[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: New Steve Wing Study



At 10:17 AM 2/4/98 -0600, Gibbs, S Julian wrote:
>I think we are being overly critical of epidemiological 
>studies.  Some have been quite helpful.  Remember that the 
>early reports linking smoking and lung cancer were 
>epidemiological.
>We must also remember that these case-control studies 
>merely show an association.  They do not demonstrate cause 
>and effect. 
>
>
I agree! Some epidemiological studies are well done, but all they show is a
correlation.  You need something like a mechanism, something in the
etiology of the disease which is at least plausible and can support a claim
for a causal connection.  

Sadly, most epidemiological studies are not well done.  Even so, if you
find a relative risk of 4, say, it would take a pretty substantial bias to
show that kind of a correlation when there is none.  The problem arises
with studies that show a relative risk of, say, 1.15 with a confidence
level of 95%.  What most epidemiologists do not discuss or even realize, is
that they have to show that their procedures are good enough, so that this
excess risk of 15% cannot arise from systematic errors; in other words,
that they have all confounding factors controlled to better than 15%.  And
that is a tall order! Too tall, in my opinion.

And it is right here, in the region of low excess risks, that most of the
"crimes" against honest science are committed.  This is also the region of
the relative risk in which the perpetrators have a good chance of getting
away with it.  Most of the dubious claims are based on relative risks below
2, and that means an excess relative risk smaller than 1 (i.e., below the
doubling dose).  An author would have to be very convincing indeed to have
me accept a relative risk of 1.5 as real.  

In addition, most associations found are of the Coney Island Fallacy type:
On a sunny summer weekend, more hamburgers are sold than on a usual
weekend, also more people drown than on a usual weekend.  A clear-cut
association which proves that hamburgers cause drowning!



*************************

Fritz A. Seiler
Principal
Sigma Five Associates
P.O. Box 14006 
Albuquerque, NM 87191,USA
Tel:	xx-505-323-7848
Fax:    xx-505-293-3911

e-mail: faseiler@nmia.com

*************************