[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: lung cancers, primary vs secondary
This is not enough reason for me to go to the substantial effort
of entering the SEER data into my data bases, unless there is reason to
believe that analyses with them would be publishable. Note that SEER data
are not available for most of my counties. One of the strengths of my
analyses is that they include data for most U.S. counties, including those
with 90% of the U,S, population. Data for a single state are much more
susceptible to statistical variations and confounding factors that are
averaged out in a national study.
Bernard L. Cohen
Physics Dept.
University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245
Fax: (412)624-9163
e-mail: blc@pitt.edu
web site: http://www.phyast.pitt.edu/~blc
On Wed, 25 Jun 2003 epirad@mchsi.com wrote:
> Dr. Cohen, you ask "Why is this not a problem for cancer incidence which you
> say is more reliable and for which autopsies are never possible?"
>
> SEER incidence data are based not only on death certificates but often of other
> sources of more reliable information. Because of the generally short survival
> of lung cancer, autopsy information is fequently availble for incidence data.
>
> Dr. Cohen, you asked, "What do you mean by surrogate? Why don't you specify
> that the issue is incidence vs mortality -- both data are from NCI? What do you
> mean by poor agreement? What is the evidence that this is correlated with
> radon levels on a national scale?"
>
> The issue is not just incidence and mortality, but rather the quality of the
> state cancer registry that provides the data. I previously told you that the
> SEER Program is the only comprehensive source of population-based information
> in the United States that includes stage of cancer at the time of diagnosis and
> survival rates within each stage.
>
> See also - http://www.seer.cancer.gov/
>
> You ask what do I mean by poor agreement of your mortality data -
>
> The rank order of county lung cancer rates varies substantially over time
> periods. This can be seen in Riggan’s county mortality data, used by Cohen,
> which exhibit little correlation across time. For example, the 1960-1969
> county lung cancer mortality data for Iowa (Riggan et al. 1983) are in
> surprisingly poor agreement with the corresponding rates from 1970-1979. The
> rank correlations are 0.398 for males and 0.163 for females. In fact, at the
> 5% level of significance, the correlation for females is not different from
> zero (p = 0.106). Cohen (1995) only considers models in which the covariate
> effects on risk are linear. Under this assumption of linearity, lung cancer
> rates will fail to correlate across time periods only if there is at least one
> risk factor for the disease which does not correlate across time. The small
> correlations observed for his lung cancer rates indicates that this is indeed
> the case. Cohen uses his risk factor information, collected after the lung
> cancer deaths, as if it was representative of earlier time periods. Since this
> is not the case, he will not be able to control for their true effects on lung
> cancer risk.
>
> Further -
>
> LUNG CANCER INCIDENCE RATES IN IOWA
>
> All lung cancer cases diagnosed in Iowa are reported to the SEER State Health
> Registry of Iowa (SHRI) (Karnell et al. 1995). Gender-specific lung cancer
> incident rates from 1988-1992 for individuals aged 40-84 were obtained from the
> SHRI. The rates are age-adjusted by the total 1970 U.S. population. Since
> lung cancer has a low survival rate, one would expect there to be little
> difference between incidence and mortality rates. Nevertheless, in assessing
> the impact of risk exposures, it is more appropriate to use incidence data.
> The age range 40-84 was selected from the Iowa data because these individuals
> had a greater cumulative radon exposure in the time period following Cohen’s
> radon measurements, and because lung cancer is rare in individuals less than 40
> years of age.
> When we replicate the regression of Riggan’s lung cancer mortality data
> on Cohen’s radon measurements, we get the same large negative associations as
> reported by Cohen (1995). However, this approach is flawed because it is based
> on temporally incorrect mortality data observed before the time of exposure.
> The results are quite different when we regress the more temporally correct
> Iowa SHRI lung cancer incidence rates on radon and smoking percentages. The
> analysis produces spurious results – positive slopes for one group, negative
> slopes for the other. In contrast, Cohen found negative slopes for both Iowa
> males and females. Iowa's negative slopes were typical of the other 33 states
> in his study. Among males and females, respectively, the slopes for Iowa
> ranked 18th and 15th overall. We used incidence rates for Iowa since these
> data were readily available to us through the Iowa SEER Cancer Registry.
> However, the impact of using their data serves to underscore the importance of
> using current incidence data, rather than retrospective mortality data when
> studying disease risk factors.
>
>
> > >
> > > More to the point, we have conclusively shown that the mortality data used by
> > > Dr. Cohen is temporally incorrect in relation to the latency period for
> > cancers
> > > and radon testing periods in Dr. Cohen's data.
> >
> > --I have responded to this before, quite recently. If anyone wants
> > a repeat, please ask.
> >
> > Actual incidence lung cancer
> > > data from a National Cancer Institute Cancer Registry is in very poor
> > agreement
> > > with Dr. Cohen's surrogate data for the time period of interest.
> >
> >
> >
> >
> > ************************************************************************
> > You are currently subscribed to the Radsafe mailing list. To unsubscribe,
> > send an e-mail to Majordomo@list.vanderbilt.edu Put the text "unsubscribe
> > radsafe" (no quote marks) in the body of the e-mail, with no subject line.
> > You can view the Radsafe archives at http://www.vanderbilt.edu/radsafe/
> >
>
************************************************************************
You are currently subscribed to the Radsafe mailing list. To unsubscribe,
send an e-mail to Majordomo@list.vanderbilt.edu Put the text "unsubscribe
radsafe" (no quote marks) in the body of the e-mail, with no subject line.
You can view the Radsafe archives at http://www.vanderbilt.edu/radsafe/