[R] Use of R in clinical trials

Emmanuel Charpentier charpent at bacbuc.dyndns.org
Tue Feb 23 09:06:08 CET 2010

Le samedi 20 février 2010 à 09:44 -0800, Dieter Menne a écrit :
> If you check
> http://n4.nabble.com/R-help-f789696.html
> you will note that this thread has the largest number of read since years.
> Looks like an encouragement to Mark to keep the mentioned CRAN document
> updated.
> To add a more serious note to my sarcastic comments earlier:
> I don't think the FDA or our national agencies in Europe are to blame. I
> know that they have eminent statisticians there who know what they are
> talking of and are much more flexible than the culprits.
> The people to blame for the "Nobody got fired for using SAS" attitude are
> reviewers and bosses with mainly medical background who make decisions. It
> the difference in the use of the term "validated" which leads to confusion.
> A method is considered "validated" in medicine when it has been compared
> with a gold standard and is non-inferior within reasonable limits. For
> example, in the diagnosis of a gastric ulcer gastroscopy is the gold
> standard, and cheaper or less invasive tests are measured against that. 
> However, sometimes gold standards are the easy way out to avoid litigation,
> and statistical evidence against these is brushed aside. Think of year it
> needed to accept the Australian bush doctor's evidence that the bacteria
> Helicobactor pylori is a main determinant for gastric ulcers; everyone
> "knew" that ulcers were caused by stress alone.
> Or gastric emptying: the "gold standard" is (was?) the use of a radioactive
> marker that was recorded after a meal. Since radioactivity cannot rise out
> of nothing, it was a well known fact that stomach content always goes down
> after a meal. After people started measuring the real volume of the liquid
> in the stomach with MRI imaging, it came out that the content INCREASED
> after the meal due to strong gastric secretion. Despite visible evidence
> from the images, the increase was considered "not validated", because is was
> in contradiction of the gold standard.
> "Validated" in medicine means: Some well-known person has made a publication
> on the subject. He or she may be right, but not always.
> Mention the word three times in a discussion, and my blood pressure is at
> 200 mmHg

Furthermore, the "gold" standard might have to be replaced by some
platinium. The FDA itself seems open to some radical changes. See :


Tho only software "products" explicitely mentioned in this documents
are ... WinBUGS, BRugs and OpenBugs. Not exactly a common component of
your average "validated" toolchain...

By the way, a "validated" acceptance of Bayesian methods for regulatory
purposes (for devices, admittedly, but FDA"s arguments are *also*
applicable to drugs assessment) might well be a "window of opportunity"
for R, which offers :
        - interfaces to WinBUGS
        - interface to JAGS, an alternative implementation of the BUGS
        language and model
        - various packages for alternative MCMC and/or conjugate
        estimation of posterior distributions
        - a good system of production of formally validatable documents
        (Sweave|odfWeave, and possibly RRPM (Bioconductor)).
Looks like R might have a head start over SAS for this "new" race...

However, Bert Gunter's remarks are *still* valid : a lot of money has
already been invested in SAS "production systems", esp. in the
pharmaceutical industry. This industry will consider switching systems
if and only if this switch allows them to save *serious* money or to
*seriously* streamline their current process. We all know it's possible,
but the economic|financial case still has to be done. An implementation
of such a system in a new sector (namely medical devices), using the
"Bayesian headstart" and the FDA incentive to consider Bayesian methods,
might help making this case.

> My message: If you hear "not validated"  or "validated", question it. 

My message : timeo apothicarios et dona ferentes

					Emmanuel Charpentier

More information about the R-help mailing list