ANS National
 Meetings/
 Sessions

June 1995

Peter G. Groer

and

Shili Xiao

(Univ. of Tennessee)

3. Lung Cancer and ²²²Rn-Progeny - Some Methodological Issues

"INTRODUCTION

                "The correlation between the incidence of lung cancer and exposure to ²²²Rn daughters has been reviewed and evaluated comprehensively by the Biological Effects of Ionizing Radiation (BEIR) IV committee.¹ A new National Research Council Committee, BEIR VI, is taking another look at old and new data to derive new estimates of lung cancer risks for residential exposures. The 'classical' approach to estimate the risks from residential ²²²Rn-daughter exposures is through extrapolation from followup studies of uranium and other hardrock miners. Direct estimation of residential risks is being done with case-control studies. The results of one such recent study do not 'agree' with past risk estimates obtained via extrapolation. In the following sections we show by examining the methodology used for analyses that lung cancer risks from ²²²Rn progeny will always be surrounded by a degree of uncertainty and that this residual uncertainty is irreducible. Disagreement between point estimates of risks from different studies is therefore to be expected. This discussion will be limited mainly to followup studies."

"ANALYSIS OF FOLLOWUP STUDIES

                "Modern analyses of followup studies use various mathematical models of the hazard function h(\D,0) (Ref. 1) to describe the dependence of lung cancer incidence on time t, the covariate radiation dose D, and a vector of parameters 0. To keep it simple, we use only one covariate (D). (Cigarette consumption would be another important covariate.) Once a model has been chosen, the model parameters have to be estimated. The so-called likelihood function plays a central role in parameter estimation. Characterization of the uncertainty of the parameter estimates depends on the approach used. We favor the Bayesian approach,² where uncertainty about parameters is characterized with probability densities. This way of describing uncertainty is similar to the way uncertainty is characterized in quantum mechanics. The 'broader' the density the greater is the uncertainty."

TABLE I

Observed and Expected Lung Cancers in Several Uranium Miner Cohorts*

*From BEIR IV report,¹ pp. 123-130.

"SOURCES OF UNCERTAINTY FOR LUNG CANCER RISK ESTIMATES

                "The model used for construction of the likelihood may be the 'wrong' model. Criteria for model selection have been developed and can be used to select a reasonable model² for analysis. The data in Table I were selected from the BEIR IV report¹ and show why the model type at low exposures is so hotly debated. Some studies do not show an excess of lung cancers at low exposures, and one does.

                "Even if the proper model is used, certain assumptions are made to construct the likelihood. Usually it is assumed that the covariates, such as dose, are precisely known. The covariate describing the radiation exposures from ²²²Rn daughters is the working-level month (WLM), which for all miner studies is only roughly estimable. Covariate uncertainty is currently handled with Monte Carlo simulation. Analytical results are sparse, and more work is needed in this area.

                "Subtler is the assumption of independent competing risks that simplifies the formulation of the likelihood. All analyses of miner data make this assumption. Parameter estimates change when this assumption is dropped.

                "These two sources of uncertainty are certainly irreducible. The WLM for past time periods cannot be dramatically improved, and the exact form of the dependence of risks cannot be determined from the data available for cohort studies.

                "Insufficient data are often solely blamed for the uncertainty of risk estimates and, in the same breath, also used to justify more expenditures for research. Additional precise data reduce the uncertainty of parameter estimates, which does not hap- pen when the data are imprecise.

                "Some of the uncertainty of lung cancer risk estimates is irreducible. Therefore disagreements between point estimates of risk are not surprising, and it is illusory to assume that a single 'golden' point estimate could be found by collecting more data. Therefore predictions of lung cancer risks have to be made under uncertainty. This has not been done so far, despite the fact that the methods to accomplish this are available."

1. "Health Risks of Radon and Other Internally Deposited Alpha Emitters," National Academy Press, Washington, D.C. (1988).

2. M. WEST, J. HARRISON, Bayesian Forecasting and Dynamic Model, Springer Verlag, New York (1989).