[R] R Software Risk Analysis
he|mut@@chuetz @end|ng |rom beb@c@@t
Fri Jun 19 17:31:29 CEST 2020
any (!) software used in regulated environments has to be validated.
Regrettably is is a misconception by many working in the pharmaceutical
industry that only studies evaluated by SAS are accepted by the FDA.
See this one-pager https://www.fda.gov/media/109552/download and a
presentation by the FDA's Paul Schuette at the useR! 2016
Note that the FDA itself uses R in modeling & simulation.
Contrary to proprietary (off-the-shelf, commercial, you name it)
software – where only a black box validation (aka, rubbish in, rubbish
out) is possible – open source SW allows – in principle – a white box
validation is possible.
Relying on proprietary SW is not necessarily a good idea ("We payed a
lot, hence, it will work.") Stephen Senn once told me (given, a good
while ago) that after an update of SAS, the Welch-test for unequal group
sizes / variances collapsed into the simple t-test. He called up SAS and
got the coder on the line. He inspected the source and after a couple of
minutes replied "Hey, you are right. We screwed up." It took SAS half a
year to roll out a corrective patch. What about clinicial studies
evaluated in the meantime?
We published a couple of papers in a specific field comparing software
doi:10.1208/s12248-020-0427-6). It turned out that one of the commercial
[sic] SWs tested was seriously flawed. Consequences: Dozens of approved
drugs taken off the market. The glitch in the software was _partly_
corrected in 2014 and the vendor stopped marketing it in 2019.
Just by two cents
Ing. Helmut Schütz
BEBAC – Consultancy Services for
Bioequivalence and Bioavailability Studies
1070 Vienna, Austria
T +43 1 2311746
M +43 699 10792458
E helmut.schuetz using bebac.at
More information about the R-help