RE: Iowa Radon Lung Cancer Study

["Brooks, Michael" <mdb7@cdc.gov>]

I finally had a chance to read the full  "Iowa Radon Lung Cancer Study" in
the June issue of the American Journal of Epidemiology from a statistical
and health physicist's perspective. In my opinion this is the best designed
case control study for residential radon that I have read to date. The study
design and methodology are probably the best one could hope for, measuring
long-term radon levels in living spaces. The State of Iowa has the highest
mean residential radon level and looking at homemakers with a minimum of 20
years in the same home have the potential to find any link between
residential radon and lung cancer risk. My concerns have nothing to do with
the study design, but with the data analysis and interpretation. The study
claims to demonstrate a STRONG statistical association between residential
radon and lung cancer, based on a corrected odds ratio of 2.14 in the
highest exposure group and excluding all cases that died during the study.
My problems with the study are as follows (in no particular order):

1. The total difference in lung cancer cases can be accounted for by natural
variation among the cases (n=413). The natural variation in the number of
cases is 20.3, while the 33% of cases exposed above 4 pCi/L and 28% of
controls corresponds to 5% of 413 cases or 20.6. 

2. The study controls have an 11% higher rate of post-secondary education
than for the cases. Highest educational level has been strongly correlated
to greater longevity and overall health. It does not appear that the odds
ratios were corrected for educational level.

3. Due to the etiology of lung cancer the mean life expectancy after
diagnosis is around 5 years, therefore it is unreasonable to exclude cases
that died during the 5 year study, but it may be reasonable to exclude only
those cases whose family's disposed of the radon measuring devices before a
radon measurement could be made.

4. If statistical significance can only be achieved by omitting cases that
died during the study period, this might "imply" a protective effect from
radon exposure.

5. There may be a synergistic effect between smoking and radon exposure for
developing lung cancer that is not accounted for in the analysis. Many of
the uranium miner studies did not clearly identify smoking status of those
with lung cancer. The uranium miner studies appear only to show a
relationship between radon exposure and cancer among the smokers and miners
of unknown smoking status.

6. The cases had an ever-smoked rate of 86% vs. a rate of 32% ever smoked
among the controls. The smoking correction is not defined and the much
higher rate of smoking among the cases is going to make the corrected odds
ratio extremely sensitive to the smoking correction.

7. The intervals of cumulative radon exposure are made at strange,
non-integer values and are not evenly spaced. No cases or controls were
exposed to ZERO pCi/L of radon. There was a threshold of exposures. What was
that value?

8. When confidence intervals are graphed for the odds ratios vs. exposure
categories, there is no clear dose response. A line requires at least two
significant points to test for linearity and the origin does not count.

Overall, this study does not appear to demonstrate any statistically
significant association or dose response between residential radon and lung
cancer. The claim of a strong association is not shown in this study. My
opinion only.

Michael D. Brooks, CHP
CDC/ATSDR
Health Physicist
Division of Health Assessment and Consultation
Federal Facilities Assessment Branch
1600 Clifton Rd (E-56)
Atlanta, Georgia 30333
(404) 639-6053, fax 6075
Email: mdb7@cdc.gov

************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html