Log in
Register|
Shuanghe ★★ Spain, 2013-06-14 12:55 (4327 d 21:00 ago) Posting: # 10788 Views: 32,007 |
|
|
Hi all, I just found this interesting article yesterday written by Barbara Davit and other 3 FDA staff from Division of Bioequivalence II. "Expectations of the US-FDA regarding dissolution data in generic drug regulatory submissions" Most interesting part for me is the following phrases in the section "Dissolution profile comparison": "...For circumstances where wide variability is observed, a statistical evaluation of f2 metric (an f2 confidence interval) can be calculated using a bootstrapping approach19. According to the dissolution guidance, "Model Independent Multivariate Confidence Region Procedure" is suggested for comparison of dissolution profiles in instances where within batch variation is more than 15% CV20. Recent data from ANDA submissions lead the DB to believe that this method is likely to be less discriminating and sensitive than the f2 test. Therefore, the DB recommends the applicant use the f2 confidence interval approach when the mean data cannot be used for the f2 test as a result of high dissolution variability. ..." Well, given that so many people were asking here about Mahalanobis calculation etc I guess this is an interesting turn of event. I'm no statistician but I'd say bootstrap f2 should be much easier than multivariate statistical distance method (or so I thought, see below) Now comes my "hidden agenda"  I used to do f2 in M$ Excel with "eyeball" as to which data point should be included etc. for 1000 bootstrap dataset that's not an option. I'm currently struggling with SAS to apply the f2 criteria such as at least 3 point, no more than 1 point > 85% etc. It's much harder than I thought.  Any help here would be really appreciated. Also, given that FDA's encouraging this method it would be nice to have R package/macro to do that. Any R-guru?  All the best Shuanghe — All the best, Shuanghe |
|
d_labes ★★★ Berlin, Germany, 2013-06-14 15:48 (4327 d 18:08 ago) @ Shuanghe Posting: # 10791 Views: 28,880 |
|
|
Dear Shuanghe, I have some spare documented and un-validated code with respect to this, written a time ago (2011). But it's too long to post here and I haven't enough spare time to convert it into a package, i.e. do all the documentation necessary to be accepted on CRAN. If you (or others) are interested, send me a mail. But as Helmut always states "Take the code with a grain of salt". It comes without any warranty not to damage your computer .@Ace: Are you there? — Regards, Detlew |
|
Helmut ★★★   Vienna, Austria, 2013-06-14 16:30 (4327 d 17:26 ago) @ d_labes Posting: # 10793 Views: 28,813 |
|
|
Dear Detlew, ❝ @Ace: Are you there? His last login was already a while ago. Shall I send him an e-mail invitation? — 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 |
|
Shuanghe ★★ Spain, 2013-06-14 17:13 (4327 d 16:43 ago) @ d_labes Posting: # 10795 Views: 28,946 |
|
|
Thanks a lot Detlew! ❝ If you (or others) are interested, send me a mail. I've sent an email to you by the Forum. ❝ But as Helmut always states "Take the code with a grain of salt". It comes without any warranty not to damage your computer I'll take it with "much more than one grain of salt", I promise.  by the way, what do you think of the following step for SAS: 1. PROC SURVEYSELECT to generate 1000 dataset with 12T and 12R each. 2. obtain mean T and R at each time point. 3. apply f2 criteria of >= 3 points, no > 1 point with >85% dissolution etc 4. calculate f2 5. 5% percentile as lower limit of 90% confidence interval. — All the best, Shuanghe |
|
d_labes ★★★ Berlin, Germany, 2013-06-14 17:55 (4327 d 16:00 ago) @ Shuanghe Posting: # 10797 Views: 28,893 |
|
|
Dear Shuanghe! ❝ by the way, what do you think of the following step for SAS: ❝ ❝ 1. PROC SURVEYSELECT to generate 1000 dataset with 12T and 12R each. ❝ 2. obtain mean T and R at each time point. ❝ 3. apply f2 criteria of >= 3 points, no > 1 point with >85% dissolution etc ❝ 4. calculate f2 ❝ 5. 5% percentile as lower limit of 90% confidence interval. Sorry, but I'm not familiar with that PROC SURVEYSELECT. If it creates the bootstrap samples from your original data the code looks correct. — Regards, Detlew |
|
yjlee168 ★★★  Kaohsiung, Taiwan, 2013-06-16 00:17 (4326 d 09:39 ago) @ Shuanghe Posting: # 10803 Views: 28,666 |
|
|
Dear all, Helmut recently introduced a free open-source program called "PhEq_bootstrap". Please see this thread. I don't know if this program fits your needs or not. Program description from SOURCEFORGE: This program was developed as a help in establishing pharmaceutical equivalence by use of FDA f2 coefficient. It was designed to help with f2 computation in cases when intra- and inter-batch variability is large, namely RSD>10%. The use of statistical bootstrap technique allows to implement confidence interval (CI) into the f2 coefficients resulting in overcoming of their major drawback in the original metrics. The algorithm provides possible “worst case scenario” of f2 values, thus supporting claim about pharmaceutical equivalence. The target users are researchers from industry and academia dealing with pharmaceutical equivalence problem....— All the best, -- Yung-jin Lee bear v2.9.2:- created by Hsin-ya Lee & Yung-jin Lee Kaohsiung, Taiwan https://www.pkpd168.com/bear Download link (updated) -> here |
|
sameep ☆ India, 2015-01-07 07:08 (3756 d 01:47 ago) @ yjlee168 Posting: # 14233 Views: 25,710 |
|
|
❝ Helmut recently introduced a free open-source program called "PhEq_bootstrap". Please see this thread. I don't know if this program fits your needs or not. Dear Friends, There is an option in DDSolver to calculate bootstrap F2. The results obtained using DDSolver and PhEq_Bootstrap are coming different. Sample DDSolver results: Observed f2 - 41.690 Number of bootstrap - 10000 Bootstrap mean - 40.393 Bootstrap median - 40.416 5% percentile - 34.344 95% percentile - 46.290 Is 5% percentile > 50 - No Similarity of R and T - Reject Sample PhEq_Bootstrap results: Number of bootstrap - 10000 Average F2 - 40.34 F2 unbiased value - 42.46 Expected F2 - 38.62 Lower confidence interval for expected f2 - 33.42 Upper confidence interval for expected f2 - 43.62 Similarity NOT confirmed: Lower CI is below the limit (f2 = 50) Now the question is, which one of the two software's perform "FDA acceptable" calculation? Here, i am assuming that "Shah VP, Tsong Y, Sathe P, Liu JP. In vitro dissolution profile comparison-statistics and analysis of the similarity factor, f2. Pharm Res. 1998 Jun;15(6):889-96." method is "FDA acceptable". OR is there any other method to calculate the bootstrap F2 (I am no SAS brainer  )?If SAS can be used, does any one have the code for it? Thanks in advance, Sameep |
|
AngusMcLean ★★ USA, 2013-06-18 18:06 (4323 d 15:50 ago) @ Shuanghe Posting: # 10811 Views: 27,882 |
|
|
Shuanghe: There is exactly such an R package and it is up on R. It was written by a Prof. in Poland. It is based on an article with Vinoh Shah's name (bootstrapping e.t.c).  You can get f2 and the confidence intervals.I can look it up for you if you are interested.? Angus |
|
d_labes ★★★ Berlin, Germany, 2013-06-19 10:20 (4322 d 23:35 ago) @ AngusMcLean Posting: # 10813 Views: 27,816 |
|
|
Dear Angus! ❝ ... There is exactly such an R package and it is up on R. It was written by a Prof. in Poland. Where did you found such a R-package .All my searches on CRAN or via  didn't succeed.— Regards, Detlew |
|
AngusMcLean ★★ USA, 2013-06-19 23:51 (4322 d 10:05 ago) @ d_labes Posting: # 10832 Views: 27,732 |
|
|
Hello: here is the reference as promised. http://sourceforge.net/projects/pheqbootstrap/ Please let me know how you get on with this program. The author is a very hard working fellow with excellent programming skills. Angus |
|
d_labes ★★★ Berlin, Germany, 2013-06-20 10:18 (4321 d 23:38 ago) @ AngusMcLean Posting: # 10833 Views: 27,625 |
|
|
Dear Angus ❝ here is the reference as promised. ❝ http://sourceforge.net/projects/pheqbootstrap/ Ok, that's the reference Helmut has originally given and Yung-jin has referrred to. But although hosted on sourceforge and is freeware PhEq_bootstrap has nothing to do with R. Neither is it a R-package nor it is implemented in that programming language. Just to cite from the sourceforge page of PhEq_bootstrap: "It was developed in Lazarus environment, therefore source code is available in ObjectPascal." (Emphasis by me) — Regards, Detlew |
|
ElMaestro ★★★ Denmark, 2013-06-20 14:45 (4321 d 19:10 ago) @ d_labes Posting: # 10838 Views: 27,703 |
|
|
Hi all, it can't be a lot of code can it (how the heck can the exe be 4 Mb)? Just wanted to check it, but when opening the zip-archive from SourceForge I learned that "source code is available" does not mean that "source code is available". Oh well...  — Pass or fail! ElMaestro |
|
yjlee168 ★★★ Kaohsiung, Taiwan, 2013-06-20 15:21 (4321 d 18:35 ago) @ ElMaestro Posting: # 10839 Views: 27,666 |
|
|
Dear ElMaestro, I think you may look the file called 'PhEq_boostrap_src.tar.gz' under 'file' of SF link. See this. Download it and unzip it. I find it is a project folder containing some files under it. So you need to install Lazarus compiler (> 100 MB, it's free pascal compiler/IDE) first. Then you can see all source codes from Lazarus IDE to open the project files. Apparently, Lazarus does not save all codes as ASCII type text files, but binaries. So we cannot read these source code files with a plain ASCII editor. However, under Lazarus, we can see all source codes. Interesting compiler. When I install Lazarus few minutes ago, I see a message from installation windows, quoted ... - How to reduce the exe file size? The default binaries are very big because they include debug information. For release builds you can switch this off in the Project settings. I don't know if the authors switch the setting off or not for their final release.[edited: yes, the authors did switch the debug setting off for release builds. It can be checked from Lazarus - 'Project - options' after opening the project.]
![[image]](img/uploaded/image62.jpg) ❝ it can't be a lot of code can it (how the heck can the exe be 4 Mb)? Just wanted to check it, but when opening the zip-archive from SourceForge I learned that "source code is available" does not mean that "source code is available". — All the best, -- Yung-jin Lee bear v2.9.2:- created by Hsin-ya Lee & Yung-jin Lee Kaohsiung, Taiwan https://www.pkpd168.com/bear Download link (updated) -> here |
|
AngusMcLean ★★ USA, 2013-06-20 17:29 (4321 d 16:27 ago) @ d_labes Posting: # 10840 Views: 27,568 |
|
|
I have a email from the author October 2012: he wrote he was going to put it up on R. So it seems that he has not done it as yet. I will ask him. Would you prefer it up on R? Angus |
|
AngusMcLean ★★ USA, 2013-06-22 02:47 (4320 d 07:08 ago) @ d_labes Posting: # 10859 Views: 27,576 |
|
|
The authors from Poland have published a paper on f2 with bootstrapping. See reference, but he never put it up on R. ....my apologies. http://www.dissolutiontech.com/DTresour/201302Articles/DT201302_A02.pdf Angus |
|
Helmut ★★★ Vienna, Austria, 2013-06-22 04:57 (4320 d 04:59 ago) @ AngusMcLean Posting: # 10861 Views: 27,505 |
|
|
Angus, do you never follow links other people give you as a matter of principle? I posted the reference already in April and both Yung-jin and Detlew linked to it in two different posts above. — 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 |
|
AngusMcLean ★★ USA, 2013-06-22 17:57 (4319 d 15:58 ago) @ Helmut Posting: # 10864 Views: 27,404 |
|
|
Helmut: Thank you for your note. I see your reference links. Usually I do follow the links. In this case I did not, since I abruptly left this site went back to my email to find the R reference. {He indicated to me that it was going to be put up on R}. I asked him and he sent me the reference in the dissolution journal as I now see you already have posted some time ago. Best Regards, Angus |
Ing. Helmut Schütz