American Journal of Epidemiology Vol. 152, No. 2 : 194-195
Copyright © 2000 by The Johns Hopkins University School of Hygiene and Public Health
LETTERS TO THE EDITOR |
RE: "PARALLEL ANALYSES OF INDIVIDUAL AND ECOLOGIC DATA ON RESIDENTIAL RADON, COFACTORS, AND LUNG CANCER IN SWEDEN"
University of Pittsburgh Pittsburgh, PA 15260
 |
INTRODUCTION |
|---|
 TOP INTRODUCTION REFERENCESINTRODUCTION REFERENCES |
|---|
The recent paper by Lagarde and Pershagen (1
) reports that the slope of the individual data on risk versus dose is positive, whereas the slope of these data aggregated by counties to give an ecologic analysis is negative. They (and others) claim that this finding invalidates the conclusions of my findings (2) from an ecologic study of lung cancer rates versus average radon levels for US counties. My study, designed only as a test of the linear-no threshold theory (LNT), reported a strong negative correlation, leading to the conclusion that LNT fails very badly in the low-dose region where it has never been tested. On the basis of the study by Lagarde and Pershagen (1), I have been asked publicly to concede that my conclusion is invalid. The purpose of this letter is to respond to these claims and requests. The most important problem with the study by Lagarde and Pershagen is the effect of confounding factors. In fact, the study by Lagarde and Pershagen even reports that one confounding factor, geographic latitude, explains the difference between the results of their analyses of individual and ecologic data. However, aside from that, it is obvious that a confounding factor that, for unrelated reasons, is strongly correlated with both radon levels and lung cancer risks can drastically alter the slopes being studied. I give many examples of this in my papers; for instance, if smoking prevalence in US counties had the maximum not-implausible width and a correlation coefficient with radon levels of -0.90, our discrepancy with LNT would be explained. However, in all cases, I was able to show that the required assumptions about correlations of confounding factors with radon levels are completely implausible.
I have searched at great length for potential confounding factors that might explain our discrepancy with LNT, investigating over 500 in all, including socioeconomic factors, environmental factors, geography (which includes latitude, a very important confounding factor in the study by Lagarde and Pershagen), etc., but nothing I have found has an important influence on our results.
Another problem with the study by Lagarde and Pershagen is its statistical weakness. The slope of lines fit to their data (excess risk per 100 Bq/m3 of radon) have 95 percent confidence intervals ranging between -0.21 and +0.15) and -0.21 and +0.21 from their ecologic analyses versus between -0.01 and +0.15 and +0.01 and +0.27 from their analyses of individual level data. This is hardly a statistically significant discrepancy. In contrast, the discrepancy between the slopes in my data and the LNT predictions are by 20 standard deviations. The reason for this dramatic difference in statistical significance is that the study by Lagarde and Pershagen is based on only 14 counties, whereas mine is based on 1,600 counties, and the average number of cases in the counties in my study is much larger.
In summary, the study by Lagarde and Pershagen, for which the conclusions are severely limited by confounding factors and by statistical significance, does not in any way impact the validity of my study or its conclusions.
|
REFERENCES |
|---|
|
TOPINTRODUCTIONREFERENCESINTRODUCTION REFERENCES |
|---|
-
Lagarde F, Pershagen G. Parallel analyses of individual and ecologic data on residential radon, cofactors, and lung cancer in Sweden. Am J Epidemiol 1999;149:268–74.
[Abstract/Free Full Text] - Cohen BL. Test of the linear-no threshold theory of radiation carcinogenesis for inhaled radon decay products. Health Phys 1995;68:157–74.[Web of Science][Medline]
THE AUTHORS REPLY
Division of Environmental Epidemiology Institute of Environmental Medicine Karolinska Institutet S-171 77 Stockholm, Sweden
|
INTRODUCTION |
|---|
|
TOPINTRODUCTIONREFERENCESINTRODUCTION REFERENCES |
|---|
Cohen (1
) has been impervious to the many theoretical arguments that explain how bias may be easily introduced in ecologic studies (2). Our empirical confirmation of the potential discrepancy of results obtained using group- and individual-level data, respectively, at least suggests that biases contemplated in theory do occur in practice. Taking advantage of the much larger precision in his ecologic estimate, Cohen points out the variability of our risk estimates. However, this remark would have been more relevant if the risk estimates had been obtained among different samples of subjects; the point is that we obtained divergent risk estimates by using the same data set for the ecologic and the analytic approaches. This was also illustrated in table 1 of the paper by Lagarde and Pershagen (2), in which the ecologic pattern shown by relative frequencies of lung cancer cases that tended to decrease with increasing county-specific average radon levels was contradicted by the fact that, within counties, lung cancer cases tended to have larger average radon levels than did controls. We agree with Cohen, as we made exactly the same comment—that confounding is very likely to occur and have a large impact on group-level analyses, especially when the main variation in exposure is between individuals rather than between counties. It is doubtful whether appropriate group-level adjustment for confounding can be made by using factors previously identified as potential confounders in individual-level studies because different factors could be relevant in the two settings and because associations between environmental variables may be stronger and more fortuitous at group level. In addition, the synergy between different exposures, which is most important when assessing an effect that is weak on its own, cannot be suitably modeled at group level. Cohen's assessment of potential residual confounding through simulations is not convincing because such methods rely on the assumption that the experimenter has identified the source of bias that needs to be specified as input in the simulations, which may be a very difficult task, indeed, as Cohen admits.
The precision of an estimate is not a guarantee against bias. Scientists unfamiliar with epidemiologic methods may naively take Cohen's statistically significant negative, group-level association between radon and lung cancer as evidence for a beneficial effect of radon exposure at low levels. This has even been interpreted as a natural consequence of the evolutionary adaptation of organisms to their habitats and the ubiquity of radon (3). The essential point overlooked here is that synergy between tobacco smoke and radon accounts for most of the excess lung cancer risk due to radon, and reciprocally, for part of the risk due to tobacco smoke. In an evolutionary context, it would not be expected for humans to have adapted to the combined effect of radon and tobacco smoke. The same point was also neglected in Cohen's ecologic approach and sensitivity analyses, which cannot appropriately model individual joint exposures.
|
REFERENCES |
|---|
|
TOPINTRODUCTIONREFERENCESINTRODUCTION REFERENCES |
|---|
-
Cohen BL. Re: "Parallel analyses of individual and ecologic data on residential radon, cofactors, and lung cancer in Sweden." (Letter). Am J Epidemiol 2000;152:194–5.
[Free Full Text] - Lagarde F, Pershagen G. Parallel analyses of individual and ecologic data on residentias radon, cofactors, and lung cancer in Sweden. Am J Epidemiol 1999;149:268–74.
- Parsons PA. Hormesis: an adaptive expectation with emphasis on ionising radiation. J Appl Toxicol 2000;20:1103–12.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  S. Greenland Ecologic versus individual-level sources of bias in ecologic estimates of contextual health effects Int. J. Epidemiol., December 1, 2001; 30(6): 1343 - 1350. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. L. Cohen, F. Lagarde, and G. Pershagen RE: "PARALLEL ANALYSES OF INDIVIDUAL AND ECOLOGIC DATA ON RESIDENTIAL RADON, COFACTORS, AND LUNG CANCER IN SWEDEN" Am. J. Epidemiol., July 15, 2000; 152(2): 194 - 195. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||