January 2000 (vol. 16, #2)
1601 N Tucson Blvd #9, Tucson AZ 85716
c 2000 Physicians for Civil Defense
The National Safety Council is a tax-exempt, nongovernmental agency, which describes itself as a consensus- builder. ``We do not have the authority to legislate or regulate. However, we can influence public opinions, attitudes, and behavior'' www.ncs.org -and it does so with tax dollars. Its Environmental Health Center produces the Climate Change Update (heavily slanted toward global warming advocates) under a cooperative agreement with the U.S. Environmental Protection Agency (EPA) and does public outreach on air quality issues (such as radon) under an EPA grant.
The media blitz happens to coincide with the promulgation of a new EPA rule on radon in water (also see CDP 9/99, available at www.oism.org/cdp). Estimating that radon in drinking water causes 168 cancer deaths per year (89% from lung cancer due to breathing radon released from water and 11% from stomach cancer due to drinking the radon-containing water), the EPA proposes a standard as low as 300 pCi/L see ``Proposed Drinking Water Standards for Radon,'' www.epa.gov/OGWDW/radon/proposal.html Even at a standard of 1,000 pCi/L, the radon contribution of water to indoor air would only be 25% of the average natural outdoor concentration.
The EPA in 1991 estimated an annual compliance cost of $180 million. The American Water Works Association (AMWA) estimated $2.5 billion, or about 14 times as high. If 168 deaths could be postponed and shifted to another cause, the cost would be between $1 million and $15 million per person so affected. (Since the radon could not be reduced to zero, the hypothetical benefit is much lower and the cost per postulated beneficiary much higher.) Such cost:benefit calculations are based on the linear no-threshold (LNT) hypothesis of radiation carcinogenesis.
Congress required the EPA to have the National Academy of Sciences (NAS) review and validate the scientific evidence supporting the proposed radon limits. The NAS is, of course, aware that the most extensive and relevant data available contradicts the LNT hypothesis, diverging from LNT predictions by at least 20 standard deviations. (The NRC does not choose to call this discrepancy to the attention of the public.)
The NAS report states: ``The ecologic study of Cohen (1995) is the most comprehensive. It encompasses about 300,000 radon measurements in 1,601 counties in the U.S. The trend of county lung cancer mortality with increasing home radon concentration is strikingly negative, even when attempts are made to adjust for smoking prevalence, and 54 socioeconomic factors....This finding contradicts the existing risk estimates at low exposure, and a sound reason for the significant negative trend should be sought.''
The possibility that the existing risk estimates are simply mistaken is not explicitly entertained. As is usual, Cohen's data are not refuted but rather brushed off with a glib reference to the Ecological Fallacy: ``the ecological studies are ambiguous because no attempt is made to determine actual exposure to individuals in the area of study and no correction is made for smoking, the strongest confounder for lung cancer.''
The NAS also looks at the eight published case-control studies. The largest, performed in Sweden, involved 1,360 cases and 2,847 controls. ``The lung-cancer excess was not statistically significant even for smokers or nonsmokers with over 400 Bq m-3 in the home for over 32 y.''
The NAS concludes: ``All that can be said about domestic risk is that it is low and difficult, if not impossible, to detect.... [N]umerical risk estimates for lung cancer from [radon and its decay products] will rely on projection models from the underground-miner experience'' (Risk Assessment of Radon in Drinking Water, available at www.nap.edu/catalog/6287.html .
The Biological Effects of Ionizing Radiation (BEIR) VI Report uses the LNT relation between radon exposure and lung cancer, conceding only that ``other relationships, including threshold and curvilinear relationships, cannot be excluded with complete confidence.'' The Committee's models were based on epidemiologic studies of miners, in which ``information was limited on other key exposures including cigarette smoking and arsenic.'' The assumption of linearity of risk down to the lowest exposures ``could not be validated against observational data.''
The official authorities on radiation risk thus have no observational data to support their onerous and expensive regulations. Moreover, if the risk of lung cancer is actually lower in homes with radon levels exceeding their standards, the EPA will be causing more lung cancer rather than preventing it. A careful look at the Ecological Fallacy is thus imperative.
``The Ecological Fallacy is a fallacy'' when applied to Cohen's study, conclude Fritz Seiler and Joseph L. Alvarez in a careful review that will appear in Human and Ecological Risk Assessment (see figure and p. 2).
Cohen's measurements are themselves not in dispute. ``They are the only copious source of information on the incidence of lung cancer fatalities correlated with the air concentrations of radon and its progenies in the United States or elsewhere,'' write Seiler and Alvarez. The few other measurements of mortality rates as a function of low- level exposure are ``taken at only a few averaged exposure levels with random and statistical errors which are dramatically larger than those of Cohen's data.'' There are no measurements of comparable quality that would be acceptable as contradictory data. Rather, the EPA is using an extrapolation (dotted line in the figure) in an effort to contradict data; moreover, the extrapolation is arbitrarily forced by the model to go through the origin.
The radon case is a classic example of assessing risk when the errors in the assessment greatly exceed the actual risk. The EPA assumes that the risk must exist, despite data to the contrary. This is religion, not science.
Although the EPA, in the past, has simply chosen not to address the lack of scientific support for a certain standard, another comment period is open until Feb. 4 (see next page). The EPA has been directed by Congress to consider scientific evidence-bus station blurbs notwithstanding.
Procedure for Notice and Comment
The EPA has invited comments on its proposed rule concerning radon in drinking water; the deadline has been extended to 2/4/2000. (Comments must be received by that date.) The rule is published in the Federal Register vol. 64, no. 244, pp. 71367-71368, 12/21/1999. Written comments may be sent to the Radon-222, W-99-08 Comments Clerk, Water Docket (MC-4101), U.S. Environmental Protection Agency, 401 M Street SW, East Tower Basement, Washington, DC 20460. Comments may be submitted electronically in ASCII, WP6.1 or WP8 format to ow-dock- et@epamail.epa.gov or on floppy disk. Be sure to identify them by the docket number W-99-08 and avoid any special characters or encryption.
For further information, see www.epa.gov or call (202) 260-3027. Sample comments are available from Physicians for Civil Defense, (520) 325-2680, and our comments will be posted on our web site www.oism.org/cdp
. A Thought Experiment
Seiler and Alvarez ask: What would have to be done to make Bernard Cohen's data consistent with the LNT hypothesis (i.e. to explain his ``discrepancy'')? We would have to assume that there is a wide spread in the frequency of tobacco smoking in different counties, about twice the best estimate, and that this frequency has a perfect negative correlation with the radon levels in these counties.
It is, of course, far more likely that the LNT hypothesis is simply wrong. Nevertheless, let us correct for all confounding factors in the test population (most of the U.S., in Cohen's data), applying the correlations in reverse to come up with the linear dose-effect relationship (postulated to be correct).
Suppose now, for the Gedankenexperiment, that we wish to make a prediction for the increase in lung cancer for a test population, say that of Denver, CO, which is suddenly subjected to a doubling in the concentration of indoor radon.
``We use the linear model, apply all confounding factors appropriate for the Denver population, and obtain the appropriately distorted expectation values. Now we explicitly make a usually implicit assumption, which we shall call the `equivalent population assumption': it states that the test population is in all important aspects representative for the prediction population. Without this equivalent population assumption, no predictions can be made.
``This equivalent population assumption is nothing new. It must hold true for any dose-effect function of any agent which is derived from results in a test population and then applied to calculate the expected effects in a prediction population. As an example, we can use the situation with the cancer risk coefficient determined in the BEIR V report. The use of this coefficient assumes implicitly that the characteristics of the Japanese test population, first in times of war and later at peace, are the same as the characteristics of today's American prediction population in times of peace. Yet, the only numerical acknowledgment of a possible difference between the characteristics of the two populations made in BEIR V is a note in the discussion of systematic errors (BEIR 1990). Here, making the equivalent population assumption means that the predicted values for the Denver population will be equal to the values of the Cohen data, except for some increase in the systematic uncertainties due to the equivalent population assumption. Thus, as a direct consequence of this near equality, the Cohen data can and should be used directly for a prospective risk assessment. Note that the equivalent population assumption means that smoking corrections are not necessary, and that it is the raw incidence data that must be used.''
Seiler and Alvarez argue that the proper function for making predictions is the experimentally determined correlation between exposure and effects, not the shape of the underlying dose-effect function. The ``ecological fallacy'' is relevant to finding the dose-effect relationship because the confounding factors must be quantitatively well known. However, it is irrelevant to a risk assessment. The best way to estimate risk is to use the Cohen data set. The data points have exceedingly small errors, given in small increments of exposure.
It is seen from the figure that the relative risk is larger than 1 at very low doses, decreasing with larger radon exposures until it is significantly less than 1 (i.e. it displays a beneficial or hormetic effect). At higher doses, the effect levels off, then rises again, becoming greater than 1 at the zero-effects point.
``Obviously, all these well determined effects are hidden in the large uncertainties of the other data, which up to now were the only ones deemed acceptable by the proponents of the linear model. Note that the two data sets are in statistical agreement, but only the Cohen data with their small errors are able to reveal the finer details of the dose dependence.''
Units
Selected References
For an audiotape of BL Cohen's lecture at the 1999 DDP meeting, call (520) 325-2680.