An article "GM corn health risks identified" appearing at Stock and Land 16 Dec 2009 features a new statistical study by French scientist de Vendomois and his colleagues. What wasn’t mentioned in Stock and Land is that de Vendomois uses a risky statistical approach called "data mining" to trawl through nearly 500 tests of GM corn safety in rats to find a small minority of test results that might have safety significance.
Data mining as used by this French group suffers a very real risk of giving false findings, as more fully documented in previous GMO statistics posts. [Note added — and by Australian food safety agency FSANZ see later Pundit post]
The Stock and Land article also quotes the opinion of anti-GM group Genethics’ that studies questioning GM safety are being ignored by the Australian food safety agency FSANZ . In fact, FSANZ and other food safety agencies frequently make detailed responses to such concerns, and contrary to Genethics assertions, the new French study illustrates that safety agencies have done substantial science while dealing with the very issues raised by the French investigators.
In a June 2007 report, the European food safety organisation EFSA made a detailed criticism of the same French group’s statistical methods, and convincingly demonstrate that de Vendomois had previously made false claims about GM safety using data dredging.
This 2007 EFSA report showed (in Appendix 5) that when there is correlation between results of different tests (which EFSA showed does occur in practice ), or when there are small real differences between groups being compared (as there are in feeding trials where animals are allowed to freely take as much food as they like, and random differences in rat appetite caused groups to differ in average weight) the numbers of positive results appearing by chance can be greater than expected from an ideal simplified model.
de Vendomois and others (2009) do not take these realistic complications into account, nor, rather oddly, do they even mention the telling 2007 EFSA critique of their approach, although it is obvious from the prominence of formal interviews of group member Gilles-Eric Seralini with the EFSA documented in it that they must know of its existence.
Genethics do not mention this crucial omission by de Vendomois.
As the latest French report does not properly correct the mistakes pointed out by EFSA, it is quite possible that it also has dredged out false results. We should all be careful about misusing or abusing statistics when it comes to understanding food safety. It’s not easy though :-).
See
Il y a 3 mesonges…
Joel Spiroux de Vendomois and other (2009). A comparison of the effects of three GM corn varieties on mammalian health. Int. J Biological Sciences 2009 5(7):706-726.
Post script.
Statistics, of course, is not the be all and end all of the issue. A statistically meaningful difference is not necessarily a toxic effect, nor is it an abnormality if it falls within the range of common variation seen in rat populations in closely similar circumstances.
Update 2. A link to a later Pundit Post on the dredging event
FSANZ make a statement— very similar to the Pundit’s view, but with nice links.
From the comments to that later Pundit posting:
The new french commitee, the HCB (Haut Conseil des Biotechnologies)published a statement concluding the "new" Séralini analysis is as misleading and wrong at the previous one.
http://www.blogger.com/The
Update 3. Monsanto have put their response on their website. It’s comprehensive.
