ANS National
 Meetings/
 Sessions

November 1995

Fritz A. Seiler

(IT Corp, Albuquerque)

And

Joseph L. Alvarez

(lT Corp, Englewood)

3. Cancer Risks, Risk-Cost-Benefit Analysis, and the Scientific Method

            “Two main changes in risk analysis are increasingly beginning to influence the manner in which, in the perception of scientists, low-dose modeling of radiation carcinogenesis is supposed to be done. In the past, efforts to model radiation risks have been carried out under the banner of scientific endeavors. On closer inspection, however, it has become obvious that these efforts were not guided by the scientific method and that a change in approach is needed.1 We realize increasingly that risk analysis is not done in a vacuum and that any action taken due to the result of the analysis not only has a benefit in the form of a risk reduction but leads inevitably to an increase in cost and an increase in the risks of persons effecting the benefit. Thus, a risk-cost-benefit analysis should be done and show a clear-cut net benefit before a remedial action is taken.

                “These two changes will require a dramatic change in the approach to low-dose risk modeling. Some important aspects of this statement may not be directly evident, so a short analysis of the situation may be helpful. In a cost-benefit analysis, the benefit of a management action, measured in numbers of human lives saved or injuries averted, has to be compared to the cost of the action measured in dollars. Many methods, such as the Multi-Attribute-Utility Theory, have been developed to solve the obvious problem of comparing cost in human lives to cost in dollars.2 A less obvious problem is the implicit assumption that both the reduction in risk and the increase in cost are estimated with models of roughly the same degree of sophistication and assuming that the numerical uncertainty analysis can account for the remaining differences in reliability.

                “In a risk-cost-benefit analysis, the costs are evaluated in a manner that is likely to yield a best possible estimate and its uncertainty. Similarly, the risk of the remedial action, often a set of different occupational risks, is based on actuarial data and is usually also done in a manner that should yield a best possible estimate. The benefit by risk reduction, however, is usually overestimated dramatically, all in the name of conservation and the benefit of mankind. This well-meaning approach thus leads to an unacceptably biased risk and risk reduction, making these risk models completely useless for risk-cost-benefit analyses.

                “In the application of the scientific method to the problem of low-dose risk modeling, the most important question is, What can we as scientists honestly state about this risk? and the only acceptable procedure is one that calculates a risk using the most appropriate model available and the best possible values for the model parameters to yield the best possible risk value according to the state of the art. This is followed by an exploration of the limits of our knowledge about each one of the model parameters and, if necessary, an error propagation calculation to arrive at the total error of the risk. Again, the question is, What can we honestly say about the standard error? Then and only then should upper limits be calculated and used and statements about the low-dose risk made. Thus, from the standpoint of the scientific method, using conservative models and conservative parameter values is not acceptable. Indeed, in the new paradigm of risk assessment, conservatism is not a virtue but a sin.

                “An example for the application of these new rationales is the definition of a minimum significant risk.3 It is based on the fact that some errors decrease us the risk decreases, but others do not. Thus, as the dose decreases, the risk decreases also but not all components of its standard error. There will thus be a dose for which the risk is comparable to its error, and simple statistical test will show that the assumption that the risk is different from zero can no longer be maintained. As with all quantities, there is a minimum scientifically meaningful risk. An inspection of the BEIR V model and its uncertainties shows that the minimum significant risk lies in the range of a few percent for a 90 to 95% confidence level. For an improved error analysis of the BEIR V data, minimum significant risks of a few times 10^-3 might be obtained, but values near or below 1 x 10^-3 are probably out of the reach of these studies. The important consequence of these evaluations is that any risk smaller than the minimum significant risk is scientifically meaningless.”

1.F. A. SEILER, J. L. ALVAREZ, "The Use of the Scientific Method in Risk Analysis," Technology, 331A, 53 (1994).

2.R. L. KEENEY, "Decision Analysis," Handbook of Operations Research: Foundations and Fundamentals, J. J. MODER, S. E. ELMAGHRABY, Eds., Van Nostrand  Reinhold, New York (1978).

3.F. A. SEILER, J. L. ALVAREZ, "The Definition of a Minimum Significant Risk," Technology, 331A, 83 (1994).