Helmut ★★★ Vienna, Austria, 2012-10-05 17:59 (4443 d 08:06 ago) Posting: # 9317 Views: 15,891 |
|
Dear all, interesting story. FDA approved generics to Wellbutrin XL 300 mg based on BE of the 150 mg strength (safety concerns of the higher dose in healthy subjects). Soon after approval (in 2006) post marketing reports showed efficacy/safety issues. 2007 FDA requested a BE study of the 300 mg strenght in patients who experienced problems after switching to the generic. Teva began the study but terminated it in 2011 due to recruitment difficulties. FDA sponsored at BE study in healthy subjects (ethical?), which showed lacking BE in August 2012. Quote: The Agency determined that it was important to conduct direct bioequivalence studies comparing Budeprion XL 300 mg to Wellbutrin XL 300 mg, in light of the reports that patients and health care professionals had submitted to FDA. FDA believed that the optimal patient population for those studies would be patients who had reported a lack of efficacy or unwanted side effects after switching from Wellbutrin XL 300 mg to Budeprion XL 300 mg. The rationale for this was two-fold. First, the Agency believed that such a study would better target the reported problems and thereby provide the best information as to why the drug did not appear to be effective. Second, because that patient population was taking the 300 mg strength of the drug already, this study design would not unnecessarily expose healthy adult volunteers to the higher strength of the drug, particularly given concerns that this strength might be associated with risk of seizures. As a consequence the rating was changed from AB to BX, patients taking Budeprion XL 300 mg as a substitute for Wellbutrin XL 300 mg should talk with their health care professionals if they have questions about taking this medication. Impax/Teva is withdrawing Budeprion XL 300 mg from the market. The API-specific guidance was removed form FDA’s website. Lessons learned: “Waiving up” is a risky business and should be avoided. References:
P.S.: Amazing how the FDA plots a plasma profile (equidistant x-axis instead of a time scale): http://www.fda.gov/AboutFDA/CentersOffices/OfficeofMedicalProductsandTobacco/CDER/ucm153270.htm. — 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 |
jag009 ★★★ NJ, 2012-10-06 00:30 (4443 d 01:35 ago) @ Helmut Posting: # 9319 Views: 12,193 |
|
ElMaestro ★★★ Denmark, 2012-10-06 01:35 (4443 d 00:30 ago) @ Helmut Posting: # 9324 Views: 11,959 |
|
Great story, Helmut, thanks for posting. What surprises me somewhat is this:
Of course, if the stats obtained with N=24 are correct then the chance of showing bioinequivalence is higher than the chance of showing bioequivalence, that's clear (and yes, how to calculate the sample size when the previous T/R was lower than 80% is an issue, but let's not go there). Bioinequivalence hasn't been shown until the CI is entirely out of the acceptance range, and as I read the announcement from FDA they also do not use wording suggesting that the products are truly inequivalent. In stead they just correctly suggest that BE has not been shown. Another case of 'aggregate weight of evidence', I imagine. — Pass or fail! ElMaestro |
Helmut ★★★ Vienna, Austria, 2012-10-06 03:14 (4442 d 22:51 ago) @ ElMaestro Posting: # 9325 Views: 13,200 |
|
Hi ElMaestro! ❝ A BE study generally has three possible outcomes: Bioeequivalent, bioinequivalent (CI entirely outside acceptance range) or inconclusive (overlap with the acceptance range, meaning not proven bioinequivalent and not proven bioequivalent). Statistically speaking there are just two outcomes. The Null is bioinequivalence, which is either rejected (consumer risk with a probability of α) or not. The “grey zone” is related to the producer’s risk (with a probability of 1–β), which is important in study planning. Post hoc we get only this ‘bad luck’ feeling (in German: dumm gelaufen). ❝ If I were the FDA and had the funds I would certainly run a new study with more subjects in order to reach a firm conclusion. Given the safety profile of budeprion I wouldn’t do that (at least not in healthy subjects as the FDA did in their study). For the sample size see below. I’m wondering why the FDA asked Teva to run a study in patients who showed safety/efficacy problems after switching from Wellbutrin XL 300 mg. It should have been foreseeable that they will run into massive recruitment problems. Why didn’t the FDA ask for a simple BE study in patients already back in 2007? ❝ […] how to calculate the sample size when the previous T/R was lower than 80% is an issue, but let's not go there. Of course conventional sample size estimation will not work. Maybe Detlew can consider adding sampleN.inf() to PowerTOST . But leaving power aside and assuming that we can exactly reproduce the ratio (75.2%) and CV (29.8%) of Cmax it would need 227 (!) subjects in order to squeeze the upper CL below 80% (Null not rejected, bioinequivalent).❝ Bioinequivalence hasn't been shown until the CI is entirely out of the acceptance range, and […] they also do not use wording suggesting that the products are truly inequivalent. In stead they just correctly suggest that BE has not been shown. Yep. ❝ Another case of 'aggregate weight of evidence', I imagine. Makes sense since already in 2007 they had 85 supportive cases of therapeutic inequivalence (though anecdotal). Likely much more as of now. — 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, 2012-10-07 15:11 (4441 d 10:54 ago) @ Helmut Posting: # 9330 Views: 11,697 |
|
Hi HS, ❝ Statistically speaking there are just two outcomes. The Null is bioinequivalence, which is either rejected (consumer risk with a probability of α) or not. The “grey zone” is related to the producer’s risk (with a probability of 1–β), which is important in study planning. Post hoc we get only this ‘bad luck’ feeling (in German: dumm gelaufen). Accept the null hypo or not, yes that dicotomic, but that's not in itself what distinguishes between bioequivalent or bioinequivalent (although it usually is what determines regulatory aceptance of a submited dosssier, but this isn't the situaiton here). A product is in practice one of these two options but weare not able to tell which one if we end up with a CI that has includes areas inside and outside the acceptance range like we saw here. In perspective, consider the ability to repeat studies. In my experience with regulators they accept repeat of a study when it is inconclusive (or some would say under-powered; but again the joker here is when the PE is outside the acceptance range - I shall not say anything generally about who accepts what in such situations). I am not afraid of saying the FDAs study here was a little underpowered since it did not tell if the product was equivalent or inequivalent. It showed a tendency, possibly, nothing more than that in my opinion. Despite alpha and beta there is still a chance that the product is BE on basis of FDA's result, cf. the repeat issue. In a country where every there is a lawyer for every 270 ordinary citizens, I understand why a regulatory agency might wish to safeguard against lawsuits just in case. I would however, be happier to see such reactions based on conclusions rather than trends and tendencies. — Pass or fail! ElMaestro |
d_labes ★★★ Berlin, Germany, 2012-10-09 10:50 (4439 d 15:15 ago) @ Helmut Posting: # 9351 Views: 11,579 |
|
Hi Gents! ❝ Of course conventional sample size estimation will not work. Maybe Detlew can consider adding Bug or a feature? library(PowerTOST) ❝ But leaving power aside and assuming that we can exactly reproduce the ratio (75.2%) and CV (29.8%) of Cmax it would need 227 (!) subjects in order to squeeze the upper CL below 80% ... Code from this thread (replacing the calls to internal PowerTOST functions to PowerTOST:::func_to_call, f.i. PowerTOST:::.design.no(design) ) gives me 228 .— Regards, Detlew |
Helmut ★★★ Vienna, Austria, 2012-10-09 15:38 (4439 d 10:26 ago) @ d_labes Posting: # 9355 Views: 11,579 |
|
Hi Detlew! ❝ Bug or a feature? power.noninf(alpha=0.05, CV=0.298, margin=0.8, theta0=0.752, n=277) Feature! I didn’t realize that it’s possible to enter the house through the back door. Of course one can cheat as well (exchanging the limit with the ratio): sampleN.noninf(alpha=0.05, CV=0.298, theta0=0.8, margin=0.752) ❝ Code from this thread (replacing the calls to internal PowerTOST functions to PowerTOST:::func_to_call, f.i. Hhm, can you be a little bit more specific (= for dummies like me)? — 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, 2012-10-09 17:58 (4439 d 08:07 ago) @ Helmut Posting: # 9359 Views: 11,543 |
|
Dear Helmut! ❝ ❝ # [1] 0.8010803 Remark 2: Meanwhile after thinking about I would call it a bug because power.noninf() claims that it calculates the power of a non-inferiority test. But it does this only for theta0 >= margin if margin is <1 and for theta0 <= margin if margin is >1. Otherwise the power of an 'inferiority' test is calculated (exchanged Null and Alternative). I consider to 'repair' the code. ❝ ... I didn’t realize that it’s possible to enter the house through the back door. Of course one can cheat as well (exchanging the limit with the ratio): ❝ Kuuhl! ❝ ❝ Code from this thread (replacing the calls to internal PowerTOST functions to PowerTOST:::func_to_call, f.i. ❝ ❝ Hhm, can you be a little bit more specific (= for dummies like me)? Your number came from a sample size planning via CI half-width if I'm right (trial to a given precision). Ben has developed an R function ss_ci() to do this in the mentioned thread. He borrowed some code from PowerTOST, namely the (internal) functions .design.no() and .design.props() . These functions are since the introduction of a mandatory NAMESPACE for packages no longer accessible in simple way from outside the package. Therefore you will get an error "function not found" if you run the code unchanged. You have to change the code and call the functions via the package:::name syntax. Type help(":::") in your R console to get some more information.Specific enough for dummies? — Regards, Detlew |
Helmut ★★★ Vienna, Austria, 2012-10-09 21:51 (4439 d 04:14 ago) @ d_labes Posting: # 9363 Views: 11,541 |
|
Dear Detlew! ❝ Your number came from a sample size planning via CI half-width if I'm right (trial to a given precision). Right guess. ❝ Ben has developed an R function […] ❝ ❝ Specific enough for dummies? Borderline. — 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 |