Response to suggestion by Mossman

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

On Thu, 30 Jan 1997, Ken Mossman wrote:

> Radsafers:
>
> It would appear that one test of the validity of Bernie's conclusions =
> would be to look at the relationship between cigarette consumption and =
> lung cancer mortality. According to the American Cancer Society, =
> cigarettes explain 80-90% of lung cancer deaths. There is over 40 years
=
> of epidemiological research (Hammond and others) clearly establishing a
=
> positive dose-response between smoking (as measured by cigarette packs =
> per day) and lung cancer mortality. I know Bernie has used cigarette =
> consumption data to correct for smoking as a confounding variable in his
=
> analysis, but I have not seen a separate analysis of  the relationship =
> between lung cancer deaths versus cigarette consumption.
>
> It would seem to me that if the resulting analysis is consistent with =
> other epidemiological studies (particularly with regard to the slope of
=
> the dose-response curve), Bernie's conclusions about a negative =
> correlation between radon and lung cancer deaths would be bolstered (not
=
> necessarily proven). However, if the analysis of cigarette consumtion =
> versus lung cancer mortality using Bernie's data are inconsistent with =
> previous findings, that would invalidate Bernie's conclusions.
>
> Ken Mossman (ken.mossman@asu.edu)=20
> Kenneth L. Mossman
> Professor of Health Physics
> Arizona State University
> Tempe, Arizona 85287-2701
> Voice: 602-965-0584
> FAX: 602-991-4998
> E-mail: ken.mossman@asu.edu=20
>

        My analysis of radon vs smoking is published in "Statistics in
Medicine", Vol 14, p.327-328; 1995. It does show the expected correlation
between smoking prevalence and lung cancer in statistically indisputable
analyses. This is the analysis Ken is requesting.
         On the other hand, this is an ecological study and I have never
claimed that an ecological study can determine a dose-response
relationship. I only use it to test the linear-no threshold theory, for
which application the "ecological fallacy" does not apply.
        The reason I published the analysis referenced above was to point
out a horrible error in a paper by Sander Greenland in which he claimed
that the ecological study did not show the smoking vs. lung cancer
relationship. He used a formula in which he used percent as a fraction --
e.g. he used 0.5 instead of 50, a factor of 100 error.
    The above referenced paper dealt only with
data on states, so I add here the data on counties that I used in most of
my analyses. With m=age-adjusted mortality rate, and S=%of adult
population that smokes cigarettes, linear regression of m on S gives

                m=1.22(+/-2.5) + 1.12(+/-0.049) S      for males

                m=1.83(+/-0.66) + 0.327(+/-0.021) S    for females

The mean values and standard deviations for the distribution of 1601
counties are
        males: m=59, SD=15.5;  S=52, SD=6.9

        female: m=12, SD=4.6;  S=31, SD=5.2

These results are clearly consistent with smoking being responsible for
nearly all lung cancer in males, and for the great majority of lung
cancers in females. Note that these rates are from 1970-1979,
before female lung cancer rates from smoking reached their present level
of importance and when smoking prevalence was much higher than at present.
        I should have said that m is in units of 10E-5/year.


Bernard L. Cohen
Physics Dept.
University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245
Fax: (412)624-9163