[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Iowa radon vs lung cancer study - response to Cohen
On Mon, 26 Jun 2000 EPIRAD@aol.com wrote:
>
> Radsafers,
>
> With hesitation, I am writing a limited response to Dr. Cohen's recent
> posting concerning the Iowa Radon Lung Cancer Study. I have been answering
> Dr. Cohen's questions about the Iowa Radon Lung Cancer Study via direct
> email. I would have preferred to continue the direct email approach to
> scientific dialogue. Obviously, I do not feel the listserv is a scientific
> forum and scientific correspondences are better left to letters-to-the-editor
> or direct correspondence. I therefore urge Dr. Cohen to either write a
> letter-to-the-editor of the AJE if he thinks his comments have any scientific
> merit or continue his direct dialogue with me.
--I felt that our private correspondence had gone about as far as
was useful for the present. Since the Iowa Study had been extensively
discussed on RADSAFE, I thought it appropriate that I state my views on
the subject.
I might note that Dr. Field has several times used RADSAFE to
publicly challenge my work.
> In the end, Dr. Cohen and I will likely have to
> agree to disagree on many of these points as we have so many times before.
--This is one issue we do agree on
> ----------------------------------------
> Dr. Cohen states, "The principal reason why the Iowa study is claimed to be
> better than previous studies is that they did a more elaborate job of
> estimating radon exposure doses, by considering exposures outside the home
> and time spent in various places inside and outside the home. I view these
> improvements as relatively unimportant because of the other much larger
> uncertainties in radon exposures".
>
>
> Field: The importance of accounting for mobility is obvious by examining the
> impact on risk estimates if the mobility is not accounted for (see Figure in
> AJE paper). Health Physicists know the importance of both concentration and
> time in exposure assessment. If you do not account for where the time is
> spent, you are left with assessing risk based solely on concentration as
> previous residential radon studies have done. Dr. Cohen may view these
> improvements as relatively unimportant, but he has no data to support such a
> claim.
--I did not mean to imply that mobility cannot be important, but I
went on to list many other things that can be more important that are
still not taken into account in their assessing exposures, especially
since there is so little difference in average mobility patterns for cases
and controls.
> The other reason the Iowa Study obtained better exposure estimates is
> completely ignored by Cohen. Previous residential studies that did not have
> the minimum residency requirement of 20 years had to impute a sizable portion
> of missing data for the previous 20 years prior to lung cancer or prior to
> enrollment. This decreased the power of previous residential radon studies.
> The importance of the minimum 20-year residency is detailed in the AJE paper.
--I certainly agree that the 20 year residency requirement is an
important feature. As I remember, there were similar requirements in other
studies; where there were not, that would be a serious deficiency in those
studies. But the 20 year requirement still falls far short of determining
lifetime exposures.
> --------------
> Dr. Cohen states, "For one example, they consider only exposures 5-19 years
> prior to cancer diagnosis, neglecting earlier exposures; among the Japanese
> A-bomb survivors, the majority of lung cancer deaths occurred more than 20
> years after exposure, and according to the BEIR-VI models, the same should be
> true of lung cancers induced by radon."
>
> Field: I would suggest that Dr. Cohen review BEIR IV to look at the temporal
> weighting factors derived from miners.
--Using the BEIR-VI model, only 37% of the lung cancers induced by
radon in homes and elsewhere is due to exposures between 5-19 years before
death. That means that 63% of the relevant radon exposure is not accounted
for. I don't have the BEIR-IV model in hand, but I assume it is superseded
by BEIR-VI on this matter.
> --------------
> Dr. Cohen states: "Moreover, the procedures they use to take into
> consideration the time spent in various places introduces a whole new set of
> uncertainties derived from problems in remembering and reporting habits from
> 5-19 years earlier."
>
> Field: The retrospective dose assessment methodology was very detailed. The
> assessment methodology was reviewed by at least two reviewers for the
> American Journal of Epidemiology and several reviewers from the Journal, Risk
> Analysis. The methodology describing the retrospective mobility assessment
> is available for review in the journal. The stable time periods in the
> subjects' life were anchored by life events. For example, women know when
> they were married, when they were working, when they had children, etc.
> These life events were used to frame the time periods. Please see the paper
> describing this methodology for more details. Nonetheless, greater
> uncertainty in retrospective mobility is attributable by assuming a constant
> home occupancy such as 75%.
--I did not mean to suggest that you did not do a competent job on
this, but I still think that there are large uncertainties in such
procedures.
> --------------
> Dr. Cohen states, "There are also uncertainties in that radon levels in a
> particular house may have changed over time. For example, a crack may have
> developed (or conversely, have become sealed) in the basement floor, causing
> the radon level to be different 15 years ago than it was in the recent
> measurement. Similar problems can arise from
> changes in ventilation; do window opening habits and use of air
> conditioning remain unchanged over 20 years, let alone between hot and
> cool Summers?"
>
> Field: This is more of a problem for studies that go back and assess the
> radon concentrations in previous homes. Obviously, the new owners may have
> different behaviors as far as opening windows etc. than did the subject.
> Also, when someone new moves into a home, they are more apt to make
> structural changes than people who have lived in their homes for over 20
> years. Also, remember - the most important predictive factor for radon in
> the home is source strength - source strength remains stable over time.
>
> We excluded cases and controls that mitigated their home for radon. We also
> looked at changes that may have affected radon concentrations over time. The
> major changes that occurred that may have affected the radon concentrations
> was the switch over in the 1960s and early 70's to central and window air
> conditioning. The window opening behavior of the subjects was stable over
> the years. We did note a higher degree of difference in repeated year
> measurements for the homes after the subject had passed away. The COV for
> repeated second year measurements is reported in the AJE paper.
>
--Again I do not dispute the fact that you did a competent job on
trying to assess the retrospective radon exposures, but there are many
things that could have affected radon levels that would not be recognized.
>
> Dr. Cohen States: "The aspect of this study that bothers me most is that the
> corrections for time spent in various places makes a very large difference
> in the results, greatly enhancing their conclusion that radon exposure is
> important, in spite of the fact that the average percentage of time spent
> in various places is essentially the same for cases and controls: for
> cases vs controls, average percent of time spent in the home is 73.2 vs
> 72.1, in another building is 14.2 vs 14.4, outside is 7.6 vs 8.5, and away
> is 5.0 vs 5.0. While I can understand that mathematically it is possible
> that correcting for where each individual spends her time can make a big
> difference from considering onle the averages, it seems very
> difficult to see how the effect can be so systematic as to grossly affect
> the results. It seems to me that the authors owe us some explanation to
> help us understand such a seemingly improbable situation."
>
> Field: As I have pointed out before in my discussions with Dr. Cohen, you can
> not always look at summary data findings and be able to explain the details
> of the individual data linkages between radon concentrations and
> retrospective mobility.
--I can understand this as a mathematical possibility, but it is
so very curious that I believe an explanation is essential
We were limited in some discussions in the AJE
> paper since we had already slightly exceeded our space limit in the AJE
> paper. We have almost completed a paper that examines the a priori risk
> assessment model used by the Iowa Study versus alternative scenarios that did
> not include consideration of where the time was spent by the subject. The
> paper in the works contains a more detailed discussion than what is included
> in the AJE paper.
--I look forward to this paper in the hope that it will provide
some understanding of the very curious situation. Please send me a copy as
soon as it is available.
>
> Dr. Cohen states. "Another problem that bothers me is that there is no
> consideration given to geography, which is a potential confounding factor
> very closely correlated with radon levels. Controls are chosen randomly from
> the entire state of Iowa, where average radon levels in different counties
> varies by a factor of 3, from 1.7 to 5.0 pCi/L. Cases come largely from
> areas where there is a high prevalence of smoking, which might easily have
> different average radon levels than the controls. It seems important to me
> that information be given at least on what counties the cases and controls
> come from."
>
> Field: Dr. Cohen, I understand your focus on average county radon
> concentrations given your previous research. However, the range of radon
> concentrations within a county is much greater than the range between
> counties. We did not use mean county radon concentrations to assess radon
> exposure, we used individual data for each subject. In fact, in a
> preliminary analysis we have found that mean county radon concentrations are
> a poor predictor of both individual radon exposures and home radon
> concentrations in Iowa.
--You seem to miss my point here. My point deals with the
selection of controls. If controls are selected from counties with lower
average radon levels than the cases, we can expect that the average
radon levels for your individual cases will be higher than for your
individual controls, which would lead to the unwarranted conclusion that
radon exposure causes lung cancer
>
> Dr. Cohen States: "The cases are nearly all smokers while the controls are
> nearly all non-smokers. There are many differences in life styles of smokers
> and non-smokers that can affect radon exposures. Some attention should be
> given to these. Also there is a substantial difference in educational
> attainment between cases and controls which could be a source of trouble.
>
> This is one example of socioeconomic variations which can be important both
> in disease vulnerability and in radon exposures, but essentially no attention
> is given to socioeconomic factors other than education."
>
> Field: The inclusion criteria of 20 years in the current home provided a
> match that made the educational status and socioeconomic status between the
> cases and controls much closer than previous residential epidemiologic
> studies. In addition, the analyses were adjusted for education. I would
> like to know what data Dr. Cohen is using to demonstrate these other
> unidentified socioeconomic factors were not adjusted for as part of the
> adjustments we made for smoking, education and age.
--As many epidemiologists have often told me, socioeconomic
factors can correlate with anything, and they are known to correlate with
lung cancer and with radon. I assume you matched cases and controls for
age. Your adjustments for smoking and for education are not evident in
their effect on cases/controls in Table 4, which are the data that impress
me most.
> Dr. Cohen States: "The basis for a case-control study is that cases and
> controls are matched in every way except for having contracted the disease.
> Any deviation from this would seem to be an issue for concern."
>
> Field: This is an erroneous statement. What Dr. Cohen describes here is
> overmatching.
>
> Dr. Cohen says in a case-control study the cases and controls are matched "in
> every way except for having contracted the disease". Dr. Cohen, I
> respectfully request a reference for this statement. For example, if your
> statement was true, we should have matched the cases and controls by radon
> concentration, mobility, shoe size, eye color, favorite ice cream flavor,
> etc.
--I should have said "in radon exposure and having contracted the
disease", and I should have said "in every way that could be relevant".
I don't see why a reference is needed for this statement -- it is simple
logic. Of course there are practical limitations to how well matching can
be done, but ignoring important confounding factors would be an important
weakness in any study. Ignoring potential confounding factors comes down
to judgements of plausibility, and to me, the plausibility of
socioeconomic differences between cases and controls causing problems is
very high. On the other hand, the plausibility of eye color, shoe size,
etc causing problems is very low.
Bernard L. Cohen
Physics Dept.
University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245
Fax: (412)624-9163
e-mail: blc+@pitt.edu
************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html