Revision 1
March 19, 1998
by Radiation, Science, and Health, Inc.,
Edited by J. Muckerheide

1.2.6.3
Radon

1.2.6.3.1
Ecological Studies

The BEIR IV report (1988) based upon a linear-no threshold extrapolation of the incidence of lung cancer in uranium mine workers exposed to high radon concentrations, predicts that the lifetime mortality risk of lung cancer is increased linearly by 10.8% per pCi/L. One pCi/L approximates the world average (UNSCEAR 1982) and is equivalent to 0.2 working-level-month (WLM). (NCRP 1984) The American Cancer Society projects for the United States 170,000 new cases of lung cancer in 1993. Accordingly, prior continued home exposure of the population to one additional pCi/L of radon would have produced 18,000 additional new cases of lung cancer in 1993. Five-year survival of treated lung cancer is only slightly more than 10%. Relying upon the BEIR IV theoretical prediction, the Environmental Protection Agency (EPA) considers radon in the home to be the nation’s leading health hazard.

"However, there is no epidemiologic evidence to support the risks predicted by BEIR IV. To the contrary, epidemiologic studies in the United States (Cohen 1986, 1987, 1989), Sweden (Swedjemark 1984), Finland (Castren et al 1984), and China (Hofman et al 1983) with increased radon concentrations up to 12 pCi/L, as well as in those areas below the average radon concentration of 1 pCi/L (George and Breslin 1980, Nero et al 1983, Wrixon 1984) have all demonstrated a negative correlation of lung cancer with radon concentration. For a variety of reasons, these studies which contradict the linear-no threshold theory have been considered invalid by the National Academy of Sciences Committee on Biological Effects of Ionizing Radiation (BEIR), National Council on Radiation Protection and Measurements (NCRP), and the International Commission on Radiologic Protection (ICRP). Criticisms have included poor statistical power, inadequate controls, and inadequate determination of the degree to which data have been altered by smoking and confounding factors such as numerous socioeconomic variables, geography, altitude, and climate. An extensive University of Pittsburgh National Survey of radon in homes was completed in 1992 that addresses these criticisms with excellent statistical power.

"The University of Pittsburgh nationwide study based upon 272,000 measurements in the homes of 1217 counties was completed in 1992. This study and nine individual state studies were normalized to the EPA National Residential Radon Survey. The combined data set compiled from Pittsburgh, states, and EPA studies includes 1729 counties containing nearly 90% of the U.S. population. After deleting Arizona, California, and Florida, states with high retirement migration, and counties with incomplete data, 1601 counties remain included (Cohen 1994). Figure 1 [see Cohen 1995 above] shows plots of mean age-adjusted lung cancer mortality rates (m) for white males (Figure 1a) and females (Figure 1c) vs mean radon levels (r) in homes of all counties within various ranges of r, along with the standard deviation of the mean, first and third quartiles, and the best linear fit to the data for individual counties, m=1a(1+br). These mortality rates are corrected for smoking by use of Bureau of Census Population Surveys of smoking prevalence and BEIR IV risk estimates for smokers and nonsmokers, and are shown together with the best linear fit, M = m/m_o = A + Br in Figures 1b and 1d. BEIR IV theory lines are normalized lines with slope B increasing mortality at a rate of 7.3%/pCi/L. After correction for variations in smoking frequency, there is a very strong tendency for lung cancer mortality to decrease with increasing mean radon level in homes, in sharp contrast to the increased mortality expected from the linear-no threshold theory. The discrepancy between theoretical and measured slopes is by >20 standard deviations. An earlier study based upon data for 965 counties furnished additional details of methodology and somewhat less steep negative slopes of m/m_o, vs r, with the discrepancy between theoretical and measured slopes by 7 standard deviations. (Cohen and Colditz 1994)"

"Many other factors in addition to smoking are carefully analyzed to see whether any can explain this discrepancy. Pittsburgh radon measurements are consistent with EPA and state measurements. Potential problems concerning outliers and sampling issues are demonstrated to be absent. Uncertainties in lung cancer mortality rates (m) and smoking prevalence (S) are given elaborate consideration and shown to be unimportant causes of the discrepancy between theoretical and measured slopes.

"A careful investigation was made of the possibility that one or more socioeconomic confounding factors other than smoking could correlate strongly and with opposite signs with both m and r. Those would introduce a strong negative correlation between m and r which would not be due to a direct causal relationship. There are 54 socioeconomic variables (SEV) which are analyzed singly and in combination. Extensive statistical analysis of the possibility that some combination of SEV may act cooperatively to confound the m-r relationship concluded that the actual effect of confounding by combinations of SEV is to reduce the discrepancy between slopes by no more than 10%. Confounding by geography was also analyzed by considering the 34 states with at least 20 counties having known radon levels. The average of B-values is -6.1 for males and -7.2 for females; reductions in the discrepancy by 8.2% and 7.1%, respectively.

"In addition to the 54 SEV and geography, also considered are the possible confounding physical features of altitude, average winter and summer temperatures, inches of precipitation, number of days per year with more than 0.01 inch precipitation, average wind speed, and percent of time with sunshine as compared with the maximum possible. Studies of these physical features concluded that none is an important confounding factor. The strong decrease in lung cancer mortality rates corrected for smoking frequency with increasing radon exposure is found in only the low altitude states or only the high altitude states; in only the warmest or only the coldest; in only the wettest or only the driest; etc. It is also found in only the states selected where the physical features are close to average. The BEIR IV theoretical prediction of lung cancer mortality from radon exposure corrected for smoking, M = m/m_o = A + Br, does not take into account two recognized r-S correlations: (1) urban houses have 25% lower radon levels than rural houses and urban people smoke more frequently, and (2) houses of smokers have 10% lower radon levels than houses of nonsmokers. An extensive statistical study of the effects of these r-S correlations leads to the conclusion that the BEIR IV prediction of B is reduced from + 7.3 to + 6.9, which contributes very little to decreasing the discrepancy with the large negative values of B, -7.3 and -8.3 obtained from the actual measured and reported data.

"Linear-no threshold theories other than BEIR IV are considered which involve different treatments of smoking. Also considered is the ‘intensity of smoking.’ Analytic statistical study of these considerations lead to the conclusion that other theoretical predictions of B could reduce the discrepancy between 30% and 81%.

"The possibility that an unrecognized confounding factor could explain the discrepancy is recognized. However, the following properties are required of an unrecognized confounder that could resolve the discrepancy: (1) It must have a very strong correlation with lung cancer, comparable to that of cigarette smoking, but still be unrecognized; (2) It must have a very strong correlation of opposite sign with radon levels; (3) It must not be strongly correlated with any of the 54 socioeconomic variables (SEV); (4) It must be applicable in a wide variety of geographic areas and independent of altitude and climate. The first property alone requires of the unrecognized confounder that it must have increased by orders of magnitude since the beginning of this century, and have been much more important in males in the first half of the century, with effects on females rapidly catching up in recent years. The remaining properties impose additional requirements that are also most difficult to meet singly, while to satisfy the four simultaneously becomes incredible. These multiple restrictions upon an unknown confounder make it extremely improbable that one exists that would resolve the discrepancy.

"These tests of the linear-no threshold theoretical prediction of lung cancer mortality induced by radon exposure, with the slope of the line determined by high dose exposures, demonstrate that the theory fails badly by gross overestimation of mortality in low dose, low dose rate range of radiation. A likely explanation is that stimulated biological defense mechanisms more than compensate for the radiation ‘insult’ and are protective against cancer in a low dose, low dose rate range."