Re: Ecologic LNT study - reply from Dr. Lubin

[Jim Muckerheide <jmuckerheide@delphi.com>]

FIELDRW@aol.com (forwarded what Jay Lubin) wrote:

> This exchange has been very useful.  The fundamental disagreement that I have
> with Dr. Cohen is simply whether one can "scale up" a model for individuals
> and
> attain a meaningful model at the county level.  The answer is no; one cannot
> "scale up" without accurate within-county information on risk factors (and the
> county-to-county variation) - information that ecological studies do not have.
> Let's state the issue in the reverse.  Suppose several counties have equal
> county-level radon "means", and equal county-level data for all other risk
> factors (the same mean ages, proportion of smokers, SES, etc), i.e., at the
> county level the covariate information is identical across counties.  Is there
> anything that can be inferred about the lung cancer rates in the counties?
> The answer is no; one cannot draw any logical inference about the lung cancer
> rates in the counties.  While the risk factor information is the same at the 
> county level (and thus no additional adjustments are ever possible using 
> county information), differences in lung cancer rates at the county level may 
> occur due to differences in the correlational relationhips among the risk 
> factors within the counties.
> 
> Population dose can be a valid indicator of population risk, and "scaling up"
> can be valid only when the risk relationship between disease outcome and all
> risk factors is linear in all risk factors.  However, there is overwhelming
> evidence that the functional relationship between radon, other risk factors
> and lung cancer is most decidedly not linear - it is linear in radon exposure, 
> but proportional to the background rate.  This non-linear relationship for 
> risk has been most clearly demonstrated in the greater than additive 
> associations between
> radon and numerous factors and lung cancer, such as radon and tobacco use (and
> very likely exposure to environmental tobacco smoke, but this has not really
> be explored), radon and attained age, radon and gender (only suggestive data),
> radon and previous lung diseases, and radon and other mining exposures (in
> populations occupationally exposed to radon).  Analyses have consistently
> demonstrated a linear relationship between lung cancer and the excess relative
> risk.  That functional form implies that conditional on other risk factors,
> lung
> cancer risk increases proportionally to the background rate with increasing
> radon progeny exposure.  While the relative risk may be linear, the absolute
> excess risk with increasing exposure is dependent on the myriad of other risk
> factors. This is the reason that even though counties may have equal (or
> greater or lesser) radon concentrations, county lung cancer rates cannot be 
> "scaled up" with any validity and be compared without information (at the 
> level of individual) on other lung cancer risk factors.
> 
> My problem is then with ecological studies.  I do accept that it is a
> perfectly valid scientific question to ask whether there is an excess risk 
> of lung cancer at the lowest levels of (residential) radon progeny exposure 
> - but for individuals.  In my view, the data are consistent with some excess 
> risk at the lowest exposures, but I will allow that this is an arguable 
> proposition. However, ecological analyses shed no light on this issue, since 
> one cannot "scale up" with validity.
> 
> Jay Lubin
> 
> Jay Lubin, PhD
> National Cancer Institute
> Biostatistics Branch, EPS/8042
> 6120 Executive Blvd
> Bethesda, MD 20892-7244
> Tel: 301-496-3357
> Fax: 301-402-0081
> Email: lubinj@exchange.nih.gov <mailto:lubinj@exchange.nih.gov>
> -----------------------------------

Surely no one with any knowledge of statistics can take this seriously. :-( 
This is simple dissembling to "justify" the LNT for the rad protection/funding
agencies.

There is some basis for these _general_ concerns in Cohen's first study of
Cumberland County PA, See, e.g.,:
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/1/Rev%202%2012631co87.html

And even a basis in the following study comparing the "10 large counties with
highest lung cancer" vs. the "10 large counties with lowest lung cancer." The
statistical strength even here makes the likelihood that the LNT for radon is
possible as the result of the influence of other risk factors minimal. But
arguable if you want to be really conservative! See:
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/1/Rev%202%2012631co89.html

Concerns about ecological studies are valid, but such 'concerns' still only
result in a small to moderate likelihood that "a study" _MAY_ not be
representative.

But the LNT premise is completely invalidated by the numbers; the statistics.
_IF_ there were other risk influences at work, there would be (have to be)
significant variation in the multiple studies. The statistics establish that
even small but significant competing or compounding risk effects would cause
significant variations if they existed. The results would not be, could not
be, strong and consistent like Cohen's results. A few notes:

o The data set reflects virtually the entire U.S. population for ALL lung
cancer deaths, with >300,000 radon data points, not some limited group that
"may not be representative."

o All of _100s_ of independent studies, with separate data sets, consistently
produce the SAME result. The study with ALL the results is _NOT_ just some
regression from widely scattered data. Such an unambiguous relationship in a
LARGE data set is the direct result of sound data - data that is both
extensive, and with a definitive causal relationship. Again, this is _NOT_
some small, ambiguous, "representative" data set.

o Graham Colditz at Harvard, a _real_ epidemiologist, one of the world's
premier epi's who works with real people to address _real_ health issues for
human benefits (not committed to the EPA and radiation research funding),
_fully_ validated Cohen's results when the data was already quite extensive,
but even then not complete. Later data has only reinforced the consistency and
certainty of the results. He found the inverse relationship to be
epidemiologically valid (even with some differences with Cohen on treating
some analyses).  [See: Environ Res 1994 Jan;64(1):65-89 "Tests of the
linear-no threshold theory for lung cancer induced by exposure to radon."
Cohen BL, Colditz GA  ABSTRACT: "The linear theory used to extrapolate the
cancer risk of radon exposure from high levels where direct data are available
to low levels encountered in homes is tested by comparing lung cancer rates,
m, and average radon levels, r, in numerous U.S. states and counties. It is
shown that most problems normally associated with ecological studies do not
apply here. The data show a very strong tendency for lung cancer rates,
corrected for smoking prevalence (S), to decrease with increasing r, in sharp
contrast to the opposite behavior predicted by the theory. It is shown that
even a perfect negative r-S correlation cannot explain this discrepancy.
Actual r-S correlations are only a few percent. Several other possible
explanations for the discrepancy are explored, but none can reduce it by more
than about 25%."]

(We worked with an independent senior epidemiologist, during a sabbatical
year, who looked at BEIR VI, Cohen's later studies, and the 1994 Cohen-Colditz
study, plus some of the confirmatory studies: Bogen, Jagger, etc. She said
Colditz was right on the epi science. She decided not to publish after she was
'warned off' by fellow epi's that she had had review her preliminary analysis
to because "you don't want to get entangled with that career damaging
subject." But, we expect such sources will come to the fore when convictions
on allegations about the such sources are to be found.)

o There is a totally false premise that there is 'no explanation' for Cohen's
'discrepancy.' There is actually no discrepancy, no "paradox." The science
literature agrees with Cohen.  Contrary to BEIR VI and implications here that
dishonestly treat Cohen as 'alone,' in fact Cohen's results are confirmed by
all other strong studies. Only some small 'junk science' case-control studies
(some of which are shown to be manipulated) have different results. And by
definition, small- to moderate-sized case-control studies can not possibly be
accurate under these conditions. See again, Cohen's 1987 paper that reviews
the literature above!

NOTE: By definition, case-control studies REQUIRE precise knowledge of actual
individual doses to produce good results. This limits the variation in the
statistics of small numbers. Knowing the actual dose eliminates the
statistical uncertainty in this primary parameter. This is what makes
case-control studies "better." This works in medical studies, etc. where doses
are controlled. 

This is _NOT_ true with radon case-control studies. That's why radon
case-control studies are all over the map (and some seem to have been dropped
when the results were not what the rad protectionists wanted). It's also why
the case-control study in Shenyang China, where the traditional life-style of
women in small residences in China make the "dose data" much better than other
radon case-control studies, is consistent with Cohen's results. See, e.g.,
Kondo's summary of the Blot (1990) study:
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/3/126331ko93.html

In fact, go further: With wide variation (high vs. low) radon, people in "high
dose" residences are LIKELY to have a much wider variation in individual
doses, even more between residents of 2 houses that have the same high-dose
measurement. 

That's not true of "low dose" residences (unless one works in a high-dose
location, which is not so likely in a low-dose area, with a low-dose work
location likely for a high-dose resident). Therefore, the high dose group have
some people with low dose, and therefore at higher lung cancer risk, but the
low dose group do not have an equivalent distribution of people at high dose
to have a low lung cancer risk. Therefore radon case control studies have a
structural bias against showing the inverse relationship.

Of course, there are really no good case-control studies that include the
truly low dose regions with higher lung cancer rates that are most significant
in Cohen's data.

However, in addition to other studies identified by Cohen in 1987, see other
studies at:
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/1263list.html
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/Rev%202%201263list.html
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/1/12631list.html
http://cnts.wpi.edu/RSH/Data_Docs/1-2/6/3/1/Rev%202%2012631list.html

Consider that the reason that Cohen's discrepancy can not be "explained" is
that radon does not cause lung cancer, but indeed environmental levels reduce
lung cancer, even among those who smoke. (We'll leave the fact that even the
miner studies are misrepresented to identify radon as a cause lung cancer.)

You can also consider animal studies that contradict the LNT for radon:
http://cnts.wpi.edu/RSH/Data_Docs/1-3/1/13114du97.html

We will get more of the data entered as we can; any help appreciated :-)

Regards, Jim
Radiation, Science, and Health
muckerheide@mediaone.net
==============================


> Bill Field
> mailto:bill-field@uiowa.edu
> ************************************************************************
> The RADSAFE Frequently Asked Questions list, archives and subscription
> information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html
************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html