9

I have worked on clinical trials in FDA submissions for many years. I use SAS almost exclusively. Recently I discussed a consulting job I had to bootstrap a Deming regression. Bootstrapping is much more easily done in R I think and several experts have suggested ways that I can do this in R. Since this is going into a resubmission to the FDA for my client I am concerned as to whether software validation might become an issue. I hear that R is gaining acceptance with the FDA and experts tell me that traceability which is an attractive feature with SAS can be dealt with easily in R as well.

My client did the original Deming regression with a package that I am not familiar with and validation was not an issue with the original submission. I want to help without adding any new headaches for them.

Can the R experts here give me some guidance? If it will help I can mention some of the programming options that were suggested?

Gala
  • 8,501
Michael R. Chernick
  • 42,857
  • 2
    Not a direct answer, but re bootstrapping in SAS, do you know of David Cassell's paper "Don't be Loopy"? – Peter Flom Oct 03 '12 at 12:28
  • 7
    This is a bit open ended, and I'm not sure what you expect to hear that hasn't already been said in other questions, R vs SAS, why is SAS prefered by private companies? and Is the R language reliable for the field of economics?. In particular, the comments on your answer to the SAS v R question seem sufficient to answer this (especially the Revo R blog post cited). – Andy W Oct 03 '12 at 13:11
  • 7
    http://www.r-project.org/doc/R-FDA.pdf might be useful. – Laurent Oct 03 '12 at 13:31
  • 1
    Thanks for all the comments thus far. I will check out all the links! @PeterFlom I don't know that paper. Does it have something to do with whether or not to imbed one procedure in another? Anyway I would be curious to see it. Can you provide a link? Personally I think the bootstrap has a useful role to play in comparing models as in variable selection in logistic regression that Gail Gong studied in the 1980s. I think it can be another way of doing sensitivity analysis – Michael R. Chernick Oct 03 '12 at 16:23
  • Hi @MichaelChernick don't be loopy – Peter Flom Oct 03 '12 at 16:26
  • @Laurent Thanks for that reference. My mention of R vs SAS was mainly just a concern. I know that knowledgeable people have written about how just about anything can be done reasonably in R and I know it has been debated on this site. I may have commented on this myself and had R programmers tell me that things like traceability can be handled in R. My understanding of the FDA review process is that the lead reviewer has a lot of authority and does not always appreciate the statistical issues and can make illogical rulings. – Michael R. Chernick Oct 03 '12 at 16:30
  • Companies can overcome these problems but it can be more of a headache than they are willing to take on. I think in my consulting situation they preferred to go with the flow rather than challenge and do the correct analysis. – Michael R. Chernick Oct 03 '12 at 16:32
  • @AndyW I recall my involvement in the post from the first link and that is why I said that the experts gave me the impression that R can do everything in a way that is/should be acceptable to the FDA. I trusted the comments and did not followup with the links provided since it wasn't my question and I wasn't as vested in the answers as I am now! Thanks for bringing it all to my attention. – Michael R. Chernick Oct 03 '12 at 16:38
  • I think the comments address a lot of my concerns. What could be left for an answer is a description of what my client would need to do to validate the R solution I give them. Is there anything in the federal regulations or FDA Guidances that can support a waiver of validation of software (just in case the issue is raised in the next submission round)? – Michael R. Chernick Oct 03 '12 at 16:41

1 Answers1

3

We have done tons of submission to the FDA with R. The Agency do not prefer a software versus the other.

Samer
  • 39