Re: Wing: Descriptive Epidemiology by Any Other Name...

[Bernard L Cohen <blc+@pitt.edu>]

On Thu, 27 Feb 1997, Ron L. Kathren wrote:
> 
> As for the difference between testing a LNT 'theory' and determining a dose
> response relationship, I'm really confused.   Is not the LNT theory simply a
> statement or representation or model of a dose-response relationship?  If
> so, then is not testing it simply a determination of its validity, and hence
> determining whether it is a (suitable) dose-response relationship?

	--I apologize for not explaining this difference clearly, although
I have done so in several of my papers. Let me try to explain it briefly:
	A dose-response relationship refers to risk to individuals, and
therefore requires data on individuals. This is not available from an
ecological study. What is often done is to assume that all individuals in
each ecological group have the average exposure of the group. This
assumption is what is called "the ecological fallacy". For example,
suppose there is a threshold for response at 10 pCi/L, and average radon
levels are: County A -1.0 pCi/L and County B -1.5 pCi/L. But these average
levels are irrelevant because they don't necessarily determine what
fractions of the populations are above the 10 pCi/L threshold.Thus, an
ecological study cannot determine a dose-response relationship, and
trying to use it for that purpose is "the ecological fallacy".
	Testing the LNT theory is a very different matter. LNT gives the
risk to individuals, and I show by rigorous mathematics that this leads to
a prediction for lung cancer rates vs average radon levels in counties.
There are no assumptions introduced, so the "ecological fallacy" does not
apply. I  then test this prediction against the data and find a tremendous
discrepancy. Incidently, the fact that LNT predicts that average dose
determines mortality rates is familiar to Health Physicists from the fact
that "person-rem" determines number of deaths --- person-rem divided by
population is average dose, and number of deaths divided by population is
mortality rate. This depends on LNT, but there is no ecological fallacy
involved.
	There are a lot of details to be worked out on treatment of
potential confounding factors, and I literally spent years of effort on
that aspect, trying to explain our discrepancy consistent with LNT. I even
published papers in 1990, 1992, and 1994 reporting on these efforts
but still holding out the hope that such an explanation is possible. I
tried to hire consultants to find such an explanation. It was only in my
1995 paper that I concluded that by far the most plausible explanation of
our discrepancy was gross failure of LNT theory. I still try to keep an
open mind and am offering large monetary awards for suggested
explanations, but none have been forthcoming. As part of this offer, I
promise to provide several plausible explanations for why any other
published ecological study can give very wrong answers. Hopefully this
shows that I understand the problems with ecological studies in general.
	Perhaps this discussion explains why I am greatly distressed when
people say "Cohen's study is an ecological study and it is well known that
ecological studies are no good". In my view, my study is completely
different than all other ecological studies and should be considered
separately from them.    

Bernard L. Cohen
Physics Dept.
University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245
Fax: (412)624-9163
e-mail: blc+@pitt.edu