RSH Data & Documents

"Low Level
Radiation Health
Effects: Compiling
the Data"

Revision 2
March 30, 1999
by Radiation, Science, and
Health, Inc.,
Edited by J. Muckerheide

1.2.6.3.1
Ecological Studies

"The very foundation of the scientific method is the requirement that any theory which is not in agreement with experimental observations must be abandoned (or modified) unless plausible explanation for the discrepancy can be provided or conflicting data which support the theory are unavailable. Cohen (1995) laid out such a situation for radon exposure as a causal agent for lung cancer. Statistically indisputable evidence for a very large discrepancy between observational data and the linear no-threshold theory of radiation carcinogenesis was presented. In spite of the fact that these results are widely known in the scientific community, there have as yet been no convincing explanations offered by others that would support linear no-threshold in this case. There are no other observational data in the low dose region treated in Cohen (1995) that can be interpreted as conflicting with its results. To continue acceptance of the theory in this situation would appear to violate the scientific method.

"Since the linear no-threshold theory is vitally important to many aspects of health physics, it is important that this discrepancy be resolved. This paper offers further analysis of the data used in Cohen(1995). It goes beyond the linear regression analysis of m/mo on r used there, and considers the shape of the m/mo vs. r relationship for countries of various characteristics. Average radon levels, r, are from a recent compilation (Cohen 1992). Age adjusted lung cancer mortality rates are derived from official health statistics (Riggan and Mason 1983). Sources of other data are given in Cohen (1995)."

"Fig. 1 shows a plot for male and female lung cancer rates in all of the 1,601 countries [the data points in it are identical to those in Fig.1 of Cohen (1995)]. The dotted and dashed lines in Fig. 1 show linear and liner + quadratic fits, respectively, to all 1,601 data points. We see that the linear + quadratic gives a much better fit to the data, and considering the error bars (1 standard deviation) that fit appears to be quite good. This indicates that a fit to the plotted points would be very similar to a fit to all 1,601 underlying data points, and therefore the plotted points are a reasonably valid representation of the data. The dashed curves in Fig. 1 will be taken as standards and will appear on all other plots presented in this paper to facilitate comparisons.

Figure 1

"The solid straight lines going upward to the right in Fig. 1 are the predictions of linear no-threshold theory (NAS 1988). If extended, they would go through the data on miners exposed to high radon levels as analyzed in the BEIR-IV Report. These lines will also appear in all other plots.

"These other plots, presented in the same manner, are for counties selected on various bases. That means, of course, that the number of counties included is far less than in Fig. 1, and, therefore, the statistics are substantially poorer."

"Fig. 2 shows the data selected on the basis of percent of population living in urban areas."

Figure 2

"The socioeconomic characteristics studied were selected from the list in Appendix B of Cohen (1995) because they were perceived a priori to be reasonable candidates for confounding the radon-lung cancer relationship. The characteristics selected were [designation in Cohen (1995) in parentheses]: total population (PT); population per square kilometer (PD); rate of population growth (PI); percent of population living in urban areas (PU); percent of population in age range 5-17 y (PY) and age >64 y (PE); average persons per household (PH); birth rate (VB); death rate (VD); physicians per capita (VP) and hospital beds per capita (VH); rate of births to teen-age mothers (VC); percent of adults that are college graduates (SU) and high school graduates (SH); dollars per capita spent on education (SE); crime rate (SC); percent of houses that are owner occupied (HO); percent of households with two or more automobiles (HA); annual income per capita (EI); percent of population below poverty level (EJ); percent unemployment (EU); average wage (farming excluded) (EW); percent of earnings from farming (EF), from manufacturing (EM), services (ES), from retail trade (ER), from government (EG); dollars per capita sales of clothing (EC), and by restaurants (EX); and percent of local government spending allocated to health (GH), welfare (GW), and roads (GR).

"For all of these cases, the plots, which are available on request from the author, are not substantially different from those shown in Fig. 2. Thus, the behavior shown in Fig. 1 applies if we consider only the most urban counties or if we consider only the most rural counties, if we consider only the richest or only the poorest, if we consider only those with the best medical care or only those with the worst medical care, if we consider only the most educated or only the least educated, if we consider only those with the youngest population or only those with the oldest population, etc. It also applies if we consider only those with average urbanicity, only those with average wealth, only those with average medical care, only those with average education, etc.

"The principal exception to similarity with Fig. 2 were cases where the data seemed to lie uniformly above or below the standard curves. Examples of this are shown in Fig. 3 for income per capita and in Fig. 4 (not shown) for physicians per capita. From these we see that the richest counties and those with the best medical care have slightly higher than average lung cancer rates (after correction for smoking prevalence) while the poorest counties and those with the worst medical care have lower than average lung cancer rates. This might be explained by the fact that diagnosis efficiency is better in the former groups and poorer in the latter groups.

Figure 3

"There were indications of data lying consistently above or below the standard curves in stratification studies of some of the other socioeconomic variables, but in no other case was the effect as strong and consistent as in Figs. 3 and 4. But the striking point is that, even in cases like Figs. 3 and 4 where the data are shifted up or down, the shape of the curve through the data is very similar to the shape of the standard curves."

"Since smoking prevalence, S, is a key factor in determining lung cancer rates, that subject was treated extensively in Cohen (1995). One of the approaches used was to stratify the data on S. If estimates of S are very poor, but the ranking of counties by S is reasonably accurate, this approach is sufficient to remove uncertainties in S as a factor in our studies. Data plots with stratification on S are shown in Fig. 6, following the format used in Figs. 2-5. It is interesting to note that S is strongly correlated with r; for the low/medium/high tertiles, mean values of m/m0 and m/mo are 2.03/1.89/1.40 and 0.96/0.98/0.99 for males and 2.00/1.75/1.57 and 0.85/0.88/0.90 for females. Note that while variations in r are large, variations in m/mo are quite small; this indicates that the correction for smoking represented by mo is effective.

Figure 6

"From Fig. 6 we see that the data agree very well with our standard curves. That means that the variation of rn/rn0 with r is the same if we consider only counties with the most smokers, or if we consider only counties with the fewest smokers, or if we consider only counties with average numbers of smokers.

"The most interesting other variable in Cohen (1995) is geography; it is more strongly correlated with radon levels than any other known factor (Cohen 1991). The U.S. Bureau of Census divides the nation into four regions: Northeast (ne), North Central (nc), South (s), and West (w). Data plots for each of these are shown in Fig. 7, for males above and for females below.

"In Fig. 7 we see larger deviations from the standard curves than any of studies discussed above, which embrace the entire country. We see that lung cancer rates in the Northeast, corrected for smoking, are substantially (perhaps 10%) higher than the national average. The decrease in m/mo between r/ro = 0.5 and 2.0 seems to be much stronger than average in the South, but very weak in the North Central region.

"Nevertheless, in all cases, the data seem to fit reasonably well to the shape of the standard curves if the latter are shifted up or down.

"Perhaps the most useful application was to understanding the slopes, B, of straight line fits to these data in Cohen (1995). States with many high radon counties like PA, OH, IL, MN, IA, ND, NE, CO, NM, and ID contribute to the standard curve in a region where it is rather flat and are therefore expected to have lower than average B-values, which they do. None of these are from the South region; many states in the South region have few, if any, counties with radon level above 2 ro, and therefore contribute to the standard curve in a region where it is steepest, and are therefore expected to have higher than average negative slopes, which they do. This partially explains the observation by Price (Price 1995) that states in the South region have substantially higher negative slopes than states in other regions."

"The principal conclusion from this work is that the relationship between m/mo and r is essentially the same for each of the 100+ groups identified here by their characteristics, as that relationship for the entire nation; i.e., all are reasonably well fit by the standard curves. None of the m/mo vs. r plots behave according to the prediction of the linear no-threshold theory. This is a disappointment in that the hope in undertaking this project was that there would be differences that could give clues to resolving the discrepancy between theory and observations noted in Cohen (1995)"

"The present conclusion from our work, including Cohen (1995) and this paper, is that there is a huge discrepancy between these data and the predictions of linear no-threshold theory, that despite our extensive efforts no plausible explanation for this discrepancy has been found, that no such explanation has yet been offered by others, and that unless such an explanation is forth-coming, it seems inappropriate to continue using the linear no-threshold theory in this dose region for predictions of the biological effects of radon and radon progeny."