BE-proff ● 2016-06-02 16:35 (3210 d 00:55 ago) Posting: # 16382 Views: 10,784 |
|
Dear All, Do you make references to calculations done by PowerTOST in documents submitted to regulatory authorities (i.e. study protocols, clarifications, etc)? I mean phrases like "sample size was calculated with PowerTOST bla-bla". ![]() AFAIK, authorities want to see validated software while R-based scripts seem not to be validated... How to prove officials that PowerTOST is reliable soft? ![]() Thx. Edit: Category changed; see also this post #1. [Helmut] |
d_labes ★★★ Berlin, Germany, 2016-06-02 17:05 (3210 d 00:25 ago) @ BE-proff Posting: # 16384 Views: 9,461 |
|
Dear BE-proff, ❝ AFAIK, authorities want to see validated software while R-based scripts seem not to be validated... ❝ How to prove officials that PowerTOST is reliable soft? I'm not a friend of software validation, whatever that is. Ask the same question to statements like: "I have calculated the sample size with SAS, PASS, nQuery, Brain-Pencil-Paper ... (or whatever soft-/wetware you prefer ![]() See also this thread. Then act. Others, more trained in this ugly field of software validation may give you more hints what to do in detail. In short:
BTW: PowerTOST is not an R-based script. It is a package providing you the possibility (via functions) to estimate power / sample size with your R-based script. — Regards, Detlew |
Helmut ★★★ ![]() ![]() Vienna, Austria, 2016-06-02 17:39 (3209 d 23:52 ago) @ BE-proff Posting: # 16385 Views: 9,453 |
|
Hi BE-proff, ❝ Do you make references to calculations done by PowerTOST in documents submitted to regulatory authorities (i.e. study protocols, clarifications, etc)? Yes, I do. In tricky cases (e.g., justification of an adjusted α in two-stage design) I give my script in an appendix to the protocol. ❝ How to prove officials that PowerTOST is reliable soft? Hart is’ hart und weich is’ weich. My translation: Hard is hard and soft is soft. — 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 |
DavidManteigas ★ Portugal, 2016-06-03 13:05 (3209 d 04:26 ago) (edited on 2016-06-03 13:54) @ BE-proff Posting: # 16389 Views: 9,453 |
|
Hi, As far as I understand, regulatory authorities in Europe are not that troublesome with software as FDA. They don't even ask the sponsor to send the datasets of the study or the Logs of the statistical software execution. I believe that in some countries, you could state in the Protocol and in the SAP that you will used Winonlin and SAS in your analysis and use R in all of them and never have a problem in submission's nor in inspections. For instance, regulatory inspections in Portugal care only about GCP compliance in the clinical site. They don't even check in detail eDC software or data management & analysis process. Lack of qualified resources I think. Nevertheless, my personal opinion is that you should always "validate" yourself the results of the software. For instance, when I use a new R package to implement a new statistical method I always validate the results with SAS, being the job for regulatory purposes or not. I also do that when I'm doing an analysis I'm not very comfortable with and I store both outputs (R and SAS) for my own protection in case of questions (the main trouble with R arises from the clients, much of them used to SAS and don't want to take the risk of using R just by "superstition"). Regards, David Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Ohlbe] |
BE-proff ● 2016-06-04 00:30 (3208 d 17:01 ago) @ DavidManteigas Posting: # 16392 Views: 9,318 |
|
Hi again, I'll make some clarifications. In Russia we have two options to prove sample size: either demonstrate calculations with formulas or provide literature reference. The 1st option is not for me because I don't have relevant background so I'd prefer the 2nd way. So, if somebody could share any literature reference in a journal (preferrable) which mentions PowerTOST I'd really appreciate it ![]() Thx |
Helmut ★★★ ![]() ![]() Vienna, Austria, 2016-06-04 17:51 (3207 d 23:39 ago) @ BE-proff Posting: # 16399 Views: 9,470 |
|
Hi BE-proff, ❝ In Russia we have two options to prove sample size: either demonstrate calculations with formulas … See this presentation (slide 8). The procedure is always iterative (since the t-value in the formula to calculate power depends on the degrees of freedom and hence, the sample size itself). The approximation1 by the shifted central t-distribution can be done with a pocket calculator (alternatively: paper-pencil-brain) and a t-table. The better approximation (noncentral t) already needs software, since no tables exist for a range of the noncentrality parameter. The exact method needs numeric integration of two integrals → software. ❝ … or provide literature reference. What do you mean by that? ❝ The 1st option is not for me because I don't have relevant background so I'd prefer the 2nd way. ❝ ❝ So, if somebody could share any literature reference in a journal (preferrable) which mentions PowerTOST I'd really appreciate it Jones & Kenward2 The power and sample size for a bioequivalence trial can conveniently be calculated using the R package PowerTOST (Labes (2013)).a PowerTOST is mentioned in Nadarajah et al.3 In the folder \library\PowerTOST\tests you find R-scripts to validate the various sample size functions against published data. See the man-pages of ct5.1 , ct5.2 , ct5.3 , ct5.4.1 , ct9.6.2 , ct9.6.4 , ct9.6.6 , ct9.6.8 , ctCW.III , ctSJ.VIII.10 , and ctSJ.VIII.20 for the references.
Stat Meth. 2015;22:23–46. doi:10.1016/j.stamet.2014.07.002. ![]() — 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 |
BE-proff ● 2016-06-07 17:15 (3205 d 00:16 ago) @ Helmut Posting: # 16405 Views: 9,129 |
|
Hi Helmut, Thank you very much for your help! ![]() |
ElMaestro ★★★ Denmark, 2016-06-04 18:12 (3207 d 23:18 ago) @ BE-proff Posting: # 16400 Views: 9,300 |
|
Hi BE-proff, as far as I know PowerTOST and other packages will generally be accepted. Power is usually the sponsor's problem: It is the chance of showing BE when a set of assumptions are met. It usually does not affect patient's risk and so isn't where you get the most deficiency letters. At the end of the day, your assumptions are guess and guesses tend to have a lot of uncertainty, so power is in that perspective not a question of having a boatload of decimals right. Besides there is no regulatory power minimum or maximum though 80% or thereabouts is quite common. Regulatory questions -or inspection triggers- may arise from bad guesses; for example it may be somewhat unethical to assume a CV of 8% and a GMR of 100.00% for most drugs (you'd be including too few patients for a realistic chance of showing BE). Thus, what the industry generally needs is a healthy dose of common sense rather than validation of power-calculating software to the 17th decimal. It is difficult to validate someone's ability to guess well anyhow. — Pass or fail! ElMaestro |
BE-proff ● 2016-06-07 17:17 (3205 d 00:13 ago) @ ElMaestro Posting: # 16406 Views: 9,121 |
|
Hi ElMaestro, We will try to use PowerTOST with literature references in protocol. Time will show if such option suits our regulators... |
mittyri ★★ Russia, 2016-06-08 09:51 (3204 d 07:40 ago) @ BE-proff Posting: # 16414 Views: 9,118 |
|
Hi BE-proff, ❝ Time will show if such option suits our regulators... We used it without any questions from experts. Good luck! — Kind regards, Mittyri |
Astea ★★ Russia, 2016-06-14 19:41 (3197 d 21:50 ago) @ mittyri Posting: # 16429 Views: 8,972 |
|
I join the opinion of mittyri. Until now, there were no pretensions in this regard. — "Being in minority, even a minority of one, did not make you mad" |