possible to use some "critical limits"? — Kind regards, Mittyri |
|
Helmut ★★★ Vienna, Austria, 2015-02-03 14:24 (3753 d 01:52 ago) @ mittyri Posting: # 14364 Views: 22,619 |
|
|
Dear Mittyri, ❝ 1. Could you explain: ❝ ❝ At 30% the chance for such a wrong decision is 50%. Therefore, we can expect the highest inflation of the TIE at 30%. ❝ But: ❝ ❝ Maximum inflation of the TIE 0.0684 (at CV 30.1%). ❝ (in your post below) ❝ Why the maximum inflation is not observed exactly at 30%? For the EMA scaling starts at CV >30% and for the FDA at ≥30%. Therefore, I explored 30% and the slightly larger value 30.1%. No big deal. Actually the maximum inflation for the EMA is expected at 30% + ε, where ε is infinitely small. ❝ 2. Where did you get the image with inflation of the TIE? My own simulations. ❝ (is it possible to find it in high resolution?) I can send you the originals. Take FDA’s results with a grain of salt. I have to rethink what Detlew posted. ❝ As I understand the TIE in this image can be changed to CI for adjusted α. (So the black line is the line of 90% CI) Yes and no. You can find an adjusted α and calculate the CI accordingly. The black lines enclose the area of significant inflation of the TIE (for 106 simulations that’s 0.05036). ❝ 3. I read this thread one more time, but didn't catch one thing ❝ For example before the study (TRT|RTR) I have estimated number of subjects 42 proceeded from expected CV (40%), required power (80%), TIE (5%) and expected Ratio (90% for HVD). ❝ OK, I'm on a lucky side, the CV and GMR were as expected. ❝ After that I should adjust the TIE (and wide the CI). And new CI could be outside the limits! Yes, that’s possible. If you plan the study for α 0.05 and adjust afterwards you might loose up to ~10% power. ❝ Could I expect it and plan the study in accordance with coming procedure of TIE adjusting? Yes, that is what I would do. Find an adjusted α based on a “best guess” CV and estimate the sample size accordingly. It seems that 0.025 for fully replicated designs and 0.0304 for the partial replicate will be conservative (i.e., cover the inflation at 30%). ❝ Yes, it is not possible to counteract the inflation by a prespecified adjusted α, but as well as for power is it possible to use some "critical limits"? Not sure what you mean by that. Run some sim’s. For really high CVs (say >40%) this entire story is not important anyway. — 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, 2015-02-03 13:02 (3753 d 03:13 ago) @ Helmut Posting: # 14359 Views: 22,817 |
|
|
Dear Helmut ❝ ... ❝ ❝ … are the present explanations more or less that ❝ ❝ a. 7% is better than what the FDA came up with. ❝ ❝ At CVWR ~30% yes. In a fully replicated 4-period design FDA’s inflation of the TIE might reach 22.4% [sic]. But: With FDA’s method there is little inflation (if any) at CVWR >30% and the method becomes quickly very conservative. At ≤30% we face inflation in both methods. ❝ ... ❝ FDA ❝ Did you consider the paper Davit et al. "Implementation of a Reference-Scaled Average Bioequivalence Approach for Highly Variable Generic Drug Products by the US Food and Drug Administration" AAPS Journal, Vol. 14, No. 4, December 2012 in the pictures? This paper wants us to neglect the discontinuity at CV~30% in looking at the GMR's we assume for bioinequivalence. Instead we are forced to look at widening implied limits from s0=0.25 on (CV~25.4%). Then the inflation at CV=30% hokus pokus vanishes. Left some small inflation around s0. F.i.: library(PowerTOST)BTW: Anybody out there to shed some light on the arguments of that paper? Desired consumer risk model (!), distinct from implementation ...  Is it sound or only an attempt to answer discussions about alpha-inflation? — Regards, Detlew |
|
Helmut ★★★ Vienna, Austria, 2015-02-03 13:22 (3753 d 02:54 ago) @ d_labes Posting: # 14360 Views: 22,652 |
|
|
Dear Detlew, ❝ Did you consider the paper ❝ Davit et al. (2012) ❝ in the pictures? No. ❝ This paper wants us to neglect the discontinuity at CV~30% in looking at the GMR's we assume for bioinequivalence. Instead we are forced to look at widening implied limits from s0=0.25 on (CV~25.4%). Then the inflation at CV=30% hokus pokus vanishes. I used
I don’t see the inflation vanishing. On the contrary:
❝ BTW: Anybody out there to shed some light on the arguments of that paper? ❝ Desired consumer risk model (!), distinct from implementation ... Not me. I’m as well…❝ Is it sound or only an attempt to answer discussions about alpha-inflation? Duno. Talked about the inflation of the TIE with Barbara – but in May 2013. — 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, 2015-02-03 14:07 (3753 d 02:09 ago) @ Helmut Posting: # 14362 Views: 22,573 |
|
|
Dear Helmut ❝ I used ❝ ❝ ❝ ❝ I don’t see the inflation vanishing. On the contrary: ❝ ❝ That's the knackpunkt: They force us not to look at theta0=0.8 or 1.25 but if sW>0.25 at exp(+-sW*ln(1.25)/0.25) i.e. at 0.7694922 | 1.299558 if CV=30%. See Fig. 2 and the paragraph "Controversies" of that paper.And then hokus pokus: power.RSABE(CV=0.30, n=24, theta0=1.299558, design="2x3x3", nsims=1e6)— Regards, Detlew |
|
Helmut ★★★ Vienna, Austria, 2015-02-03 14:41 (3753 d 01:34 ago) @ d_labes Posting: # 14365 Views: 22,724 |
|
|
Dear Detlew, ❝ That's the knackpunkt: They force us not to look at theta0=0.8 or 1.25 but if sW>0.25 at ❝ ❝ And then hokus pokus: ❝ ❝ Heck, the “implied BE limits”!1 THX, for reminding me (the 10th time?)… I think that this is really hokus pokus!2 Wish to have more spare-time.
— 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 |
|
nobody nothing 2015-01-31 12:25 (3756 d 03:50 ago) @ Helmut Posting: # 14334 Views: 23,169 |
|
|
Hi again! I wanted to learn something about the New Zealand approach to HVD mentioned in Wonnemann, 2014 (largely identical to Wonnemann, 2015 I guess, haven't read it completely yet). Unfortunately I could not find something at the NZ reg homepage, except for a brand new BE guideline, which, however, mentions HVD only in the context of some "isotope application or replicate design" (emphasize by me) studies, apparently without providing further details: http://www.medsafe.govt.nz/regulatory/Guideline/GRTPNZ/bioequivalence-of-medicines.pdf However, it apparently allows the use of sequential design "for studies expected to require a large number of subjects" (p. 20/44). But it states that it is "(n)ormally not practical to use more than 40 subjects in a bioavailability study", without any justification for this statement... This appears pretty close to the design outlined in Wonnemann, 2014 for NZ, but no statistical details are provided in the NZ guideline. Edit: Subject line changed. [Helmut] — Kindest regards, nobody |
|
Helmut ★★★ Vienna, Austria, 2015-02-01 17:08 (3754 d 23:08 ago) @ nobody Posting: # 14338 Views: 23,032 |
|
|
Hi nobody, ❝ I wanted to learn something about the New Zealand approach to HVD mentioned in Wonnemann, 2014 (largely identical to Wonnemann, 2015 I guess, haven't read it completely yet). There is just one paper. The epub ahead of print of July 2014 and the published version of January 2015. ❝ Unfortunately I could not find something at the NZ reg homepage,… Wonnemann’s ref. (30) point to the 5th edition of 2001 which is no more online. If you want I can send you a copy. ❝ …except for a brand new BE guideline,… That’s the 6th edition of 2014. ❝ …which, however, mentions HVD only in the context of some "isotope application or replicate design" (emphasize by me) studies, apparently without providing further details: ❝ http://www.medsafe.govt.nz/regulatory/Guideline/GRTPNZ/bioequivalence-of-medicines.pdf No changes it section 2.5.5 anyhow. ❝ However, it apparently allows the use of sequential design "for studies expected to require a large number of subjects" (p. 20/44). But it states that it is "(n)ormally not practical to use more than 40 subjects in a bioavailability study", without any justification for this statement... OK, this section is a little bit weired. First they tell us that a study should be designed with sufficient power “normally ≥80%”. Fine. In line with ICH E9 “The number of subjects in a clinical trial should always be large enough to provide a reliable answer to the questions addressed”. Now NZ states “If the calculated number of subjects appears to be higher than is ethically justifiable, it may be necessary to accept a statistical power which is less than desirable. Normally it is not practical to use more than about 40 subjects in a bioavailability study.” On the contrary! It’s not ethical if a regulator suggests to perform a study in a sample size which will have low power to demonstrate BE. I suspect that n=40 was derived from T/R 0.95, CV 30%, 80% power in a 2×2 crossover. In a 4-period fully replicated study the same values would allow for a CV of 44%. For a CV of 63% and 40 subjects power would already be <50% [sic]. Toss a coin instead! Cheaper. Talking about sequential designs the GL is ambiguous:
— 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 |
|
nobody nothing 2015-02-01 23:08 (3754 d 17:08 ago) @ Helmut Posting: # 14341 Views: 22,832 |
|
|
Hi Felix What was the exact wording for this "40-volunteers is too much" thing in the 5th edtition of this guideline? Or send it to me and I will post it here... Schönsonntachabend — Kindest regards, nobody |
|
Helmut ★★★ Vienna, Austria, 2015-02-02 00:54 (3754 d 15:22 ago) @ nobody Posting: # 14342 Views: 22,803 |
|
|
Hi nobody, ❝ What was the exact wording for this "40-volunteers is too much" thing in the 5th edtition of this guideline? The entire paragraph of the 6th ed, p20 was exactly the same in the 5th (section 15.5.5 p143). If the calculated number of subjects appears to be higher than is ethically justifiable, it may be necessary to accept a statistical power which is less than desirable. Normally it is not practical to use more than about 40 subjects in a bioavailability study. Where calculations suggest that an excessive number of subjects is required, clinical efficacy and/or safety studies are an alternative. The use of a co-administered active ingredient labelled with non-radioactive isotope as a reference, or studies in which treatments are replicated within each subject, may improve discriminatory power for highly variable medicines. In the 6th ed. you find also If the calculated number of subjects appears to be higher than is ethically justifiable, it may be necessary to accept a statistical power which is less than desirable. It is usually not ethically justifiable to use more than about 40 subjects. on p32. Same text in the 5th, p153.Personal remark: If >40 subjects in a PK study are not ethical, I don’t comprehend how “clinical efficacy and/or safety studies“ (with extreme sample sizes!) could be “an alternative”. There are very few examples of PD endpoints having less variability than PK. A surrogate – or even more a true – clinical endpoint? Wellington, show me an example! — 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 |
|
nobody nothing 2015-02-02 09:07 (3754 d 07:09 ago) @ Helmut Posting: # 14345 Views: 22,889 |
|
|
I don't know any PD endpoint validated for systemic BE, especially with clearly established sensitivity. And to be true I think there never will be some. And this FDA model for locally applied (dermal) corticosteroids is a nightmare. So NZ is not a model to look for... Too bad... — Kindest regards, nobody |
Ing. Helmut Schütz