This question is directed at people who work with Torsten and Stan in drug development.
We are doing some work for dose justification from the Phase II study to inform dose selection in Phase III, which will be submitted to the FDA. Does anyone know if Torsten or Stan had been validated for this purpose and/or if you had any experience using the output for submissions?
Thanks!
3 Likes
You might want to ask the folks at Metrum directly. I don’t know if they follow the Stan forums. Bill Gillespie (@billg on these forums) would be a good person to ask. @charlesm93 may also know.
1 Like
I don’t have such experience but I’d be curious to know! Also pinning @yizhang.
First, I am not aware that FDA requires any formal validation of software for data analyses submitted to them. Data analyses performed with Stan by Metrum Research Group scientists have been included in submissions to the FDA and other regulatory bodies. I am not aware of any pushback from the regulators. On the other hand I don’t believe any of those analyses would be characterized as pivotal for proof of efficacy.
I have not done any work using Torsten functions that has been submitted to the FDA. I don’t know if anyone outside of MetrumRG has done so. I don’t anticipate that it would present any serious problems.
If you use Stan or Torsten for Bayesian analyses that cannot be readily performed with software more familiar to a regulatory scientist that may review a submission (e.g., NONMEM), then I would not expect much resistance.
8 Likes
Thanks, Bill. We tried doing some work in nlmixr2 and found it quite limiting, particularly when expanding the model. I am trying to avoid buying NONMEM and Monolix licenses even though the workflow will likely be faster for simple one/two-compartment models. If we learn anything relevant from this submission, I am happy to report back.
P.S. Thanks for developing Torsten! I hope someone will keep maintaining it ))
6 Likes