[R] Regulatory Computing with R

From: <anthony.rossini_at_novartis.com>
Date: Tue, 21 Aug 2007 12:30:51 +0200


Dear all -

Cody and Bert have some amusing points.

The problems with R that Cody states are no different than those that any organization has with any programming work. Period.

We've mostly solved them through appropriate approaches, addressing through quality management, some of the issues raised by Cody with respect to 3-rd party packages, etc.

Quality Management != Absolute Quality.

It's about risk management, as David's presentation of our work at UseR hopefully explained.

It's about common sense.

Combined, this can result in reasonable statements like (real names used in artificial examples to make a point):

        Martin M does X with R, I trust Martin, therefore I trust X done with R because the risk of wrong results in Y will have a low impact.

If Y happened to have an impact of $500m, then a reasonable approach might be to reconsider and find an additional expert, say Doug B, to confirm.

Alternatively, if you don't believe in expert opinions (or subjective probability, or mechanistic modeling), and feel like an empirical frequentist, you might just get a team of monkeys to verify that X in R seems correct, based on a project management strategy that incorporates someone's favorite IT risk mitigation approach.

With respect to Bert's points about 21CFR Part 11, please read the documents on the R WWW with respect to such things for a pretty informed opinion as to what is really happening.

I may not speak for Novartis, but it is possible that we'll be using a non-commercial version of R at some point in the future and we've been looking into the risk management strategies. Some people are annoyed at the packages we will not let people use, but code review suggests that we really don't want people to use them (risk management, again). The supporting infrastructure will be nice, but it'll also be a PITA to build.

But it really is just a matter of codification of common sense -- you should always put anything that you want to reproduce under version control, you should always have test cases to confirm that the implementations that you are using work in a few average cases (no one can cover every corner case) and you should make sure that you align your data and computer code with your reporting workflow. It's the implementation of common sense that seems to be hard, as most R-help readers should be aware of by now. I can't claim to implement it all the time, as readers of this list are probably additionally aware.

Best regards / Mit freundlichen Grüssen, Anthony (Tony) Rossini
Novartis Pharma AG
MODELING & SIMULATION
Group Head a.i., -- EU Statistical Modeling CHBS, WSJ-027.1.012
Novartis Pharma AG
Lichtstrasse 35
CH-4056 Basel
Switzerland
Phone: +41 61 324 4186
Fax: +41 61 324 3039
Cell: +41 79 367 4557 (to send an SMS from Lotus Notes put the following: 0041793674557_at_sms.novartis.com in the To box : -> only the content of the subject is sent)
Email : anthony.rossini_at_novartis.com

        [[alternative HTML version deleted]]



R-help_at_stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. Received on Tue 21 Aug 2007 - 10:46:00 GMT

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.2.0, at Tue 21 Aug 2007 - 12:34:09 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help. Please read the posting guide before posting to the list.