3
Models and Risk Projections

INTRODUCTION

This chapter presents the committee's risk models relating lung-cancer to radon exposure and applies the models to exposures of the general population to estimate the burden of lung-cancer due to exposure to indoor radon. We discuss both the committee's models describing lung-cancer risk in miners and the application of the models in projecting lung-cancer risks in the general population. We also describe prior risk models and the basis for our approach to developing new risk models. The committee decided to use primarily miner-based data for risk estimation and to use models in which risk is linearly related to dose at low doses. Those two decisions follow those of the BEIR IV committee. However, the rationale for our model is supported more strongly than was that of the BEIR IV committee, being grounded in the biologic considerations developed in chapter 2 and in the stronger body of observational evidence provided by the pooled data from the studies of underground miners, as well as a meta-analysis of the reported 8 case-control studies of residential radon exposure and lung-cancer.

In this chapter, we provide risk projections that describe both the increment in lifetime risk of lung-cancer mortality for various exposure scenarios and the population burden of lung-cancer attributable to exposure to indoor radon. This chapter also addresses uncertainties associated with the models and with risk projections based on the models. Appendix A describes the modelling and uncertainty analysis procedures in detail.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI 3 Models and Risk Projections INTRODUCTION This chapter presents the committee's risk models relating lung-cancer to radon exposure and applies the models to exposures of the general population to estimate the burden of lung-cancer due to exposure to indoor radon. We discuss both the committee's models describing lung-cancer risk in miners and the application of the models in projecting lung-cancer risks in the general population. We also describe prior risk models and the basis for our approach to developing new risk models. The committee decided to use primarily miner-based data for risk estimation and to use models in which risk is linearly related to dose at low doses. Those two decisions follow those of the BEIR IV committee. However, the rationale for our model is supported more strongly than was that of the BEIR IV committee, being grounded in the biologic considerations developed in chapter 2 and in the stronger body of observational evidence provided by the pooled data from the studies of underground miners, as well as a meta-analysis of the reported 8 case-control studies of residential radon exposure and lung-cancer. In this chapter, we provide risk projections that describe both the increment in lifetime risk of lung-cancer mortality for various exposure scenarios and the population burden of lung-cancer attributable to exposure to indoor radon. This chapter also addresses uncertainties associated with the models and with risk projections based on the models. Appendix A describes the modelling and uncertainty analysis procedures in detail.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI RISK-ESTIMATION APPROACHES This section briefly reviews alternative approaches to estimating lung-cancer risk associated at radon exposure levels typically found in homes and provides the rationale for the committee's selected approach. Figure 1-3 showed the alternative approaches considered and the related data sources. Dosimetric Approach The dosimetric approach applies the well-characterized radiation data from human exposures to γ rays, in particular data from the atomic-bomb survivors, to derive estimates of the risk associated with exposure to radon (ICRP 1990). This approach has the following steps: Use physical dosimetric models of the lung to estimate alpha-particle dose to lung-airway epithelium for indoor radon exposure. Convert the alpha-particle dose to an equivalent low-linear-energy-transfer (low-LET) dose for low-LET radiations, using an appropriate weighting factor for radon-progeny alpha particles in the bronchial epithelium. Convert the equivalent dose to an effective dose, using the appropriate tissue-weighting factor for lung (ICRP 1990). (It is possible to omit step 3 and use lung-specific γ-ray-based risk estimates in step 4). Use risk coefficients per unit of effective dose, based primarily on atomic-bomb survivor data, to estimate the risk per unit of cumulative exposure to radon. One strength of this dosimetric approach is its use of the wealth of data from the continuing epidemiologic study of the atomic-bomb survivors in Hiroshima and Nagasaki. Lung-cancer risk has been well characterized in that cohort in relation to dose. However, the approach is weakened by the need for scaling factors to convert from the acute, whole-body, primarily γ-ray exposure to the chronic, localized, alpha-particle exposure of the lung from indoor radon. In addition, the data from Hiroshima and Nagasaki are subject to uncertainty owing to limitations of the dosimetry and the need to extrapolate from an exposed population in Japan to other population groups with differing background cancer rates. Biologically Motivated Approach Biologically motivated models are intended to provide realistic representation of the steps in radon carcinogenesis from energy deposition to the appearance of cancer. In this context, the parameters of the model have a direct biologic interpretation. One such model is the Moolgavkar-Venzon-Knudson 2-stage clonal expansion model, which incorporates both tissue growth and cell kinetics (Moolgavkar and Luebeck 1990). Such approaches to cancer risk estimation

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI have been proposed and reviewed by various authors (for example, Little and others 1992, 1994; Moolgavkar and others 1993; Crump 1994a,b; Moolgavkar 1994; Goddard and Krewski 1995; Little 1995). This committee did not pursue biologically motivated cancer-risk models for several reasons. First, the mechanisms of radon-induced carcinogenesis must be known with sufficient certainty before an appropriate biologically motivated model can be constructed. Despite the considerable amount of information summarized in chapter 2, the committee recognized that current knowledge of radiation cancer mechanisms remains incomplete and any postulated model would necessarily be an oversimplification of a complex process. Second, application of a fully biologically motivated model requires information on fundamental biologic events, such as mutation rates and cell kinetics, that is not readily available in the present application. Third, a comprehensive biologically motivated model involving many parameters, such as the 2-stage clonal-expansion model used by Moolgavkar and others (1993) to describe the Colorado miner data, cannot be fruitfully applied without comprehensive longitudinal data on personal exposures to both radon progeny and tobacco. When the various steps in radon-induced carcinogenesis are more fully understood, the biologically motivated approach might become the preferred approach. However, the committee considered an empirical approach to be preferable at present. Empirical Approach Statistical methods for the analysis of epidemiologic data, particularly cohort data, have evolved rapidly since the 1970s. These statistical methods can be used to estimate lung-cancer risks directly from epidemiologic data, as done by the BEIR IV committee. To implement the now-common empirical approach, it is assumed that disease rates in narrow time intervals are constant, or at least can be accurately approximated by mean disease rates in the time intervals. Epidemiologic cohort data are summarized in a multidimensional table, in which each cell contains information on person-years at risk, number of events (lung-cancer deaths) occurring within the cell, and variables that identify the cell, such as age, cumulative exposure, and exposure rate. For each cell, the observed number of events is assumed to follow a Poisson distribution, with a mean equal to the underlying disease rate for the cell multiplied by the person-years at risk. Poisson events are assumed to be infrequent and have a distribution in which the variance equals the mean. In the development of an empirical risk model to describe rates of radon-induced lung-cancer in miners, several a priori assumptions are needed about either the shape of the exposure-response function or the factors that influence risk. In its most general implementation, empirical modeling is sufficiently flexible to offer some degree of biological plausibility with only minimal assumptions needed about the structure of the model. That generality, as well as the

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI ability to model without assuming any underlying biologic mechanism of disease, leads to the characterization of the modeling approach as empirical or descriptive. The empirical modeling approach also allows for evaluation of diverse factors that modify risk, such as attained age and exposure rate, through formal statistical testing. Given the limitations of the available data and the resulting difficulty in discriminating among plausible alternative models, the empirical approach undoubtedly results in models that are relatively crude and at best yield rough approximations of actual patterns of risk. While the committee relied on data on lung-cancer mortality in underground miners to construct its proposed risk models, a series of assumptions is needed to extend the miner-based model to the general population. For example, the committee used models in which the exposure-risk relation is linear at low exposures, based on the mechanistic considerations discussed in chapter 2. Other assumptions made in projecting population risks are described later in this chapter. RATIONALE FOR THE COMMITTEE'S CHOSEN METHOD FOR RADON RISK ESTIMATION The committee critically assessed the principal approaches (see Figure 1-3) that could be used to estimate the risk associated with exposure to indoor radon, with respect both to sources of data for developing risk models and to techniques for modeling. The combinations of data resources and risk estimation approaches of present interest are as follows: Biologically motivated analysis of miner data. Dosimetric approach using low-LET data (for example, atomic-bomb survivor data). Empirical analysis of miner data. Empirical analysis of data from residential case-control studies. The strengths and limitations of the three different data sources are summarized in Table 3-1. With regard to the first approach, the committee recognized that use of biologically motivated risk models is a highly desirable goal, but it felt that such models have not reached a stage at which they can be used for radon risk assessment. Specifically, the complexity and multiplicity of the processes involved in radiation carcinogenesis were noted, as were the gaps in knowledge of the most-basic relevant processes. The paradigms describing carcinogenesis in general and radiation carcinogenesis in particular are changing rapidly. For example, the potential importance of delayed genomic instability (Chang and Little 1992; Kadhim and others 1992; Morgan and others 1996), not incorporated in currently formalized biologically motivated models, was not apparent until within the last few years. The second approach, the dosimetric approach based on the atomic-bomb

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI TABLE 3-1 Relative strengths of data for alternative approaches for estimating the risks posed by indoor radon Criteria Atomic-bomb survivor data Residential data Miner data Exposure estimation ••• • •• Potential power of study ••• • ••• Dose range • ••• • Exposure-time • ••• •• Women/children ••• ••• • Effects of smoking • •• • Scaling factors required • ••• •• ••• = adequate •• = fair • = problematic survivors, has both strengths and weaknesses. Its strengths include the availability of estimates from a large cohort of men, women, and children exposed to a wide range of doses; the extensively characterized dose estimates for the survivors; and the 45-year period follow-up. For the present application to radon progeny, weaknesses include the very different types of radiation and exposure patterns to which the bomb survivors were exposed—acute whole-body doses of gamma rays and, to a lesser extent, neutrons. In particular, the radiation weighting factor needed to relate gamma-ray risks to alpha-particle risks is probably not known to better than within a factor of about 5 (Burchall and James 1994; Brooks and others 1994; Brenner and others 1995). The risk estimates from the study of atomic-bomb survivors are also subject to uncertainty (NRC 1990). The committee reasoned that the uncertainties in extrapolating risks from acute whole-body γ-ray exposure to prolonged, localized alpha-particle exposure were too great to justify use of this approach. Over the last decade, considerable resources have been devoted to case-control studies of residential radon exposure. A number of studies have been completed, and some are still in progress. These studies are reviewed in appendix G. In principle, residential studies yield the most relevant risk estimates, because they relate directly to the population of interest. However, because of the very low risk associated with exposures at residential levels, risk estimates obtained from these studies, even estimates based on meta-analysis of several studies, are very imprecise. Furthermore, the residential studies offer little opportunity for evaluating with the modifying effects of such factors as smoking and time since exposure. For those reasons, the committee rejected a model based on the residential-radon studies (the fourth approach) as the primary source of risk estimation. However, the committee did compare risk estimates based on the residential data with the low-exposure risk estimates that it generated from the miner data.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI Having thus considered the various alternative approaches, the committee chose to follow the general approach of the BEIR IV committee and of Lubin and others (1994a) to the analysis of pooled miner data and to base risk estimates on an empirically derived model (the third approach). This approach provided the committee with well-established databases and methods as a starting point. Empirical or descriptive modeling of risk allows a unified approach for testing the validity of the form of the model and of the significance of model parameters. For example, the committee used a relative risk model rather than an absolute risk model to describe lung-cancer risk to radon exposure. It had been observed that a relative risk model, with time-varying covariates, provided a more parsimonious description of the miner data than an absolute excess-risk model (Lubin and others 1994a). The flexibility of the modeling approach allowed the incorporation of specific biologically based patterns of risk. Two important choices in the committee's analysis are the incorporation of an inverse exposure-rate effect and the assumption of linearity of the exposure-response relationship at low cumulative exposure. For radon-induced lung-cancer, those choices have a plausible biologic rationale, as well as some experimental justification (see chapter 2). PREVIOUS MODELS A number of models have been previously developed for estimating lung-cancer risk posed by exposure to radon and its progeny. Models developed through the middle 1980s were described in the BEIR IV report (NRC 1988). These and other models are discussed in detail in appendix A to this report. With the exception of preliminary reports from 2 studies which later changed, these models have all assumed linearity of the exposure-response relationship. All models used risk estimates derived from the studies of miners. The earliest risk models specified effects of exposure in terms of the absolute excess risk of lung-cancer from radon-progeny exposure. The absolute (excess) risk model represents lung-cancer mortality as r (x, z, w) = r0 (x) + g (z, w), where r0 (x) is the background lung-cancer rate and g (z, w) is an effect of exposure. (Here, w denotes cumulative exposure, x represents covariates that determine the background risk, and z denotes covariates that modify the exposure-response relationship.) The model proposed by the BEIR III committee allowed the absolute excess risk to vary by categories of attained age with allowance for different minimal latent periods for each category. A descriptive model for the absolute excess lung-cancer risk, proposed by Harley and Pasternack (1981), served as the basis of risk estimates in Reports 77 and 78 of the National Council of Radiation Protection and Measurements (NCRP 1984a,b). That model assumed that exposure has no effect on risk before age 40 years and that, after a latent period, the absolute excess risk declines exponentially with time since exposure. The model was proposed specifically to address risk associated with radon-progeny expo-

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI sure, and, although miner data were not used to define its form, published results of analyses of miner data were used to specify latent periods and parameter values thought to be reasonable and appropriate (NRC 1988). A meta-analysis of miner-study results by Thomas and McNeil (1982), the BEIR IV committee analysis of pooled data from 4 miner cohort studies (NRC 1988), and the findings from a number of individual cohort studies of miners suggested that models of the relative risk (RR) were preferable to models of the absolute excess risk. Recent descriptive models for lung-cancer risk associated with radon-progeny exposure have also modeled the relative risk rather than the absolute risk. Under the general relative risk model, the lung-cancer rate r (x, z, w) can be written as r(x, z, w) = r0 (x)RR(z, w), where r0 (x) is the lung-cancer rate among nonexposed, and RR (z, w) is the exposure-response function. Of particular interest is the linear relative risk model: RR = 1 + ßw, (1) where ßw estimates the excess relative risk (ERR), w is exposure, and ß estimates the increment in ERR for unit change in exposure w. The simplest of the relative-risk models was proposed in Report 50 of the International Commission on Radiological Protection (ICRP 1987). The ICRP model for extrapolation to indoor exposures was a linear model for ERR in relation to cumulative exposure. The value of ß was derived by reducing a value thought to be representative of the miner studies to reflect differences in conditions between mines and homes. On the basis of findings in the atomic-bomb survivors and dosimetric considerations, ß was increased by a factor of 3 for exposures occurring before age 20 years. The assumption of a constant relative risk and the higher risks assigned to exposures in childhood can be questioned. Detailed analyses of miner data, however, have indicated that the exposure-response relationship is not constant but varies with other factors (NRC 1988; Thomas 1981). In addition, there is little evidence of enhanced effects of exposure at young ages in the miner data (Lubin and others 1994a). The BEIR IV committee analyzed pooled data from 4 cohort studies of radon-exposed miners (NRC 1988). It found that the simple linear ERR model did not fit the data adequately and that the exposure-response parameter ß varied with time since exposure and attained age. Since its publication in 1988, the BEIR IV model has served as the primary basis for assessing risks for underground miners and the general population. Using the BEIR IV model as a starting point, Jacobi and others (1992) proposed a related "smoothed" model for the relative risk of lung-cancer from radon-progeny exposure, which served as the basis of risk estimation in ICRP Report 65 (ICRP 1993). Expanding the analytic approach in the BEIR IV report, Lubin and others (1994a, 1995b) pooled data from 11 cohort studies of miners, including the 4 studies used in the BEIR IV analysis, and fitted similar types of models for the ERR. Lubin and colleagues

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI (1994a) again found that the exposure-response relation varied with time since exposure and attained age, but they also found variation with exposure rate. Lower exposure rates were associated with increased risk. The BEIR VI committee used the work of Lubin and colleagues (1994a) as a starting point for the analyses described in this chapter. BEIR VI RISK MODEL FOR LUNG CANCER IN MINERS Introduction This section considers the sources of data, methods of combining data from diverse populations, and assumptions that underlie the lung-cancer risk model developed by the committee in its analysis of miner data. The committee used a relative-risk model that relates lung-cancer rate in miners to their occupational exposure to radon. In the analysis, exposure refers to occupational exposure to radon progeny during employment in underground mines, and relative risks refer to the additional risks associated with occupational exposure to radon progeny beyond the background risk from lung-cancer, which reflects other exposures, including indoor radon. Residential radon-progeny exposures of the miners are not considered in the analysis data and are implicitly assumed to be the same, on average, at all levels of occupational exposure. Any bias in the modeling due to ignoring nonmine exposures is likely to be small, because residential radon concentrations are generally much lower than mine concentrations. The committee's model is based on a linear relationship between exposure and the relative risk of lung-cancer. This linear relationship was based on an empirical evaluation of the 11 individual miner studies. In analyzing the miner data, Lubin and others (1994a) explored various models for describing the form of the relative risk in relation to radon exposure. Within the range of exposures in miners, linear models provided an adequate characterization of each cohort except the Colorado Plateau uranium miners. In the Colorado data, the authors found a relative-risk pattern that was concave at high cumulative exposures. Accordingly, in the analysis of pooled data, data from the Colorado study were limited to exposures below 11.2 Jhm-3 (3,200 WLM), below which relative risks were consistent with linearity. Sources of Data Pooled data from 11 cohort studies of radon-exposed underground miners were used to develop the committee's risk models; these data were derived from all the major studies with estimates of exposure for individual miners (Table 3-2). Data were available from 7 studies in addition to those considered by the BEIR IV committee. These data are described in detail in appendixes D and E.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI TABLE 3-2 Epidemiologic studies of underground miners used in the BEIR VI analysisa Location Type of mine Numbers of miners Period of follow-up Data available on smoking China Tin 17,143 1976–87 Smoker: yes/no (missing on 24% of subjects, 25 (out of 907) nonsmoking lung-cancer cases) Czechoslovakia Uranium 4,320 1948–90 Not available Colorado Uranium 3,347 1950–90 Cigarette use: duration, rate, cessation (unavailable after 1969) Ontario Uranium 21,346 1955–86 Not available Newfoundland Florspar 2,088 1950–84 Type of product, duration, cessation (available for 48% of subjects, including 25 cases) Sweden Iron 1,294 1951–91 Type of product, amount, cessation (from 35% sample of active miners in 1972, supplemented by later surveys) New Mexico Uranium 3,469 1943–85 Cigarette use: duration, rate, cessation (available through time of last physical examination) Beaverlodge Uranium 8,486 1950–80 Not available Port Radium Uranium 2,103 1950–80 Not available Radium Hill Uranium 2,516 1948–87 Smoking status: ever, never, unknown (available for about half the subjects, 1 nonsmoking case) France Uranium 1,785 1948–86 Not available a Lubin and others 1994a. Since the 1994 publication of the original pooled analysis by Lubin and colleagues (1994a), data from 4 studies (Chinese tin miners and the Czechoslovakia,1 Colorado and French uranium miners) have been updated or modified (Lubin and others 1997). In assembling the original data for the China study, the original investigators (Xuan and others 1993) assumed that all miners worked 285 d/yr until the early 1980s, which corresponded to the end of the follow-up less the 1   For historical reasons, the study is referred to as the Czechoslovakia or Czech cohort, although the country is now 2 independent states, the Czech Republic and Slovakia. The mining area was located in what is now the Czech Republic. About 25% of the miners were of Slovak origin and most later returned to Slovakia.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI 5-yr lag period. Recent information has indicated, however, that miners worked 313 d/yr before 1953, 285 d/yr in 1953–1984, and 259 d/yr after 1984. Estimates of exposures have been updated accordingly. An extensive reevaluation of exposure histories and of follow-up and vital status has been carried out for the Czech cohort (Tomásek and others 1994a). There were 705 lung-cancer cases in the updated data, compared with 661 in the previous data set, and the cohort was enlarged from 4,284 to 4,320 miners. For the Colorado study, follow-up has been extended from December 31, 1987, through December 31, 1990 (Hornung and others 1995). In the updated data used by the committee, there were 336 lung-cancer deaths at exposures under 11.2 Jhm-3 (under 3,200 WLM) in a total of 377 cases, compared with 294 lung-cancer deaths at exposures under 11.2 Jhm-3 in a total of 329 total cases in the prior pooled analysis. For the French miner data, the investigators made small corrections in exposure estimates and in health outcomes other than lung-cancer. In addition to the data changes for those cohorts, there has been a reassessment of estimates of exposure of a nested case-control sample within the Beaverlodge cohort of uranium miners, including all lung-cancer cases and their matched control subjects (Howe and Stager 1996). For these Beaverlodge miners, exposure estimates were about 60% higher than the original values. Because of the computational difficulties of merging case-control data with cohort data, only the data from the Beaverlodge cohort study with the original exposure estimates were used in the committee's analysis. Analysis of Pooled Data from Different Studies In the development of risk models, it is important to take account of the totality of evidence from all relevant studies. When data from many different sources are available, this is most effectively accomplished by analyzing combined or pooled data. The models developed by Lubin and others (1994a) were based on analyses of data from 11 miner cohorts. Other examples of analyses of pooled data are those by Cardis and others (1995) on cohorts of externally irradiated nuclear workers in the United States, the United Kingdom, and Canada and Lubin and others (1994b) on data from 3 case-control studies of indoor radon exposure and lung-cancer. Analyses of pooled data can provide more precise estimates of parameters than those based on individual studies—an advantage that is especially important for investigating modifying factors, which requires comparing risks among subsets of the data. They can also test whether differences in findings among studies represent true inconsistency or simply result by chance. The application of similar methods to data from all studies and the presentation of results in a comparable format facilitate comparisons of results from different studies. Analysis of pooled data from diverse sources must, however, be done with care because the data might not be fully comparable. In the present context, the

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI cohorts differ with respect to the methods used to estimate radon-progeny exposure, the completeness of mortality follow-up, and the accuracy of disease diagnosis. The cohorts also differ with respect to demographic characteristics, other exposures encountered in the mines, and smoking patterns. Such differences can lead to heterogeneity in risk estimates. Heterogeneity can be partially addressed by adjusting for modifying factors, such as exposure rate, on which data are available. However, lack of adequate data on all covariates that affect risks and biases in the data can result in residual heterogeneity even with extensive adjustment for covariates. It is important to take account of heterogeneity in analyzing the data, particularly in expressing the uncertainty in the risk estimates obtained. Statistical methods for analyzing data sets derived from different sources, taking into account heterogeneity among sources, are described in appendix A. Random-effects models (Davidian and Giltinan 1995) provide a natural statistical approach for combining data from different sources in the presence of heterogeneity. Specifically, heterogeneity is accommodated by allowing for random perturbations in parameter values from cohort to cohort, and this results in a random-effects distribution of parameter values across cohorts. The mean of the distribution constitutes an overall summary of the parameter value across cohorts and its variance describes the component of uncertainty due to unaccounted for differences between cohort studies. Two-stage statistical methods have also been used in analysis of pooled data from different studies. With the 2-stage approach, estimates of the model parameters specific for each cohort are derived, and an overall estimate is then obtained by an appropriately weighted linear combination of the cohort-specific estimates, taking into account variation within and between cohorts. The 2-stage approach was used in recent analyses by Lubin and others (1994a) and also by Burnett and others (1995) in combining data on air pollution and respiratory health in 16 Canadian cities. Both the random-effects and 2-stage approaches were used to combine data from the 11 miner cohorts (see appendix A), but the results presented in this chapter are based on the 2-stage method. In simple modeling situations, the 2-stage and random-effects models were generally found to be in good agreement. In the more-complex modeling conducted by the committee, however, the random-effects approach proved to be computationally more burdensome, and convergence was not always obtained with the iterative numerical methods required in model fitting. Consequently, the committee relied primarily on the 2-stage method in conducting its combined analyses. The committee's 2-stage approach can be viewed as a simplification of, and an approximation to, the full random-effects approach. The committee recognized that each of the 11 miner studies has certain unique characteristics that contribute to the observed differences in risk among cohorts. In the presence of such differences, the desirability of pooling data from heterogeneous populations can be questioned. Pooling makes maximal use of all relevant data in an objective manner and provides an overall summary measure of

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI tion of radon concentration in U.S. homes, and the dosimetric K-factor. The methods adopted by the committee to evaluate uncertainty in the AR require specification of the prior uncertainty in each of the factors affecting risk. As detailed in appendix A, the uncertainty in the parameters in the BEIR VI risk models was described by lognormal distributions, with dispersion at least as great as the sampling error in the estimated parameter value. Variability among radon concentrations in U.S. homes was characterized by a lognormal distribution used to describe the results of the National Residential Radon Survey; uncertainty in individual radon measurements (measurement error) was not addressed in this analysis. Variability in the K-factor was also described by a lognormal distribution based on a sample of observations in U.S. homes; uncertainty in K-factors for specific homes was described by a log-uniform distribution. Although the committee exercised some judgment in specifying distributions, the result represents a best attempt to allow for some degree of uncertainty in a number of the critical factors affecting the AR. Because the AR is not a simple multiplicative function, Monte Carlo methods were used to evaluate uncertainty under the committee's preferred risk models. As shown in Figure 3-3, this analysis leads to an uncertainty distribution reflecting the likelihood of different possible values for the AR, centered roughly at the best estimates of the AR given in Table 3-7. Because the uncertainties in the model parameters are largely statistical, the uncertainty distributions reflecting only uncertainty in the parameters of the committee's risk models (Figure 3-3, case I) can be used to obtain approximate confidence intervals for the AR. For males (Figure 3-3a), 95% of the mass of this distribution falls in the range 0.09–0.24 for the exposure-age-concentration model, and 0.07–0.16 for the exposure-age-duration model. For females (Figure 3-3b), the corresponding limits are similar: 0.10–0.26 for the exposure-age-concentration model and 0.08–0.18 for the exposure-age-duration model. In this analysis, a constant value of K = 1 was used. A second uncertainty analysis was conducted in which variability in K was taken into account (Figure 3-3, case II). Allowing for variability in K does not increase the dispersion of the uncertainty distribution for the AR, but does shift the distribution to the right. For males, the 95% uncertainty intervals for the exposure-age-concentration and the exposure-age-duration models were 0.10–0.22 and 0.08–0.18, respectively. For females, the corresponding limits were 0.10–0.28 and 0.08–0.19, respectively. The final analysis of uncertainty in the AR further acknowledged uncertainty in the observed radon concentrations in U.S. homes as well as uncertainty in K (Figure 3-3, case III). This increased the range of uncertainty in the AR under both risk models. For males, the 95% uncertainty intervals were 0.10–0.26 for the exposure-age-concentration model and 0.08–0.19 for the exposure-age-duration model. For females, the corresponding limits were 0.10–0.28 and 0.09–0.29, respectively.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI FIGURE 3-3a Uncertainty distributions for the population attributable risk (AR) for males. I: uncertainty in model parameters. II: uncertainty in model parameters; variability in K; variability in radon levels. III: uncertainty in model parameters; uncertainty/variability in K; variability in radon levels.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI FIGURE 3-3b Uncertainty distributions for the population attributable risk (AR) for females. I: uncertainty in model parameters. II: uncertainty in model parameters; variability in K; variability in radon levels. III: uncertainty in model parameters; uncertainty/variability in K; variability in radon levels.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI The uncertainty in the estimated values of the AR gives bounds to predictions of the number of lung-cancer cases attributable to residential radon exposure in the United States. For the exposure-age-duration model, the approximate 95% confidence limits on the AR imply a range of 11,400–26,200 lung-cancer cases for males and females combined. This range reflects statistical uncertainty in the central estimate of 15,400 cases given in Table 3-10. For the exposure-age-concentration model, the corresponding range is 14,800–38,600 cases, with a central estimate of 21,800 cases. The uncertainty ranges that take into account variability in K (case II in Figure 3-3) and uncertainty/variability in K (case III in Figure 3-3) are comparable in width but shifted slightly to the right because of the off-centering effect apparent in Figure 3-3 when variability in K is incorporated in the analysis. It is important to note that although these uncertainty limits encompass 95% of the mass of the uncertainty distributions in Figure 3-3, the uncertainty distributions place most of the mass nearer to the central values, indicating that values closer to the center of the distribution are most likely. Because both the exposure-age-duration and exposure-age-concentration models fit the miner data equally well, the committee was unable to express a preference for either model. However, the committee noted that in as much as these 2 models are based on exposure levels in mines that generally exceed those in homes, model-based projections of the number of lung-cancer cases due to the presence of radon in U.S. homes are appropriate only if the models apply equally well at residential exposure levels. To address that issue, the committee also calculated 95% uncertainty intervals for the projected number of lung-cancer cases attributable to residential radon exposure by using the constant-relative-risk (CRR) model restricted to exposures less than 0.175 Jhm-3 (50 WLM) (Table 3-9). Because this simple CRR model involves only a single unknown parameter ß (estimated to be 0.0117/WLM), 95% confidence limits on ß (0.002–0.225/WLM) can be used to obtain corresponding confidence limits on the AR. This simple uncertainty analysis, which focuses on the subgroup of miners with exposure levels closest to those in U.S. homes, provided 95% confidence limits of 3,300–32,600 lung-cancer cases about the central estimate of 17,500 cases based on the estimates of the AR given in Table 3-9. Although these confidence limits are wider than those based on the committee's 2 preferred models because of the smaller sample, the CRR model is based on observations closest to residential exposure levels. As discussed previously and in appendix A, other factors might contribute to uncertainty beyond those included in this analysis. Nonetheless, this limited analysis does indicate that the population AR of lung-cancer due to radon in homes is subject to considerable uncertainty. The committee acknowledges that this analysis of uncertainty and variability depends on the specific assumptions made about uncertainty and variability in each of the factors affecting the AR. Because characterization of variability and especially uncertainty in the factors is

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI difficult, these particular assumptions reflect to a large extent the committee's best judgement. COMPARISONS WITH BEIR IV The BEIR VI committee's risk models are closely related to the model developed in the BEIR IV report. The BEIR IV committee combined data from 4 cohort studies of underground miners (Colorado Plateau, Ontario, Sweden, and Beaverlodge studies) and applied Poisson regression methods in model fitting. The starting point for the BEIR VI committee was the recent pooling of 11 studies by Lubin and others (1994a, 1995a), which included the same or updated data from the original 4 cohorts and data from seven additional cohorts. Closely comparable statistical methods were applied by both the BEIR IV and BEIR VI committees. BEIR IV AND BEIR VI RISK MODELS The committee's models are a direct extension of the BEIR IV model, which included parameters for time since exposure and attained age, but not exposure rate or exposure duration, as in the BEIR VI models. The form of the BEIR IV model is obtained from the BEIR VI models by setting and γz = 1, that is, Parameter values for the BEIR IV model and the BEIR VI models show the same general declining patterns for increasing time since exposure and attained age. The decline in the ratio of ERR per unit exposure with attained age, however, is more pronounced in the current models. Note that the values for β are not directly comparable, because the values reflect different baseline levels due to the inclusion of different modifying factors in the BEIR IV and BEIR VI models. The overall ERR per unit of exposure in the absence of all modifying factors is not an adequate description of the relative risk from the miner studies and should not be used for formal comparisons. Nevertheless, the estimate of the overall ERR/Jhm-3 for the 4 cohorts used in the BEIR IV report was 3.8 Jhm-3 (0.0134/WLM), whereas the value was 1.4 Jhm-3 (0.005/WLM) for the pooled analysis of the 11 miner cohorts (Lubin and others 1994a). Those values afford a somewhat crude comparison, suggesting that the combined risk for miners in the 11 studies was less than the BEIR IV estimate. Figures 3-4 through 3-6 show more-direct comparisons of estimated LRR for selected exposure patterns in miners. Figure 3-4 shows LRRs by exposure rate from 0–5.95 Jm-3 (0–10 WL) for 5, 10, and 20 yr of exposure. BEIR IV estimates of LRRs for exposure rates of 0.60 Jm-3 (1.0 WL) and greater were higher than estimates from current models. Similar patterns are seen in Figure 3-5, which shows LRRs by duration

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI FIGURE 3-4 Predicted lifetime relative risk of lung cancer by radon progeny exposure rate for male miners exposed starting at age 25, for 5, 10, and 20 years.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI FIGURE 3-5 Predicted lifetime relative risk of lung-cancer by duration of radon-progeny exposure for male miners exposed at various rates.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI of exposure for exposures at constant rates of 0.30 Jm-3 (0.5 WL), 0.60 Jm-3 (1.0 WL), and 2.98 Jm-3 (5.0 WL). Figure 3-6 provides comparisons of the various projections of LRR for lifetime exposure to radon at concentrations found in homes. A K-factor of 1.0, an equilibrium ratio of 0.4, and 70% home occupancy are assumed. Slightly higher LRRs are estimated with the exposure-age-concentration model than with the exposure-age-duration and BEIR IV models; however, estimates of LRRs are generally similar for concentrations of 1000 Bqm-3 (27.03 pCiL-1) and below, levels that include the large majority of dwellings. SUMMARY AND CONCLUSIONS Radon is one of the most extensively studied known human carcinogens. The series of cohort mortality studies of underground miners in countries throughout the FIGURE 3-6 Predicted lifetime relative risk of lung-cancer for males and females by ''residential" radon concentration. Exposure occurs over a lifetime at a constant radon concentration.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI world is highly informative with respect to the risk of lung-cancer associated with exposure to radon. In each of those studies, miners have been shown to be at excess risk for lung-cancer under past conditions of exposure. The quantitative estimates of exposures experienced by the miners, although subject to error, allow characterization of exposure-response relationships for radon and lung-cancer. These data formed the basis for the development of the committee's risk models. Case-control studies of residential radon exposure and lung-cancer have also been conducted in various countries. Although also informative, the lower exposures of people in these studies and methodologic problems make it very difficult to identify the relationship between residential radon exposure and lung-cancer mortality in an individual study. However, the estimate of lung-cancer risk based on a recent meta-analysis of these 8 studies is in close agreement with the risk predicted on the basis of miner data. The committee was fortunate to have available an update of the data on the 11 miner cohorts previously analyzed by Lubin and others (1994a). The most-recent data were used in developing the committee's risk models. The committee recognized that great care is needed in combining data from different cohorts of underground miners around the world. The levels of exposure to radon and other relevant covariates, such as arsenic and tobacco smoke, differed appreciably among groups of miners. The completeness and quality of the data available on relevant exposures also differed notably among the cohorts. Information on tobacco consumption was available for only 6 of the 11 cohorts; of these 6, only 3 had information on duration and intensity of exposure to tobacco smoke. Lifestyle and genetic factors that influence susceptibility to cancer might also account for heterogeneity among cohorts. Despite those differences, the committee concluded that the best possible estimate of lung-cancer risk associated with radon exposure would be obtained by combining the available information from all 11 cohorts in a judicious manner. The committee used statistical methods for combining data that both allowed for heterogeneity among cohorts and provided an overall summary estimate of the lung-cancer risk. Confidence limits for the overall estimate of risk allow for such heterogeneity. The committee's risk models described the ERR as a simple linear function of cumulative exposure to radon, allowing for differential effects of exposure during the periods 5–14 years, 15–24 years, and 25 years or more before lung-cancer death. The most weight was given to exposures occurring 5–14 years before death from lung-cancer. The committee entertained 2 categorical risk models in which the ERR was modified either by attained age and duration of exposure or by attained age and exposure rate. The ERR decreased with both attained age and exposure rate and increased with duration of exposure. For cumulative exposures below 0.175 Jhm-3 (50 WLM), a constant-relative-risk model without these modifying factors appeared to fit the data as well as the 2 models that allow for effect modification.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI Lung-cancer risks associated with radon exposure were characterized in several ways. The LRR was used to describe the lifetime risk of lung-cancer among people continually exposed to radon throughout the course of their lifetime relative to the risk among unexposed individuals. The percentage of lung-cancer cases that can be attributed to residential exposure to radon is of particular interest for risk management. The committee used data from the National Residential Radon Survey in combination with its 2 categorical risk models to estimate the AR posed by residential radon exposures. The ARs were estimated to be in the range 10–15%. These estimates are somewhat higher than the estimate of about 8% based on the data and methods of BEIR IV. About 30% of the AR was associated with homes having concentrations above 148 Bqm-3 (4 pCiL-1). Although the AR percentages were comparable for ever-smokers and never-smokers under the multiplicative model, the number of radon-related lung-cancer cases was much higher among ever-smokers than for never-smokers under the multiplicative model. Of the approximately 157,000 lung-cancer deaths occurring annually, radon was estimated to play a role in about 15,000 to 22,000 cases. Of these, 13,000 to 19,000 were in ever-smokers and 1,000 to 3,000 in never-smokers, depending on the choice of the model. These computed values represent the best estimates of the lung-cancer risk attributable to radon that can be made at this time. The committee recognized that these estimates are subject to uncertainty, including kinds of uncertainty that are not captured by statistical confidence limits on risk estimates. Consequently, the committee attempted a quantitative analysis of the uncertainty associated with estimates of the population AR. This analysis was itself limited, inasmuch as characterization of such sources of uncertainty as exposure measurement error in the miner data is difficult. Using data whenever possible and expert judgment otherwise, the committee attempted to describe the sources of uncertainty in its 2 categorical risk models. The best estimates of the population AR were in the range 10 to 14% on the basis of the committee's preferred risk models. The quantitative analysis conducted by the committee provided limits within which the AR was considered to lie with 95% certainty. For the exposure-age-concentration model, the uncertainty interval ranged from about 9 to 25%, with central estimates of about 14%. This reflects a substantial degree of uncertainty in the AR, although the uncertainty distributions indicated that values near the central estimates were much more likely than values near the upper and lower limits. For the exposure-age-duration model, the AR ranged from 7 to 17%, and centered at about 10%. The committee also computed uncertainty limits for the simple constant-relative-risk model fitted to the miner data below 0.175 Jhm-3 (50 WLM), which is based on observations in miners closest to residential exposure levels. The latter analysis, which minimizes the degree of extrapolation outside the range of the miner data, led to uncertainty limits of 2–21%, with a central estimate of about 12%.

OCR for page 69
Health Effects of Exposure to Radon: BEIR VI The committee also noted that its quantitative estimates of risks posed by residential radon exposure depend strongly on the assumption of a linear relationship, without a threshold, between low-dose exposure to radon and risk. As reviewed in chapter 2, that assumption is based on our current understanding of the mechanisms of radon-induced lung-cancer, although it is recognized that this understanding is incomplete. The committee did not attempt to quantitatively address uncertainty due to the linear, no-threshold assumption, because specific mechanistically plausible alternative dose-effect relationships were not identified in the committee's review. Despite the uncertainty, the committee concluded that the weight of evidence of the available data supports a finding that residential-radon exposure increases lung-cancer risk. The best estimate of risk that can be obtained at this time is based on the committee's analysis of the combined updated data on the 11 cohorts of underground miners. The committee noted that this estimate is consistent with that derived from a recent meta-analysis of summary relative risks from the 8 residential case-control studies conducted to date. The committee questioned whether further residential studies are likely to clarify the uncertainty surrounding residential-radon lung-cancer risks. The case-control studies conducted to date have been somewhat inconsistent. While most are compatible with the hypothesis of an elevated cancer risk, their results could also be interpreted as compatible with the hypothesis of no increase in risk in light of inherent uncertainties in data. Further, the residential studies offer only very limited information on never-smokers. Clear evidence of an increased lung-cancer risk in miners, many of whom were exposed to levels of radon only about two-fold higher than associated with residence in some homes, played an important role in supporting the committee's conclusions about the likelihood of an increased lung-cancer risk due to residential radon exposures. The committee agreed with the recommendations of workshops conducted by the Department of Energy and the Commission of European Communities: further studies should not be initiated until studies now in progress are completed and the data are pooled from these studies and studies already completed are pooled. The committee examined the effect of reductions in radon levels in U.S. homes on lung-cancer risk, assuming different scenarios of the efficiency of reduction. On the basis of the committee's categorical risk models, reducing radon concentration in all homes that are above 148 Bqm-3 (4 pCiL-1) to below 148 Bqm-3 (4 pCiL-1) is estimated to result in the avoidance of about 3 to 4% of lung-cancers.