United States
Environmental Protection
Agency
Air and Radiation
(6608J)
EPA 402-R-03-003
June 2003
&EPA
EPA Assessment of Risks from
Radon in Homes
-------
EPA ASSESSMENT OF RISKS FROM RADON IN HOMES
June 2003
Office of Radiation and Indoor Air
United States Environmental Protection Agency
Washington, DC 20460
-------
11
-------
PREFACE
Radon has been classified as a known human carcinogen and has been
recognized as a significant health problem by groups such as the Centers for Disease
Control, the American Lung Association, the American Medical Association, and the
American Public Health Association. As such, risks from in-home radon exposure
have been a major concern for the EPA. In 1992, EPA published its Technical Support
Document for the 1992 Citizen's Guide to Radon, which included a description of its
methodology for estimating lung cancer risks in the U.S. associated with exposure to
radon in homes. That methodology was primarily based on reports published by the
National Academy of Sciences (MAS). In one of those reports, known as "BEIR IV"
(MAS 1988), a model was derived for estimating the risks from inhaled radon progeny,
based on an analysis of epidemiologic results on 4 cohorts of occupationally exposed
underground miners. In 1994, the EPA sponsored another study, "BEIR VI", to
incorporate additional information that had become available from miner cohort and
residential studies. In early 1999, the MAS published its "BEIR VI" report (MAS 1999),
which presented new risk models based on information from 11 miner cohorts. A
major conclusion of the BEIR VI report was that radon is the second leading cause of
lung cancer after smoking.
In light of findings and recommendations in BEIR VI, this report presents a
revised risk assessment by EPA's Office of Radiation and Indoor Air (ORIA) for
exposure to radon in homes. In response to a request by ORIA, the Radiation Advisory
Committee (RAC) of the Science Advisory Board (SAB) has reviewed the methodology
used in this report for estimating cancer risks from radon. An initial advisory, finalized in
July, 1999 (SAB 1999), found the methodology to be generally acceptable but included
recommendations for some adjustments. The RAC met again in November, 1999 to
consider ORIA's response to their recommendations. The RAC report (SAB 2000)
concluded that "ORIA has produced a credible risk assessment and has responded well
to the recommendations provided by the RAC in its Advisory." They also offered
additional comments and suggestions. Responses to those comments were provided
in a letter of October 5, 2000 from Robert Perciasepe, Assistant Administrator of the
Office of Air and Radiation.
This report was prepared by EPA staff members David J. Pawel and Jerome S.
Puskin, ORIA, Radiation Protection Division. The authors gratefully acknowledge the
invaluable assistance provided by Christopher B. Nelson, the constructive review
conducted by the RAC, and helpful review comments by Dr. Nancy Chiu and Dr.
William Brattin.
The mailing address for the authors is:
U.S. Environmental Protection Agency
Office of Radiation and Indoor Air (6608J)
Washington, DC 20460
in
-------
ABSTRACT
Background. The U.S. Environmental Protection Agency (EPA) updates its
assessment of health risks from indoor radon, which the National Academy of Sciences
(MAS) has determined to be the second leading cause of lung cancer after cigarette
smoking. This risk assessment is based primarily on results from a recent study of
radon health effects (BEIR VI) by the MAS, with some technical adjustments and
extensions. In BEIR VI, the MAS projected 15,400 or 21,800 excess lung cancer
deaths in the U.S. each year, using two preferred risk models developed from data from
11 cohorts of miners.
Methods. EPA modified and extended the approach used in BEIR VI. First, a single
model is constructed that yields numerical results midway between what would be
obtained using the two BEIR VI preferred models. Second, noting that the BEIR VI
definition of excess risk effectively omits premature deaths caused by radon in people
who would otherwise have eventually died of lung cancer, EPA modifies the BEIR VI
calculations to include all radon-induced lung cancer deaths. Third, EPA uses more
detailed smoking prevalence data and more recent mortality data for its calculations
than was used in BEIR VI. Fourth, whereas BEIR VI estimated the fractional increase
in lung cancers due to radon, EPA also provides numerical estimates of the risk per unit
exposure [lung cancer deaths per working level month (WLM)].
Results. Based on its analysis, EPA estimates that out of a total of 157,400 lung
cancer deaths nationally in 1995, 21,100 (13.4%) were radon related. Among NS, an
estimated 26% were radon related. Estimates of risk per unit exposure are 5.38x10"4
per WLM for the U.S. population; 9.68x10'4/WLM for ever smokers (ES); and 1.67xlO'4
per WLM for never smokers (NS). The estimated risks from lifetime exposure at the
4 pCi/L action level are: 2.3% for the entire population, 4.1% for ES, and 0.73% for NS.
A Monte Carlo uncertainty analysis that accounts for only those factors that can be
quantified without relying too heavily on expert opinion indicates that estimates for the
U.S. population and ES may be accurate to within factors of about 2 or 3.
Conclusions. The effects of radon and cigarette smoking are synergistic, so that
smokers are at higher risk from radon. Consequently, if projected reductions in U.S.
smoking rates hold up, some decrease in radon-induced lung cancers is expected,
concomitant with decreases in lung cancer, generally; nevertheless, it is anticipated that
indoor radon will remain an important public health problem, contributing to thousands
of lung cancer deaths annually.
IV
-------
CONTENTS
Section Page
PREFACE iii
ABSTRACT iv
LIST OF TABLES vii
LIST OF FIGURES ix
EXECUTIVE SUMMARY 1
I. Introduction 5
II. Scientific Background 6
III. Previous Methodology for Calculating Risks 9
IV. BEIR VI Risk Models 11
A. Statistical Fits to the Miner Data 11
B. Extrapolation from Mines to Homes 12
C. Smoking 13
D. Calculation of Attributable Risk and Lung Cancer Deaths 14
V. Residential Studies 16
VI. Methodology for Calculating Radon Risk 17
A. Overview 17
B. Life-Table Derivation of Lifetime Risks of Radon-Induced Lung
Cancer Death 19
1. Lung cancer death rates for male and female ES and NS 19
2. Choice of a relative risk model 22
3. Applying the concentration and duration models 23
4. Averaging the age-specific risks of lung cancer death 31
5. Combined risk estimates for the U.S. population 32
C. Etiologic Fraction 33
D. Risk per Unit Exposure and Unit Concentration 37
E. Age at Cancer Death and Years of Life Lost 41
F. Comparison with Previous Estimates 44
1. Exposure parameters 44
2. Baseline rates 44
3. Mortality data 45
-------
Section Page
4. Relative risk model 45
G. Effects Other than Fatal Lung Cancers 45
H. Current Smokers 46
I. Dependence of Lung Cancer Death Rates on Smoking 48
J. Summary 51
VII. Uncertainties 52
A. Background 52
B. Uncertainties in the Miner Data 53
1. Errors in exposure estimates 53
2. Confounding by other exposures 54
3. Smoking by miners 55
C. Uncertainties in Extrapolating to Residential Exposures 55
1. K-factor 55
2. Dependence of risk on gender 56
3. Dependence of risks on age at exposure 56
4. Smoking patterns in the U.S. population 58
D. Uncertainty in the Estimate of Average Residential Exposure 58
1. Uncertainty in the average radon concentration (C) 58
2. Uncertainty in equilibrium fraction (F) 59
3. Uncertainty in the average occupancy factor (Q ) 59
E. Monte Carlo Simulations 59
F. Uncertainty in Extrapolating to Low Exposure Rates 65
G. Sensitivity Analysis of Risk Estimates to Assumptions about Health
Effects from Exposures to Radon 67
Appendix A: Age-Specific, Ever-Smoking Prevalence Estimates 71
Appendix B: Smoothing the BEIR VI Relative Risk Functions 76
Appendix C: Notation and Formulas 78
Appendix D: Lung Cancer Risks by Radon Level and Smoking Status 82
REFERENCES 83
VI
-------
LIST OF TABLES
Table Title Page
1 Miner cohorts, number exposed, person-years of epidemiologic
follow-up, and lung cancer deaths 8
2 Miner cohorts, radon exposure, and estimates of excess relative risk per
WLM exposure with 95% Cl 9
3 Parameter estimates for BEIR VI models 12
4 Estimated AR for domestic radon exposure using 1985-1989 U.S.
population mortality rates 15
5 Estimated number of lung cancer deaths in the U.S. in 1995 attributable
to indoor residential radon progeny exposure 16
6 Estimated risk per WLM from BEIR VI concentration and duration models ... 23
7 Estimated etiologic fraction by smoking category and gender 33
8 Estimated fraction of lung cancer deaths in 1995 attributable to radon 34
9 Estimated etiologic fraction by smoking category and gender for a stationary
population in which 53% of males and 41% of females are ES 35
10 Estimates of risk per WLM by smoking category and gender for a stationary
population in which 53% of males and 41 % of females are ES 38
11 Estimated average age at lung cancer death 41
12 Estimated years of life lost per lung cancer death 42
13 Dependence of risk estimates on changes in methodology since 1992 45
14 Estimating radon-induced lung cancer deaths for current and former
smokers 47
15 Age-specific and age-adjusted relative risks of fatal lung cancers for current
and former smokers versus never smokers for whites 50
Vll
-------
Table Title Page
16 Sensitivity of risk per WLM estimates to assumptions about the relative risk
of fatal lung cancers for ES compared to NS 51
17 Effects of age at first radon exposure on ERR per WLM 57
18 Parameters for uncertainty distributions for excess relative risks based
on the concentration model 62
19 Monte Carlo simulation of risk per WLM, EF, YLL, average residential
exposure, and number of radon-induced fatal cancers 63
20 Monte Carlo simulation of EF, YLL, average residential exposure, and
number of radon-induced fatal cancers with exposure factors fixed
at nominal values 64
21 Dependence of the risk per WLM and EF estimates on the NS risk
coefficient 68
22 Dependence of the risk per WLM and EF estimates on the childhood risk
coefficient 69
23 Dependence of the estimated risk per WLM and EF estimates on
assumptions on how relative risks fall off with time-since-exposure ... 70
A1 Ever-smoking prevalence estimates for males by age group 73
A2 Ever-smoking prevalence estimates for females by age group 74
A3 Smoothed age-specific ES prevalence estimates for males and females .... 75
B1 Spline smoothed values for cj), from the BEIR VI concentration and duration
models 77
D1 Lifetime risk of lung cancer death by radon level for never smokers,
current smokers, and the general population 82
Vlll
-------
LIST OF FIGURES
Figure Title Page
1 Ever smoking prevalence by age and gender 21
2 BEIR VI (unsealed) concentration model age-specific excess risks from
a 0.181 WLM/y radon exposure 26
3 Smoothed age-specific excess relative risks from a constant radon
exposure at rate 0.181 WLM/y 27
4 Rates of lung cancer death for ES males and females 28
5 Rates of lung cancer death for NS males and females 29
6 Rates of lung cancer deaths for a stationary population in which 53%
of males and 41 % of females are ES 30
7 Etiologic fraction by ES prevalence from a lifetime exposure of 0.181
WLM/y 36
8 Probability of a premature lung cancer death from a lifelong exposure to
radon at 1 pCi/L as a function of ES prevalence 39
9 Risk per WLM as a function of age at exposure 40
10 Density function for years of life lost from a radon-induced death 42
11 Years of life lost per fatal radon-induce lung cancer 43
IX
-------
EXECUTIVE SUMMARY
Radon-222 is a noble gas produced by radioactive decay of radium-226, which is
widely distributed in soils and rocks. Radon-222 decays into a series of short-lived
radioisotopes. These decay products are often referred to as radon progeny or
daughters. Because it is chemically inert, most inhaled radon is rapidly exhaled, but the
inhaled decay products readily deposit in the lung, where they irradiate sensitive cells in
the airways, thereby enhancing the risk of lung cancer.
In 1999, the National Research Council of the National Academy of Sciences
published the BEIR VI report, Health Effects of Exposure to Radon (NAS 1999), which
assessed the risks to the U.S. population from radon in homes. The authors of this
study, sponsored by the EPA, had the benefit of extensive new information not
available to the authors of the Academy's previous BEIR IV report on the risks from
radon and other alpha emitters (NAS 1988). On the basis of epidemiologic evidence
from miners and an understanding of the biologic effects of alpha radiation, the
committee concluded that residential exposure to radon is expected to be a cause of
lung cancer in the population. Based on a statistical analysis of epidemiologic data on
11 cohorts of occupationally exposed underground miners, the committee developed
two preferred risk models from which they projected, respectively, 15,400 or 21,800
excess lung cancer cases in the U.S. each year. An analysis of the uncertainties
suggested a range of 3,000 to 33,000 cases per year. The committee concluded that
"this indicates a public health problem and makes indoor radon the second leading
cause of lung cancer after cigarette smoking."
Both of the preferred BEIR VI models are framed in terms of excess relative risk
(ERR), which represents the fractional increase in lung cancer risk due to a specified
exposure.1 To estimate the risk at any given age from a past exposure, one multiplies
the ERR times the baseline lung cancer rate for an individual of that age (and, if
appropriate, sex or smoking category). The lifetime risk from an arbitrary exposure can
be calculated using a specified risk model in conjunction with life-table methods that
incorporate competing causes of death. In both of these BEIR VI models the ERR falls
off with time-since-exposure and with age at risk; nevertheless, because of the
increasing baseline rate of lung cancer with age, the calculated risk from a given
exposure often increases with increasing age.
An important finding in BEIR VI, based on updated and expanded miner data, is
that risk from a given exposure tends to increase when that exposure is more spread
1 Exposures are measured in units of working level months (WLM), a measure of potential alpha
particle energy that will be released by short-lived radon decay products per liter of air.
1
-------
out over time. For the relatively low exposure rates or long time durations of most
concern for EPA, the risk per unit (WLM) exposure is maximal and increases linearly
with radon exposure.
Another new finding is that the estimated ERR is about twice as high for never
smokers (NS) as for ever smokers (ES). Estimates indicate that radon exposure
accounts for about 1 in 8 ES lung cancer deaths and 1 in 4 NS lung cancer deaths.
However, since ES have a much higher baseline lung cancer rate than NS, the risk of a
radon-induced lung cancer, on an absolute scale, is still much higher than for NS.
Although there is a growing body of data from epidemiological (case-control)
studies showing a correlation between lung cancer and radon exposures in homes,
these results do not conclusively demonstrate an excess risk in homes with elevated
radon and are inadequate as a basis for quantitative risk estimation. Thus, estimates of
risk for indoor exposures must still be extrapolated using models derived from the miner
data. There are a number of important differences between mine and indoor exposures
that must be considered in making this extrapolation.
First, due to physical and physiological factors, the alpha particle dose to target
cells in the lung per WLM could be higher or lower in the case of residential exposures
than for mine exposures. Since the risk is presumed to be proportional to dose, a
model derived from the miner data might need to be adjusted to account for these
differences. The BEIR VI risk estimates were based on the premise that the effects of
these differences approximately counterbalanced each other in such a way that no
adjustment was warranted. Doubts about this premise were expressed by Cavallo
(2000). Cavallo correctly noted inconsistencies in portions of BEIR VI relating to how
doses from exposures in mines and homes compare, and suggested that as a result
the BEIR VI report may have overstated risks from residential exposures. More
recently, James et al. (2003) submitted a report which carefully reexamined issues
raised by Cavallo. James et al. reaffirmed that the effects on doses of differences
between homes and mines do approximately counterbalance each other so that no
adjustment would be needed for in-home risk calculations. It follows that the
inconsistencies in BEIR VI noted by Cavallo did not lead to an overestimate of the risks
from radon.
Second, other agents in the atmosphere of underground mines, such as arsenic,
silica, and diesel fumes, could modify the lung cancer risk associated with exposure to
radon progeny. BEIR VI cited evidence that the latter two types of exposures were
probably not strong modifiers of risk but that arsenic might be a source of positive bias
in the risk estimates.
Third, the exposure rates in homes are generally lower than the lowest levels for
which we have clear evidence of excess risk in mines. Consequently, assessment of
indoor radon risks requires an extrapolation to lower exposure rates. Although the
miner data and radiobiological data are both suggestive of a constant risk per unit
-------
exposure as one extrapolates downward from the lowest miner exposures, this
assumption has been questioned. An ecological study has indicated that lung cancer
rates are negatively correlated with average radon concentrations across U.S. counties
(Cohen 1995), suggesting that the risks from very low levels of radon have been
overestimated, or that such exposure levels might even protective against lung cancer.
Biologically based models have also been proposed that could project substantially
reduced carcinogenicity at low doses (for example, Moolgavkar and Luebeck 1990,
Elkind 1994). Numerous critics, including the BEIR VI committee, have discounted the
ecological study results because of methodological limitations, and the biologically
based models remain highly speculative. The BEIR VI committee adopted the linear
no-threshold assumption based on our current understanding of the mechanisms of
radon-induced lung cancer, but recognized that this understanding is incomplete and
that therefore the evidence for this assumption is not conclusive.
In this document EPA updates its assessment of the health risks from indoor
radon, based primarily on the BEIR VI report, with some technical adjustments and
extensions. First, EPA constructs a single model that yields numerical results midway
between what would be obtained using the two BEIR VI preferred models. Second,
noting that the BEIR VI definition of excess risk effectively omits premature deaths
caused by radon in people who would otherwise have eventually died of lung cancer,
EPA modifies the BEIR VI calculations so as to include all radon-induced lung cancer
deaths. Third, whereas the BEIR VI committee assumed that a fixed percentage of
adult males or females were ES, EPA uses age-specific smoking prevalence data.
Fourth, whereas BEIR VI estimated the fractional increase in lung cancers due to
radon, EPA also provides numerical estimates of the risk per unit exposure [lung cancer
deaths per working level month (WLM)] and the number of years of life lost per cancer
death.
Based on its analysis, EPA estimates that out of a total of 157,400 lung cancer
deaths nationally in 1995, 21,100 (13.4%) were radon related. Although it is not
feasible to totally eliminate radon from the air, it is estimated that about one-fourth of the
radon-related lung cancers could be averted by reducing radon concentrations in
homes that exceed EPA's recommended 4 picocurie per liter (pCi/L) action level (MAS
1999).
It is estimated that 86% of the radon-related lung cancer deaths were in ES,
compared to 93% for all lung cancer deaths. The projected average years of life lost
are higher for the radon-related cases (17 y) than for lung cancer deaths generally
(12 y). Estimates of risk per unit exposure are as follows: 5.38xlO"4/WLM (all);
9.68x1 Q-4/WLM (ES); and 1.67xlO-4/WLM (NS). Based on an assumed average
equilibrium fraction of 40% between radon and its decay products and an indoor
occupancy of 70%, the estimated risks from lifetime exposure at the 4 pCi/L action level
are: 2.3% (all), 4.1% (ES), and 0.73% (NS). Although estimated absolute risks are
much higher for ES than NS, estimated relative risks are higher for NS. It is estimated
that among NS about one-quarter (26%) of lung cancers are due to radon compared to
-------
about one-eighth (12%) among ES. It was more difficult to estimate risks for current
smokers. Because of limitations of the data from the miner cohorts, the BEIR VI
models did not specify excess relative risks for current smokers. Estimates of risk for
current smokers (calculated by presuming that they start smoking at age 18 y and do
not quit) are 1.5xlO"3 per WLM, or over 6% for a lifetime exposure at 4 pCi/L.
EPA also reexamines the issue of uncertainty in the risk estimates. Emphasizing
the uncertainty in extrapolating risk estimates from observations on miners exposed to
higher levels of radon than are ordinarily found in homes, BEIR VI derived its preferred
uncertainty bounds (95% confidence limits 3,300 to 32,600) using a constant relative
risk model obtained by a statistical fit to a restricted set of data on miners exposed to
less than 50 WLM — levels that are comparable to lifetime residential exposures. The
sampling errors are large with this limited data base; as a consequence the resulting
confidence range may be overly broad. EPA adopts an alternative approach, deriving
its estimates of uncertainty using the BEIR VI preferred models, with some explicit
consideration of model uncertainties. However, like BEIR VI, EPA was unable to
quantify all the relevant sources of uncertainty. These uncertainties are discussed
qualitatively (orsemi-quantitatively) and, for perspective, results of sensitivity analyses
for some of these variables are included. From a Monte Carlo analysis of those
uncertainties that could be quantified, EPA estimates a 90% subjective confidence
interval of 2 to 12 xio~4 lung cancer deaths per WLM, for the general population. The
corresponding 90% interval for radon-induced lung cancer cases in 1995 is 8,000 to
45,000. Since the interval would be wider if additional sources of uncertainty had been
accounted for in the analysis, it is plausible that the number of radon-induced deaths is
smaller than 8,000 (but unlikely that it would be as small as 3,300). However, given the
predominant role smoking is known to play in the causation of lung cancer, it is unlikely
that radon accounts for as many as 45,000 deaths or 12 xio~4 lung cancer deaths per
WLM. Risk estimates for exposures to specific subgroups, especially children, NS and
former smokers, have a higher degree of uncertainty than estimates for the general
population.
The effects of radon and cigarette smoking are synergistic, so that smokers are
at higher risk from radon. Consequently, if projected reductions in U.S. smoking rates
hold up, some decrease in radon-induced lung cancers is expected, concomitant with
decreases in lung cancer, generally; nevertheless, it is anticipated that indoor radon will
remain an important public health problem, contributing to thousands of lung cancer
deaths annually.
-------
I. Introduction
In 1992, EPA published its Technical Support Document for the 1992 Citizen's
Guide to Radon, which included a description of its methodology for estimating lung
cancer risks in the U.S. associated with exposure to radon in homes. That
methodology was primarily based on two reports published by the National Academy of
Sciences (MAS), referred to here as "BEIR IV" (MAS 1988) and the "Comparative
Dosimetry Report" (MAS 1991). In BEIR IV, a model was derived for estimating the
risks from inhaled radon progeny, based on an analysis of epidemiologic results on 4
cohorts of occupationally exposed underground miners. In the Comparative Dosimetry
Report, estimates of radiation dose to potential target cells in the lung were calculated
under mine and residential conditions, respectively. Results were expressed in terms of
a ratio, K, representing the quotient of the dose of alpha energy per unit exposure to an
individual in a home compared to that for a miner in a mine. It was concluded that the
dose per unit exposure was typically about 30% lower in homes than in mines (K-0.7),
implying a 30% reduction in the risk coefficient applicable to home environments from
what would be estimated from miner data.
Subsequently, EPA sponsored another MAS study (BEIR VI), which provided
new risk models and estimates of the K-factor, based on much more complete
information (MAS 1999). Data on 11 miner cohorts were now available, including
further follow-up of the 4 cohorts upon which the BEIR IV model was based. In
addition, some new information had become available regarding exposure conditions in
mines and homes that led to a revised estimate of K. In response to questions raised
about issues relating to the K-factor in BEIR VI (Cavallo 2000), the EPA sponsored a
study in which it was concluded that, under the exposure assumptions employed in
BEIR VI, the value used for the K- factor was appropriate (James et al. 2003).
EPA is now revising its assessment of risks from indoor radon in light of the
findings and recommendations in BEIR VI. The revised methodology includes some
extensions and modifications from the approach in BEIR VI. These extensions and
modifications were made after an advisory review from the Agency's Radiation Advisory
Committee (RAC). Taken together, these adjustments have only a minor impact on the
estimated number of radon induced lung cancers occurring each year.
This document will serve as a technical basis for EPA's estimates of risk from
radon in homes. It provides estimates of the risk per unit exposure and projects the
number of fatal lung cancers occurring in the U.S. population each year due to radon. It
also provides separate estimates for males and females, and for ever- and never-
smokers. Finally, it discusses the uncertainties in these estimates. It is anticipated that
the methodology and results presented here will be used in developing guidance for the
members of the public in addressing elevated radon levels in their homes. These
results may also be used for regulatory purposes: e.g., to set cleanup levels for radium
in soil or to set maximum concentration levels for radon in drinking water.
-------
II. Scientific Background
Radon-222 is a noble gas produced by the radioactive decay of radium-226,
which is widely distributed in uranium-containing soils and rocks. The radon readily
escapes from the soil or rock where it is generated and enters surrounding water or air.
The most important pathway for human exposure is through the permeation of
underlying soil gas into buildings, although indoor radon can also come from water,
outside air, or building materials containing radium. Radon-222 decays with a half-life
of 3.82 days into a series of short-lived radioisotopes collectively referred to as radon
daughters or progeny. Since it is chemically inert, most inhaled radon-222 is rapidly
exhaled, whereas inhaled progeny readily deposit in the airways of the lung. Two of
these daughters, polonium-218 and polonium-214, emit alpha-particles. When this
happens in the lung, the radiation can damage the cells lining the airways, leading
ultimately to cancer. (Nuclear decay of radon decay products also releases energy in
the form of beta particles and high energy photons, but the biological damage resulting
from these emissions is believed to be small compared to that from alpha particles.)
Two other radon isotopes - radon-219 (actinon), and radon-220 (thoron) - occur
in nature and produce radioactive radon daughters. Because of its very short half-life
(3.9 s), environmental concentrations of actinon and its daughters are extremely low, so
their contribution to human exposure is negligible. The half-life of thoron is also
relatively short (56 s), and a lower fraction of released alpha-particle energy is absorbed
within target cells in the bronchial epithelium than in the case of radon-222. As a result,
thoron is thought to pose less of a problem than radon-222, but we have rather limited
information on human exposure to thoron, and no direct information on its
carcinogenicity in humans . For the remainder of this document, we shall focus only on
radon-222 and its daughters. Following common usage, the term radon will in some
cases refer simply to radon-222, but sometimes to radon-222 plus its progeny. For
example, one often talks about "radon risk" when most of that risk is actually conferred
by inhaled decay products.
Radon concentrations in air are commonly expressed in picocuries per liter
(pCi/L) in the U.S., but in western Europe, they are given in SI units of bequerels per
cubic meter (Bq/m3), where a Bq is 1 nuclear disintegration per second. By definition, 1
picocurie is equal to 0.037 Bq; hence, 1 pCi/L corresponds to 37 Bq/m3.
Radon progeny concentrations are commonly expressed in working levels (WL).
One WL is defined as any combination of short-lived radon daughters in 1 liter of air
that results in the ultimate release of 1.3xl05 million electron volts of alpha energy. If a
closed volume is constantly supplied with radon, the concentration of short-lived
daughters will increase until an equilibrium is reached where the rate of decay of each
daughter will equal that of the radon itself. Under these conditions each pCi/L of radon
will give rise to (almost precisely) 0.01 WL. Ordinarily these conditions do not hold: in
homes, the equilibrium fraction is typically 40%; i.e., there will be 0.004 WL of progeny
-------
for each pCi/L of radon in air (MAS 1999).
Cumulative radon daughter exposures are measured in working level months
(WLM), a unit devised originally for occupational applications. Exposure is proportional
to concentration (WL) and time, with exposure to 1 WL for 170 h being defined as
1 WLM. To convert from residential exposures expressed in pCi/L, the BEIR VI
committee assumed that the fraction of time spent indoors is 70%. It follows that an
indoor radon concentration of 1 pCi/L would on average result in an exposure of
0.144 WLM/y = (1 pCi/L) [(0.7)(0.004) WL/(pCi/L)] (51.6 WLM/WL-y).
There is overwhelming evidence that exposure to radon and its decay products
can lead to lung cancer. Since the 1500s, it has been recognized that underground
miners in the Erz mountains of eastern Europe are susceptible to high mortality from
respiratory disease. In the late 1800s and early 1900s, it was shown that these deaths
were due to lung cancer. The finding of high levels of radon in these mines led to the
hypothesis that it was responsible for inducing cancer. This conclusion has been
confirmed by numerous studies of radon-exposed underground miners and laboratory
animals.
The most important information concerning the health risks from radon comes
from epidemiological studies of underground miners. In these "cohort" studies, lung
cancer mortality is monitored overtime in a group of miners and correlated with the
miners' estimated past radon exposure. The BEIR VI committee analyzed results from
11 separate miner cohorts, each of which shows a statistically significant elevation in
lung cancer mortality with increasing radon exposure. Summary information on the
epidemiologic follow-up of the 11 cohorts is provided in Table 1.
Table 2 summarizes information on the miners' exposure and the excess relative
risk (ERR) per unit exposure in each cohort. The ERR represents the multiplicative
increment to the excess lung cancer mortality beyond background resulting from the
exposure. From Table 2 it is clear that there is heterogeneity in the estimates of the
ERR per unit exposure derived from the various studies. Some of this heterogeneity is
attributable to random error, and some to exposure rate or age and temporal
parameters discussed below. There is, however, unexplained residual heterogeneity,
possibly due to systematic errors in exposure ascertainment, unaccounted for
differences in the study populations (genetic, lifestyle, etc.), or confounding mine
exposures.
-------
Table 1: Miner cohorts, number exposed, person-years of epidemiologic
follow-up,and lung cancer deaths (MAS 1999).
Study
China
Czechoslovakia
Colorado Plateau3
Ontario
Newfoundland
Sweden
New Mexico
Beaverlodge (Canada)
Port Radium (Canada)
Radium Hill (Australia)
France
Total"
Type
of
Mine
Tin
Uranium
Uranium
Uranium
Fluorspar
Iron
Uranium
Uranium
Uranium
Uranium
Uranium
Number
of
Workers
13,649
4,320
3,347
21,346
1,751
1,294
3,457
6,895
1,420
1,457
1,769
60,606
Number
of person-
years
134,842
102,650
79,536
300,608
33,795
32,452
46,800
67,080
31,454
24,138
39,172
888,906
Number
of lung
cancers
936
701
334
285
112
79
68
56
39
31
45
2,674
3 Exposure limited to <3,200 WLM.
"Totals adjusted for miners and lung cancers included under both Colorado and New Mexico
studies.
-------
Table 2: Miner cohorts, radon exposure, and estimates of excess relative risk
per WLM exposure with 95% Cl (MAS 1999).
Study
China
Czechoslovakia
Colorado Plateau
Ontario
Newfoundland
Sweden
New Mexico
Beaverlodge
Port Radium
Radium Hill
France
Total
Mean
WLMa
286.0
196.8
578.6
31.0
388.4
80.6
110.9
21.2
243.0
7.6
59.4
164.4
Mean
duration
(y)
12.9
6.7
3.9
3.0
4.8
18.2
5.6
1.7
1.2
1.1
7.2
5.7
Mean
WLa
1.7
2.8
11.7
0.9
4.9
0.4
1.6
1.3
14.9
0.7
0.8
2.9
ERR/WLM
%
0.16(0.1-0.2)
0.34 (0.2-0.6)
0.42 (0.3-0.7)
0.89(0.5-1.5)
0.76(0.4-1.3)
0.95(0.1-4.1)
1.72(0.6-6.7)
2.21 (0.9-5.6)
0.19(0.1-1.6)
5.06(1.0-12.2)
0.36(0.0-1.2)
Weighted by person-years; includes 5-year lag period.
III. Previous Methodology for Calculating Risks
EPA's previous methodology for calculating the risks from indoor radon
exposures was described in the Technical Support Document for the 1992 Citizen's
Guide to Radon (EPA 1992). That methodology made use of the risk model derived in
the 1988 National Academy of Sciences' BEIR IV Report, based on a statistical
analysis of results from four epidemiologic studies of radon-exposed underground
miners (NAS 1990). The preferred model in the BEIR IV Report expresses the excess
relative risk (ERR) of lung cancer death at age a, as a function of past exposure:
ERR(a) = 0.025 y(a) (W, + 1/2 WJ (1 )
where \(a) is an age-specific adjustment to the relative risk coefficient, as follows:
-------
Y(a) = 1.2 when a < 55 y
= 1.0 whe n 55 y < a < 65 y
= 0.4 when a > 65 y
W1 is the cumulative exposure received 5-15 y before age a, and W2 is the cumulative
exposure up to age a-15. Thus, the model incorporates a fall-off in the ERR with age
at expression and, independently, with time-since-exposure.
In extrapolating risk estimates from mine to home exposures, EPA, MAS and
others have assumed that the risk is proportional to the dose to target cells lining the
airways of the lung. Thus, in order to estimate risk from home exposures, the right-
hand side of Equation 1 is multiplied by a factor K, which is equal to the ratio of the
dose per WLM exposure in homes relative to mines. Numerous parameters affect
estimates of the dose per WLM and, therefore, K. These include breathing rates,
location of target cells in the lung, mucus thickness and mucocilliary clearance rates,
the size distribution of aerosol particles to which radon decay products are attached, the
relative concentrations of radon decay products, and the proportion of decay products
existing as an unattached (ultrafine) fraction. The BEIR IV committee concluded that K
was reasonably close to 1 and recommended that Equation 1 be applied for the case of
residential exposures. A subsequent MAS committee examined this issue in greater
depth and determined that a best estimate for K was about 0.7 (MAS 1991).
Accordingly, EPA adopted the following risk model for residential exposures (EPA
1992):
ERR(a) = 0.0175 y(a) (W1 + V* WJ (2)
The risk of a radon-induced lung cancer death at age a was then calculated as
the product of ERR(a) times the baseline lung cancer mortality rate at age a. With the
aid of life-table techniques (EPA 1992), the average risk to a member of the 1989-91
life-table population was found to be approximately 2.24x10"4 per WLM. Using this
value in conjunction with an estimated annual average exposure in the U.S. of 0.242
WLM/y, the number of radon-induced lung cancer deaths each year in a population of
250 million was estimated to be 13,600. In that report, EPA employed a correction that
subtracted off the estimated radon-induced lung cancer deaths occurring at each age
from the reported lung cancer mortality. This "baseline correction" had the effect of
reducing the population risk estimate by about 10%.
Consistent with the limited evidence available at the time of the BEIR IV Report's
publication, the model assumed a multiplicative interaction between smoking and radon
exposure; consequently, the ERR was independent of smoking status. Also, while
there was some indication of an increased risk at low exposure rates and longer
exposure durations in the Colorado Plateau miners, these effects were not consistent
across the four cohorts analyzed. As a result, the BEIR IV committee assumed that the
10
-------
risk was not explicitly dependent on exposure rate or duration.
Soon after publication of BEIR IV, the International Council for Radiological
Protection (ICRP) published ICRP Report 65 (ICRP 1993), which relied on essentially
the same data as in BEIR IV. ICRP's risk projection model was also a relative risk
model that depended both on time-since-exposure and age at exposure, but not
exposure rate or duration.
IV. BEIR VI Risk Models
A. Statistical Fits to the Miner Data
In 1998, the MAS published a new report, BEIR VI, that updated the findings on
radon risk presented in BEIR IV. Two preferred models were developed by the BEIR VI
committee based on a combined statistical analysis of results from the latest
epidemiologic follow-up of 11 cohorts of underground miners, which, in all, included
about 2,700 lung cancers among 68,000 miners, representing nearly 1.2 million person-
years of observations. Both preferred BEIR VI models, like the preferred model in BEIR
IV, incorporate a 5-y minimum latency period and a fall-off in the ERR with age at
expression and time-since-exposure, but the BEIR VI models provide a more detailed
break-down of the risk for ages over 65 y and times since exposure greater than 15 y.
Unlike what was found with the more limited BEIR IV and ICRP analyses, the
BEIR VI committee was able to conclude that the ERR per WLM increased with
decreasing exposure rate or with increasing exposure duration (holding cumulative
exposure constant). To account for this "inverse dose rate" effect, the committee
introduced a parameter dependent on the radon-daughter working level (WL)
concentration or, alternatively, the duration of exposure. Respectively, this gave rise to
the two alternative preferred models - the "exposure-age-concentration model" and the
"exposure-age-duration model." For brevity, these will generally be referred to here as
the "concentration" and "duration" models.
Mathematically, the ERR in the two models can be represented as:
ERR = P (w5_14 + Q15_24 w15_24 + 625+ w25J4)age Yz (3)
where: p is the exposure-response parameter (risk coefficient); the exposure windows,
w5_14, w15_24 and w25+, define the exposures incurred 5-14 y, 15-24 y and >25 y before
the current age; and Q15.24 and 625+ represent the relative contributions to risk from
exposures 15-24 y and >25 y before the attained age. The parameters 4)age and yz
define effect-modification factors representing, respectively, multiple categories of
attained age (4)age) and of either exposure rate or exposure duration (yz). The values for
these parameters are summarized in Table 3.
11
-------
Table 3: Parameter estimates for BEIR VI models (MAS 1999).
Duration Model Concentration Model
pxlOO 0.55 pxlOO 7.68
Time-since-exposure
6,5.24 0.72 Q15_24 0.78
625+ 0.44 625+ 0.51
Attained age
4><55 1-00
4>55-64 0.52
4>65-74 0.28
4>75+ 0.13
Duration of exposure
Y<5 LOO
Ys-M 2.78
Y«-24 4.42
Y25-34 6-62
Y35+ 10.2
4><55
4>55-64
4>65-74
4^75+
Exposure
Y<0.5
\0.5-1
YM
Y3-5
Ys-«
Y,5+
1.00
0.57
0.29
0.09
rate (WL)
1.00
0.49
0.37
0.32
0.17
0.11
B. Extrapolation from Mines to Homes
The analysis of the miner studies provides models for estimating the risk per unit
exposure, as a function of age-at-expression, time-since-exposure, and exposure rate
or duration. However, exposure conditions in homes differ from those in mines, with
respect to both the physical properties of the inhaled radon decay products and the
breathing patterns in the two environments. Using the terminology employed in the
MAS "BEIR IV" and "Comparative Dosimetry" reports (MAS 1988, 1991), the risk per
unit exposure in homes, (Risk)h /(WLM)h, can be related to that in mines,
(Risk)m /(WLM)m, by a dimensionless factor, K,
12
-------
(Risk)h /(WLM)h
K=
(Risk)m /(WLM)m
In extrapolating from mine to residential conditions, it is assumed that the risk is
proportional to the alpha particle dose delivered to sensitive target cells in the bronchial
epithelium. Then, Kcan be written as the ratio of the estimated doses per unit
exposure in the two environments:
(Dose)h /(WLM)h
K=
(Dose)m /(WLM)m
Previously, the MAS estimated that the dose from residential exposures was
typically 30% lower than from an equal WLM exposure in mines (MAS 1991); hence,
EPA applied a K-factor of 0.7 in calculating the risk in homes based on the models
derived from miner studies (EPA 1992).
In BEIR VI the MAS derived a revised estimate of K equal to 1. The most
important changes in assumptions from the previous report was a reduction in the
breathing rate for miners and an increase in the size of particles associated with mine
exposures. However, in BEIR VI, the K-factor was defined in terms of radon gas rather
than radon daughter exposure (MAS 1999, Appendix B). This value appeared to have
been misapplied in projecting risk from radon exposure in homes (Cavallo 2000).
Under the sponsorship of EPA, James has reexamined the issue and concluded that,
under the exposure assumptions employed in BEIR VI, a "best estimate" of K- as
properly defined by the equation above - is in fact approximately 1 (James et al. 2003).
Hence, the risk projections made for residential exposures in BEIR VI do not require
modification (James et al. 2003, Krewski et al. 2002). Nominal estimates of risk for
residential exposures in this report are therefore also calculated using a value of K
equal to 1.
C. Smoking
The BEIR VI committee had smoking information on five of the miner cohorts,
from which it concluded that there was a submultiplicative interaction between radon
and smoking in causing lung cancer. That is, the ERR per WLM was higher for never
smokers2 (NS) than for ever smokers (ES), although the absolute risk per WLM was still
much higher in the latter, given their much higher rate of lung cancer. The data on
never-smoking miners are rather limited, and there is considerable uncertainty in the
2Never smokers are defined as those persons who had not yet smoked 100
cigarettes; ever smokers include all those who had smoked 100 cigarettes or more.
13
-------
magnitude of the risk among this group. As a best estimate, the BEIR VI committee
determined that the NS should be assigned a relative risk coefficient (p) twice that for
the general population, in each of the two models defined above. For consistency, the
value of p for ES in the respective models was adjusted downward by a factor of 0.9
from that for the general population.
D. Calculation of Attributable Risk and Lung Cancer Deaths
The two MAS preferred models described above can be used to estimate lung
cancer risks in any population for which radon exposure rates and vital statistics can be
specified. As will be seen in a Section VI.C., the fraction of lung cancer deaths due to
radon — referred to in BEIR VI as the attributable risk (AR) — is only weakly dependent
on lung cancer rates in the population. The BEIR VI committee chose to focus primarily
on AR calculations. Unlike BEIR IV, the BEIR VI report contains no estimate of the
lifetime risk per WLM, which would be a strong function of the lung cancer rate in the
population.
The BEIR VI committee first calculated AR for sub-populations of male and
female ES and NS. For this calculation, they presumed a steady state population
governed by 1985-1989 mortality rates and an average annual exposure of 0.181
WLM/y. The exposure estimate was based on: (1) an average residential radon level of
1.25 pCi/L derived from EPA's National Residential Radon Survey (Marcinowski et al.
1994); (2) an estimated average equilibrium fraction (F) of 40%; and (3) an assumed
70% occupancy factor (Q), representing the estimated fraction of time spent indoors at
home by the population. The age-specific mortality rates for ES and NS were modified
from those for the general population to account for the higher lung cancer mortality in
ES. For males, the age-specific lung cancer rate for ES was taken to be 14 times that
for NS; for females, the ratio was assumed estimated to be 12. It was further estimated
that, among adults, 58% of all males and 42% of all females are ES (independent of
age).
The attributable risks estimated in this way by the BEIR VI committee are given
in Table 4.
14
-------
Table 4: Estimated AR for domestic radon exposure using 1985-1989
U.S. population mortality rates (MAS 1999).
Model
Concentration
Duration
Concentration
Duration
Population
Males
0.141
0.099
Females
0.153
0.108
ES
0.125
0.087
0.137
0.096
NS
0.258
0.189
0.269
0.197
Assuming that 95% and 90% of all lung cancers in males and females,
respectively, occur in ES and that the attributable risks are applicable to the 1995 U.S.
population, radon-attributable lung cancer deaths were estimated for that year by the
MAS. The results are given in Table 5.
15
-------
Table 5: Estimated number of lung cancer deaths in the U.S. in 1995 attributable
to indoor residential radon progeny exposure (MAS 1999).
Radon-Attributable Lung Cancer Deaths
Smoking Status
Total
Ever smokers
Never smokers
Total
Ever smokers
Never smokers
Lung Cancer
Deaths
95,400
90,600
4,800
62,000
55,800
6,200
Concentration
Model
Males
12,500
11,300
1,200
Females
9,300
8,300
1,700
Duration Model
8,800
7,900
900
6,600
5,400
1,200
Males and Females
Total
Ever smokers
Never smokers
157,400
146,400
11,000
21,800
18,900
2,900
15,400
13,300
2,100
V. Residential Studies
Two types of epidemiologic studies of the association between lung cancer and
radon exposure in homes have been performed and are reviewed in BEIR VI: ecologic
and case-control. In the former, variations in average radon levels between geographic
areas are correlated with corresponding variations in lung cancer rates. In the latter,
measured radon levels in the homes of lung cancer cases are compared with those of
control subjects who do not have the disease.
The most extensive ecologic study has been carried out by Cohen, who collected
a large data base of short-term radon measurements in residences across the U.S.
(Cohen 1990, 1995). Grouping the data by county, Cohen found a negative correlation
between average radon level and age-adjusted lung cancer rate. This has led some to
conclude that radon, at typical indoor levels, presents no risk for lung cancer.
16
-------
A number of criticisms have arisen regarding this use of an ecologic study (MAS
1999). Aside from the biological implausibility of the results and the apparent
disagreement with the results from miner cohort studies and residential case-control
studies (see below), the most serious of these revolve around the question of possible
confounding with smoking, which contributes to a very high percentage of lung cancer
cases. In particular, if radon levels were inversely correlated with smoking across
counties, it is easy to see that one can have a spurious inverse correlation between
average radon level and lung cancer rate. A more subtle bias can arise from the
synergism between radon and smoking in causing lung cancer if smoking and radon
levels are correlated within counties (Greenland and Robins 1994, Lubin 1998). Cohen
has argued that the likely magnitude of these kinds of biases is too small to explain his
negative correlation, and the controversy continues (Smith et al. 1998, Cohen 1998,
Cohen 1998a, Lubin 1998a, Field et al. 1998, Goldsmith 1999). The BEIR VI committee
sided with the critics and concluded that Cohen's inverse correlation "was considered to
have resulted from inherent limitations of the ecologic method" and "was considered to
be an inappropriate basis for concluding that indoor radon is not a potential cause of
lung cancer." Most recently, Puskin (2003) found that Cohen's radon levels have
quantitatively similar, strongly negative correlations with cancer rates for cancers
strongly linked to cigarette smoking, weaker negative correlations for certain cancers
weakly dependent on smoking, and no such correlation for cancers not linked to
smoking. These results support the hypothesis that the negative trend reported by
Cohen for lung cancer can be largely accounted for by a negative correlation between
smoking and radon levels across counties.
Numerous case-control studies of radon exposure and lung cancer were begun in
recent years, and most are now either completed or nearing completion. A meta-
analysis of eight published case-control studies showed an enhanced risk for lung
cancer associated with elevated radon exposure, but the enhancement was barely
statistically significant (Lubin and Boice 1996, MAS 1999). The lack of significance is not
surprising in view of the limited statistical power achievable at the modestly elevated
radon levels generally found in homes. Indeed the observed excess risk is very close to
what is expected based on the miner data; moreover, the results deviate significantly
from a projection based on the ecologic data discussed above (MAS 1999). Additional
results from case-control studies have been reported subsequent to the BEIR VI
analysis that provide further support for an increase in lung cancer risk due to radon
exposure in homes (Lubin 1999).
VI. Methodology for Calculating Radon Risk
A. Overview
Described here is the newly developed EPA method for calculating lifetime radon-
related risk estimates based on the findings of BEIR VI. These include estimates of the
etiologic fraction (radon-induced fraction of lung cancer deaths), the lifetime risk per
17
-------
WLM (probablility of a radon-induced cancer death), years of life lost (YLL) per (radon-
induced) cancer death, and numbers of radon-induced cancer deaths per year. The
BEIR VI committee provided estimates of numbers of excess lung cancer deaths and
the excess fraction of lung cancer deaths due to radon exposure, but did not provide
estimates of risk per WLM or YLL per cancer death. Their estimates were based on
two different models for relative risk: "the concentration model" and "the duration model,"
as described in Section IV.A. The concentration model risk estimates were about 40%
higher than the duration model estimates. As discussed below in Section B, EPA is
basing its estimates on a scaled version of the BEIR VI concentration model. The
scaling results in estimated numbers of lung cancer deaths intermediate between the
BEIR VI concentration and duration model estimates. Other refinements and
extensions to the BEIR VI analysis to meet EPA's needs include:
1) The BEIR VI committee used life-table methods to calculate their risk estimates
for NS and ES. These estimates were based on the assumption that 58% of adult
males and 42% of adult females are ES, regardless of age. EPA uses age-specific
smoking prevalence data (DHHS 1997) shown in Appendix A.
2) The BEIR VI committee calculated the "excess risk" or the increase in the
probability of dying from a lung cancer. EPA uses an etiologic definition of radon-related
risk: the probability of dying prematurely from a radon-induced lung cancer. The
difference is that the BEIR VI method omits that proportion of radon-related lung cancer
mortality occurring in individuals who would have died later from lung cancer in the
absence of radon exposure. BEIR VI presents estimates of "attributable risk," by which
was meant the difference between the lung cancer mortality in an exposed and
unexposed population, divided by the mortality in the exposed population; in contrast,
EPA here presents estimates of the "etiologic fraction" (EF), which represents the
fraction of lung cancer deaths in the exposed population in which radon played some
causative role.
3) EPA adds to the discussion found in BEIR VI on how changes in smoking
patterns might impact estimates of risk. It will be shown that estimates of EF are much
less sensitive to changes in smoking prevalence than are estimates of risk per WLM.
Section B details life-table methods for deriving lifetime risks. We present
results for EF, risk per WLM, and YLL per cancer death in Sections C through E.
Section F compares current estimates to the previous EPA estimates. Section G
discusses health risks other than lung cancer mortality. Section H considers the
problem of estimating radon-induced lung cancer deaths among current smokers.
Section I offers a discussion on estimation problems related to smoking. A very short
summary is given in Section J.
18
-------
B. Life-Table Derivation of Lifetime Risks of Radon-Induced Lung Cancer Death
Lifetime risk estimates such as risk per WLM can be derived using a life-table
method. Life-table methods account for the effects of competing causes of death, which
is necessary because the probability of dying from a radon-induced lung cancer
depends on the age-specific rates of death from all causes as well as lung cancer death
rates. The death rates from lung cancer and from all causes are determined from U.S.
vital statistics. The risk per WLM and EF estimates are calculated assuming stationary
populations for male ES, female ES, male NS, and female NS. This results in risk
estimates for each of these four stationary populations.
Calculating risk per WLM is essentially a four-step process. First, the age-specific
(baseline) lung cancer death rates are determined for each of the four stationary
populations. As described in detail below, the rates are derived from the vital statistics
on lung cancer death, recent data on ever-smoking prevalence, and by assuming that
the ES age-specific lung cancer rates are 14 (males) or 12 (females) times higher than
the rates for NS. Second, a model for age-specific relative risks is chosen and applied
to the baseline rates to determine the age-specific lung cancer risk due to a constant,
lifelong radon exposure. The third step is to calculate a weighted-average of these age-
specific risks using weights equal to the probability of survival (to each age). This step is
used to yield separate risk per WLM estimates for the four gender- and smoking-specific
populations. The final step combines these estimates to obtain the risk per WLM for the
entire U.S. population. Details on each of these steps follow.
1. Lung cancer death rates for male and female ES and NS: Baseline lung
cancer death rates for the general population are derived from 1989-91 vital statistics
(NCHS 1992, 1993a, 1993b). To obtain the lung cancer death rates for ES and NS, we
assume, as in BEIR VI, that the lung cancer death rates are 14 times (males) or 12
times (females) greater for ES than NS, independent of age. The lung cancer death
rates are then calculated from the age-specific proportions of ES in the general
population, as is shown below. First, note that
hpop(x)= (1-PW) hNS(x) + p(x) • hES(x),
where hNS(x), hES(x) and hpop(x) are the respective lung cancer death rates for NS, ES,
and the general population, and p(x) is the proportion of ES at age x. Letting RR denote
the smoking related relative risk (14 for males, 12 for females), and substituting for hES(x)
yields:
hNS(x) + p(x) • RR • hNS(x),
orequivalently:
ftwsW = "pcpW [(1-PW) + PW • RR\\ (4a)
19
-------
We used Equation 4a to calculate the rates for NS, and Equation 4b for ES:
hES(x) = RR • hNS(x) (4b)
In BEIR VI, it was assumed that 58% of males and 42% of females are ES, for all
ages > 18 y. As an illustration, consider how this formula would be applied for males of
age 70 y. In the U.S., the lung cancer death rate, hpop(70) for such males was 0.0044.
If 58% were ES, the corresponding rates for NS and ES would be, according to
Equations 4a and 4b:
0.000515 = 0.0044 [0.42 + 14(0.58)]-1 for NS
0.0072 =14x0.000515 for ES
Extending the basic approach in BEIR VI, we allow ES prevalence to depend on
age. Estimates of smoking prevalence in 1990 for males and females, shown in Figure
1, are based on data from six NHIS surveys (DHHS 1997). Details are given in
Appendix A. Figure 1 clearly indicates that for the cancer-prone ages between 50 and
80 y, the male ever-smoking prevalence substantially exceeded 58%. As a result, our
corresponding estimates of NS and ES male lung cancer death rates for these critical
ages are somewhat smaller than in BEIR VI. For example, our estimate of ever-smoking
prevalence for males of age 70 y is 74%. Applying this prevalence to Equations 4a and
4b yields the NS and ES lung cancer death rates:
0.000414 = 0.0044 [0.26 + 14(0.74)]-1 for NS
0.0058 = 14 x 0.000414 for ES
These rates are about 21% smaller than the BEIR VI rates, which were based on the
assumption that prevalence rates are age-independent.
20
-------
0.8
0.7
£Z
_0>
CD 0 .5
>
O>
CL
O) 0 .4
o
E 0.3
CO
LJJ 0.2
0.1
10
Males
20
30
40
50
60
70
80
Age(years)
Figure 1: Ever smoking prevalence by age and gender.
90 1 00
21
-------
2. Choice of a relative risk model: As in BEIR VI, radon-induced lung cancer
death rates were obtained simply as a product of the modeled age-specific excess
relative risks, ERR(x), and the baseline lung cancer rate from all causes, h(x). For
modeling the relative risks we used a scaled version of the concentration model, one of
the two models preferred by the BEIR VI committee. The scaling results in lifetime risk
estimates intermediate between results that would be obtained from the BEIR VI
concentration and duration models.
The concentration model assumes that the risk per unit exposure increases as
the radon decay product concentration (i.e., the exposure rate) decreases down to some
limiting value, whereas the duration model assumes that the risk increases as the
exposure duration is increased to some limiting value. Obviously these two approaches
are closely related, since, for fixed total exposure, increased duration means decreased
exposure rate. Under some exposure conditions the two approaches are essentially
equivalent, and the BEIR VI committee found that the two models fit the miner data
equally well.
One might try to select one of these two models on the basis of biological
plausibility. An inverse dose rate effect has been seen in cellular studies of alpha-
particle induced mutations and transformation. If one postulates that the carcinogenic
action of radon stems from the mutagenicity of alpha radiation, the critical factor in
determining the risk per unit exposure would be the exposure rate (concentration), and
only secondarily, the duration. On the other hand, the potency of a promoter may
depend directly on exposure duration, as well as concentration. It turns out, however,
that one cannot distinguish the two models on this basis because the BEIR VI analysis
was carried out on highly averaged data, not reflective of the day-to-day, or even the
year-to-year, variations in concentrations to which miners were exposed to. Moreover,
the categorization of exposure rates and exposure durations are somewhat arbitrary,
and these categorizations may have had some effect on the limiting value for the risk per
unit exposure projected with each of the models. Thus, the difference in risk projections
from the two models may be largely an artifact of the analysis, and neither projection has
more credibility than the other. Therefore, to arrive at a "best estimate" of risk, it is
reasonable to average the two models in some way.
One approach would be for EPA to calculate risk with both models, on a case-by-
case basis, and average the results. This would be cumbersome, and it is preferable to
have a single model for calculating risks. Since the two models recommended in BEIR
VI exhibit very similar dependencies on age and time-since-exposure (see Table 3), as
well as the same two-fold higher risk for never smokers, a simple approach to averaging
is to adjust one of the models in such a way as to yield results approximately midway
between those obtained using the two unmodified models.
We chose to modify the concentration model for this purpose because, as will be
shown in the next section, the concentration model avoids ambiguities that may arise
when assessing health impacts from residential exposures at levels that change over
22
-------
time. As shown in Table 6, the risk per WLM is 6.52 x 10"4 for the concentration model
and 4.43 x 10"4 for the duration model. We scaled the concentration model so that the
risk per WLM would equal the geometric mean of these two values (5.38 x 10"4). This is
easily achieved since (see Section VLB.4), the risk per WLM is approximately
proportional to the risk coefficient p. The risk coefficient for the EPA's model (scaled-
concentration model) is:
p = 0.0768 x (4.43 / 6.52f2 = 0.0634, (5)
and the risk per WLM is 5.38 x 10'4 « (6.52 x 10'4) x (4.43/6.52)1/2.
Table 6: Risk per WLM based on BEIR VI concentration and duration models
Model Risk per WLM (1Q-4)
Concentration 6.52
Duration 4.43
Details on how the concentration and duration models were applied to obtain the values
in Table 6 are given in the next section.
3. Applying the concentration and duration models: As described in
Part IV, the BEIR VI concentration model specifies that the excess relative risk (relative
risk -1) depends on time-since-exposure, attained-age, and rate of exposure
(concentration) according to the formula:
ERR = p (w5.14 + Q15.24 w15.24 + Q25+ w25+) 4)ageYz, (3)
The 6-parameters detail how relative risk depends on time-since-exposure, and 4)age
describes the dependency on attained age. The YZ, ranging from 1 for radon
concentrations below 0.5 WL to 0.11 for concentrations above 15 WL, define the
dependency on exposure rate. This formula can be simplified by noting that \z is almost
always equal to 1, because residential exposure rates are almost always below 0.5 WL.
Letting p* = p $age, and using the (unadjusted) parameter estimates from BEIR VI given
in Table 3, the formula for the excess relative risk may then be expressed as:
23
-------
ERR = P* (w5_14 + 0.78 w15_24 + 0.51 w25+),
where p* = 0.0768 for attained age (x) < 55 y
= 0.0438 for 55 y< x < 65 y
= 0.0223 for 65 y< x < 75 y
= 0.0069 for x> 75 y.
This formula might be applied, for example, to estimate health effects at age 60 y
from a residential radon exposure at level 6 pCi/L (0.867 WLM/y) up to age 45 y, and
2pCi/L (0.289WLM/y) for the next 15 y. The estimated proportional increase in the risk
of a fatal lung cancer at age 60 y would be about 110%:
1.10= 0.0438[2.89+0.78(8.67)+0.51 (30.35)]
Figure 2 shows how the modeled excess relative risks for a constant lifetime
exposure depend on attained age. Up to age 55, the relative risks increase because
cumulative (weighted) exposures increase with age. The excess relative risks then drop
(discontinuously) at ages 55, 65, and 75. To avoid such biologically implausible
discontinuities, we use splines to smooth this function (see Figure 2) for our calculations.
The excess relative risk function is then multiplied by baseline rates to yield age-specific
rates of radon-induced lung cancer death. These were then averaged as described in
the next sections to yield the estimate of 6.52 deaths per 10,000 WLM in Table 6.
We now turn our attention to the duration model. For constant exposures and
attained ages greater than 35 y, the duration model (and simple algebra) simplifies to:
ERR = $*(w5_14 + 0.72 w15_24 + 0.44 w25+),
where now p* = 0.0561 for attained age (x) < 55 y
= 0.0292 for 55 y< x < 65 y
= 0.0157 for 65 y< x< 75 y
= 0.0073 for x > 75 y.
Unfortunately, the duration model does not adequately specify how to calculate risks that
result from exposures with changing radon levels. Returning to the example in which the
residential exposure level changes at age 45 y, it is not clear whether the risk from radon
exposures received by age 55 y should be calculated as the sum of risks from two
separate exposures of duration 35 y (YZ = 10.2) and 10y(yz = 2.78), or whether the
appropriate duration is 45 y (yz = 10.2).
Figure 3 shows how the ERR's from a constant lifetime exposure depend on
attained age for smoothed versions of the duration and scaled concentration models
(see also Appendix B for details). Figures 4 and 5 show age-specific estimates of lung
cancer death rates for male ES, male NS, female ES and female NS. Estimates of rates
24
-------
from exposure to radon were derived using the scaled concentration model. Not
surprisingly, radon-related rates of lung cancer death are many times higher for ES than
NS. Figure 6 shows the lung cancer death rates for a stationary population that
comprises all four subpopulations. These rates are weighted averages of the four sets
of age-specific rates. Formulas for averaging death rates and survival functions from
different populations are given in Appendix C.
Caution is warranted in interpreting the lung cancer death rates shown in Figures
4-6, especially those linked to exposure to radon. One would infer from those figures
that whereas lung cancer death rates from all causes would increase consistently from
age 40 y to about age 85 y, the rates of premature lung cancer death due to exposure to
radon are greatest between ages 55 y and 75 y. However, the precise form of the
temporal dependence of the risk is less certain than the estimate of lifetime risk. This is
because estimates of lifetime risk are determined using the mortality experience of
miners at all ages, whereas age-specific estimates are largely determined by the miners'
mortality experience in restricted age intervals.
25
-------
0.5
0.45
0.4
0.35
0.3
m
U
X
0.2
0.15
0.1
0.05
20
40 60 80
Attained age (years)
1 00
120
Figure 2: BEIR VI (unsealed) concentration model age-specific excess risks
from a 0.181 WLM/y radon exposure. Smoothed version also shown.
26
-------
0.4
0.35
0.3
in
•c 0.25
0.15
X
LJJ
0.1
0.05
Scaled age-concentration
20
40 60 80
Attained age (years)
oo
120
Figure 3: Smoothed age-specific excess relative risks from a constant radon
exposure at rate 0.181 WLM/y.
27
-------
1 0
10
<_> 1 0
O)
-
ct 10
1 0
40
Males (from radon)
Females (from radon)
45
so
55
60
65
70
75
80
85
90
Attained age (years)
Figure 4: Rates of lung cancer death for ES males and females. Estimated rates of
premature lung cancer death due to a constant exposure to radon of 0.181 WLM/y
are also shown. See the text for a discussion of uncertainties associated with these
estimates.
28
-------
Males (from radon)
Females (from radon)
1 o
Attained age (years)
Figure 5: Rates of lung cancer death for NS males and females. Estimated rates of
premature lung cancer death due to a constant radon exposure at rate 0.181 WLM/y
also shown. See the text for discussion of uncertainties associated with these
estimates.
29
-------
1 0
Attained age (years)
Figure 6: Rates of lung cancer deaths for a stationary population in which 53%
of males and 41% of females are ES. Rates of premature lung cancer death
due to a constant radon exposure at rate 0.181 WLM/y also shown. See the
text for discussion of uncertainties.
30
-------
4. Averaging the age-specific risks of lung cancer death: Weighted averages
of the age-specific excess lung cancer death rates shown in Figures 4 and 5 are
calculated to yield the risk estimates for male and female ES and NS. The weights are
the probabilities of survival, and the averaging is accomplished through integration.
Details follow.
Let S(x) be the probability of survival to age x for one of the gender- and smoking-
specific stationary populations, and assume a constant excess residential radon exposure
rate (WLM/y) equal to A. The survival function accounts for the increased probability of
smoking-related lung cancer death, but, for reasons discussed in Section I below, is not
adjusted for smoking-related risks other than lung cancer. Let S(x, A) be the probability
of survival to age x, adjusted to account for a small incremental lifetime excess rate of
radon exposure equal to A (for our calculation we used A = 0.00181 WLM/y). Also, let
h(x) be the baseline lung cancer death rate, adjusted to account for effects of smoking,
and e(x, A) be the ERR at age x due to the excess exposure (at a rate = A). The formula
for the lifetime risk per WLM (RWLM), is:
th(x)-e(x,A)-S(x,A)-dx
RWLM =- -
sS(x,A)-dx
The formula for calculating lifetime etiologic fraction (EF) is similar. The EF is the
risk of a premature lung cancer death from the background exposure of gb(x) (measured
in WLM/y) divided by the baseline lifetime risk of lung cancer death from all causes. (See
Greenland and Robins (1998) for an interesting discussion of problems associated with
estimating the etiologic fraction). A formula for the risk (R) of a premature cancer death
due to radon is:
= £h(x}-e(x,gb(x))'S(x,gb(x))'Cix
However, for constant gb(x) = g0, the following linear approximation for R holds:
R * (g 0 / A) • pi( *) ' e( x, A) • S(x, A) • dx
The formula for the baseline risk is:
31
-------
Our estimate of EFfrom an exposure of 0.181 WLM/y is:
re,
^ AM
EF-
h(x)S(x)'dx
The average years of life lost per radon-induced lung cancer death (YLL) is obtained
through:
(
(S(x}-S(xlA})'dx
I
5. Combined risk estimates for the U.S. population: A combined risk per WLM
estimate for the entire population is calculated as a weighted average of the male ES,
male NS, female ES, and female NS risks. The weights are proportional to the expected
number of person-years for each gender-and-smoking category. Similarly, the combined
EF and combined YLL estimates are weighted averages of the corresponding gender- and
smoking-specific estimates. For EF, the weights are proportional to the lifetime baseline
cancer death probabilities. For YLL, the weights are proportional to the lifetime risks of a
radon-induced lung cancer death. Details are given in Appendix C.
32
-------
C. Etiologic Fraction
Table 7 shows estimates for the EF, or the proportion of lung cancer deaths
induced by radon, for male and female ES and NS. These estimates have been
calculated using life-table methods applied to the BEIR VI age-concentration model as
detailed in Section B. We assumed a constant rate of radon exposure of 0.181 WLM per
year, as detailed in Section F. The estimates indicate that radon exposure accounts for
about 1 in 8 ES lung cancer deaths and about 1 in 4 NS lung cancer deaths. These
estimates are subject to uncertainties, which are quantified when feasible in Section
VIII.E. For example, 90% uncertainty bounds calculated for the ES suggest that the EF
for this group is between 0.05 and 0.3, or that the estimates shown in Table 7 for ES may
be accurate within a factor of about 3. Estimates for NS would be subject to greater
uncertainties since most of the miners were ES.
Table 7: Estimated etiologic fraction3 by smoking category and gender.
Smoking Category
Gender
Male
Female
ES
0.129
0.116
NS
0.279
0.252
'Based on 1989-91 vital statistics and mortality data (NCHS 1992, 1993a, 1993b, 1997). See the text for a
discussion of uncertainties.
The EF estimates in Table 7 for male and female ES and NS have been multiplied
by the corresponding estimates, shown in Table 8, of the lung cancer deaths in 1995
(NAS 1998). The result of these calculations are estimates of the lung cancer deaths due
to radon progeny for male and female ES and NS. The calculated total number of radon-
induced lung cancer deaths in 1995 was about 21,100: 13,000 males and 8,100 females;
18,200 ES and 2,900 NS. The uncertainties in these estimates are quantified in Section
VII.E.
33
-------
Table 8: Estimated fraction of lung cancer deaths in 1995 attributable to radon.
Gender Smoking
Category
ES
Male NS
ES and NS
ES
Female NS
ES and NS
ES
Male & Female NS
ES and NS
Number of
Lung Cancer
Deaths in 1995
90,600
4,800
95,400
55,800
6,200
62,000
146,400
11,000
157,400
Fraction Due to
Radon3
0.129
0.279
0.136
0.116
0.252
0.131
0.124
0.263
0.134
Number of
Radon-induced
Deaths in 1995
11,700
1,300
13,000
6,500
1,600
8,100
18,200
2,900
21,100
Estimates of the fraction due to radon are subject to uncertainties as discussed in the text.
An estimated 13.4% of lung cancer deaths in 1995 were radon-related. This
percentage depended on the proportion of ES among adults of lung-cancer prone ages,
because (see Table 7) etiologic fractions are about 2 times greater for NS than ES.
Theoretically, this EF could change, because the EF depends on the age-specific ES
prevalences, and these prevalences change. For example, the proportion of male ES is
much greater for people of lung cancer prone ages in 1995 than for the entire male adult
population. Over 70% of males between ages 50 and 80 years are ES compared to an
average of 58% for all adult males. It also seems likely that the proportion of children
less than 18 y who will take up smoking will be considerably lower than the ES proportion
among adults.
To calculate the EFfor all living males and females, we first assume that 37% of
males and 36% of females of ages less than 18 y will be ES. These percentages are
derived by noting that, for the youngest cohort (born 1965 to 1969) for which we have
reliable data, the proportion of ES in 1990 was about 37% for males and 36% for females
(DHHS 1997). It then follows that, since the ES prevalence for adults is about 58% for
males and 42% for females, and about 27% of males and 24% of females are of ages
less than 18 y, about 53% of living males and 41 % of females would be ES. We can then
use life-table calculations based on a stationary population for which the same 53% of
males and 41 % of females would be ES. Results are given in Table 9.
34
-------
Table 9: Estimated etiologic fraction by smoking category and gender for a
stationary population in which 53% of males and 41% of females are ES.
Gender Smoking Category Etiologic Fraction3
Male
Female
Male and Female
Male and Female
Male and Female
ES and NS
ES and NS
ES
NS
ES and NS
0.139
0.132
0.124
0.264
0.136
Based on 1990 adut (ages >18 y) ever-smoking prevalence data (58.7% males and 42.3% females are ES)
and assumption that 37% (males) and 36% (females) of children (ages < 18 y) will become ES.
Of course we do not know what percentage of children will become smokers.
Results of national surveys indicate that smoking prevalence among high school students
has increased since 1990, and is only recently showing signs of leveling off. Current
smoking estimates in youth, defined as tobacco use on at least one of the last 30 days,
has increased from about 28% in 1991 to about 36% in 1997 (Bergen and Caporaso
1999) and 1998 (CDC 2000).
The striking similarity of the EF estimates in Tables 8 and 9 reflect the fact that the
EF is generally insensitive to changes in smoking prevalence. This is true unless the ES
prevalence is very small. As shown in Figure 7, the EF decreases relatively rapidly until
the ES prevalence is about 0.2, and then gradually flattens out. For ES prevalence
between 0.2 and 1.0, the EF decreases from about 0.16 to 0.13 for males, and 0.15 to
0.12 for females. The next section shows that, in contrast to the EF, the risk per WLM is
sensitive to changes in ES prevalence.
35
-------
0.3
0.25
0.2
o
'-ip
CO
tj °-15
^5
UJ 0.1
0.05
Females
i i
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ES prevalence
Figure 7: Etiologic fraction by ES prevalence from a lifetime exposure of 0.181
WLM/y
36
-------
D. Risks per Unit Exposure and per Unit Concentration
Table 10 presents estimates of risk per WLM by smoking category and gender.
These estimate the number of expected radon-induced cancers for the current population
divided by the corresponding total of past and expected future radon exposures. The
estimates have been derived using life-table methods assuming, as in BEIR VI, that radon
exposure rate is constant during the life of each individual. Risk estimates for NS and ES
have been combined by assuming a stationary population for which 53% of males and
41% of females would be ES. For the entire population, the risk estimate is 5.38 xiO"4
fatal lung cancers per WLM. This estimate is subject to uncertainties, as described in
Chapter VII. Ninety percent uncertainty interval for the risk per WLM ranges from 2x10"4
and12xlQ-4.
The estimated risks for ES and NS are, respectively, about 1.8 and 0.3 times that
for the general population. Thus, ES are estimated to have about 6 times the risk from
radon as NS. Figure 8 shows that the estimated risk per WLM is sensitive to changes in
ES prevalence. Our risk estimates are based on the premises that 36% - 37% of children
will smoke and that children make up about a quarter of the population. Calculations
show that the proportion of those now alive who would smoke sometime during their
lifetime would be about 53% for males and 41 % for females. On the other hand, if all
children were to remain never smokers, the corresponding risk per WLM estimates would
be about 15% lower than those given in Table 10. Besides ES prevalence, the baseline
lung cancer rates, and thus also risk per WLM, will be affected by other changes in
smoking patterns, including quit rates and number of cigarettes smoked.
What do these risk estimates mean for homeowners who have had the radon level
in their home measured? Such measurements are usually given in picocuries per liter
(pCi/L) of radon gas. Assuming that, on average, people spend about 70% of their time
indoors at home and that the equilibrium fraction between radon and its daughters is 40%
(NAS 1999), it follows from the definitions in Section II that, at 1 pCi/L of radon gas, the
radon daughter exposure rate is 0.144 WLM per year.
From Table 10, the average risk of a fatal lung cancer due to lifetime exposure at 1
pCi/L is then:
(0.144 WLM/y) (75.4 y/lifetime) (5.38x10'4 /WLM) = 0.58%
In general, if the concentration in pCi/L is C, the estimated risk from lifetime exposure will
be 0.0058 C. Hence lifetime exposure at the EPA action level of 4 pCi/l corresponds to an
estimated risk of 2.3%. Similarly, forES and NS, the lifetime risks are 0.0103 Cand
0.0018 C, respectively. Again, risks for ES are almost 6 times higher than for NS. Risks
for current smokers would likely be higher than for ES. Estimates of lifetime risks for NS
and current smokers at constant concentrations are tabulated in Appendix D.
37
-------
Figure 9 provides information on how risks may depend on age at exposure.
Plotted there is the calculated lifetime risk per WLM as a function of age at exposure for
"average" members of the population. Results are also shown for ES and NS. We can
apply the results shown in Figure 9 to approximate risks for specific exposure intervals.
For example, consider an individual exposed to radon at 1 pCi/L between ages 40 y and
41 y. The estimated risk per WLM for exposures received between ages 40 y and 41 y is
7.71 xlO"4 WLM"1 so the risk for such an exposure would be about:
(0.144 WLM y1) (1 y) (7.71 x10'4 WLM'1) = 0.011%
Calculations for ES and NS, or for newborns destined to be ES or NS, would be done in a
similar manner. Again, on an age-specific basis, the model estimates for ES and NS are
approximately 180% or 30% of the general population estimate, respectively. Caution
must also be applied for assessing risks because of uncertainties associated with age-
specific risks, particularly with childhood exposures (see: Sections VII.C.3 and VII.C.4).
Table 10: Estimates of risk per WLM by smoking category and gender for a
stationary population in which 53% of males and 41% of females are ES.
~ , o , • ^ x Risk per WLMa Expected Life Span
Gender Smoking Category ^Q.4) K (years)
Male
Female
Male & Female
ES
NS
ES and NS
ES
NS
ES and NS
ES
NS
ES and NS
10.6
1.74
6.40
8.51
1.61
4.39
9.68
1.67
5.38
71.5
72.8
72.1
78.0
79.4
78.8
74.2
76.4
75.4
Based on 1990 adut (ages >18 y) ever-smoking prevalence data (58.7% males and 42.3% females are
ES) and assumption that 37% (males) and 36% (females) of children (ages < 18 y) will become ES.
38
-------
0.012
0.01
CO
0.008
0.006
Q
2?
0.004
_Q
o
0.002
_l I I I
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ES prevalence
Figure 8: Probability of a premature lung cancer death from a lifelong exposure to
radon at 1 pCi/L as a function of ES prevalence.
39
-------
x 10
1 .5
Ct
0.5
0
Ever Smokers
0 10 20 30 40 50
Age at exposure (years)
Figure 9: Risk per WLM as a function of age at exposure.
60
70
40
-------
E. Age at Cancer Death and Years of Life Lost
Table 11 shows that, according to the concentration model, radon-induced lung
cancer deaths tend to occur earlier than other lung cancer deaths. The estimated
average age for radon-induced lung cancer deaths is about 65 y compared to 72 y for all
lung cancer deaths. Years of life lost per death would then be greater for radon-induced
lung cancers than other cancers, as Table 12 shows. For both males and females, the
concentration model predicts an average of about 17 y of life lost per death when the
cancer is radon-induced.
Age at lung cancer death and years of life lost per death depend on the shape of
the ERR as a function of attained age (see Figure 3), but not the scaling. Figure 10 shows
the probability density function for years of life lost for the three different relative risk
models discussed in BEIR VI. Since the shapes of the concentration and duration ERR
functions are so similar, the resulting density functions for YLL are also similar. Not
surprisingly, the average YLL estimated using the duration model is also about 17 y. In
both cases, the ERR function is relatively large for young ages (between 35 and 55 y),
implying a greater likelihood that radon-induced cancers occur earlier. In contrast, the
constant relative risk function predicts relatively few early cancers, and as a result the
constant relative risk function would predict fewer YLL (about 12 y). Our "best" estimate of
average YLL is 17 y, which is derived using either of the two BEIR VI preferred models. In
contrast, the constant relative risk estimate of 12 y presents a reasonable lower bound for
describing the uncertainties in this estimate.
Figure 11 indicates how YLL depends on age at exposure. For both males and
females, YLL appears to be relatively constant for exposures up to about age 40 y, and
then YLL decreases with age.
Table 11: Estimated average age at lung cancer death.
Gender All Lung Cancer Deaths Radon-Induced Deaths
Males 70.6 y 64.5 y
Females 73.1 y 66.1 y
Both 71.7 y 65.2 y
41
-------
Table 12: Estimated years of life lost per lung cancer death.
Gender
All Deaths
Average
Radon-Induced Deaths
Average
Median
Male
Female
Both
13.2 y
14.4 y
13.7 y
16.1 y
18.6y
17.2 y
14.9 y
17.6y
16.4 y
0.06
0.05
0.04
\
ve- Constant RR
\
Ag^-duration
Age-concentration
Figure 10: Density function for years of life lost from a radon-induced death
42
-------
30
25
CD
— 20
CD
-t—"
CD
1 5
en
O
o
LO
CD
1 0
Females
•10
20
30
40
50
60
70
80
Age at exposure (years)
Figure 11: Years of life lost per fatal radon-induced cancer. Estimates based on
scaled concentration model for exposures of one year duration as a function of age
at exposure (midpoint of exposure interval).
43
-------
F. Comparison with Previous Estimates
Our current estimate of risk per WLM (5.38 x 10"4) is more than double the previous
EPA estimate: 2.24 x 10"4 (EPA 1992). The corresponding proportion of lung cancer
deaths estimated to have been radon-induced also increased, from 8.5% to 13.4%. Table
13 illustrates how changes in exposure parameters, the determination of baseline lung
cancer rates, mortality data used, and the relative risk model, affected our estimates of
risk per WLM and etiologic fraction.
1. Exposure parameters: The average radon daughter exposure rate is 0.181
WLM y"1. This is based on BEIR VI determinations that: (1) on average, people spend
70% of their time indoors at home (occupancy factor, Q = 0.70) and (2) in homes, the
average equilibrium fraction for radon daughters is F = 0.4. Taken together with the
estimated average radon concentration of C=1.25 pCi/L in the U.S. (Marcinowski et al.
1994), the estimated average exposure rate is:
w = C [F x 0.01 WL (pCi/L)-1] [Q x 51.6 WLM / (WL-y)'1]
= (1.25 pCi/L) [(0.7)(0.4) WL (pCi/L)'1] [0.516 WLM (WL-y)'1]
= 0.181 WLM/y
This value is about 25% lower than EPA's previous estimate of 0.242 WLM/y based on
Q = 0.75 and F= 0.5.
Changing the exposure from 0.242 WLM/y to 0.181 WLM/y has little effect on risk per
WLM but decreases the EF almost proportionally. EPA's previous risk estimate was
2.24 x 10"4 per WLM with an etiologic fraction of about 8.5%. Based on the same 1992
assumptions, but using 0.181 WLM/y, the estimated risk per WLM would be essentially
unchanged, but the estimated etiologic fraction would be 6.5%.
2. Baseline rates: Because EPA's radon risk estimates are determined using
relative risk models, they depend directly on the baseline lung cancer death rates that are
used. In 1992, we adjusted the observed lung cancer death rates downward to obtain
baseline rates from all causes otaerthan residential radon exposure. For our current risk
estimates, we have not adjusted the baseline rates. A discussion of this issue is given in
(Nelson et al. 2001).
Table 13 indicates that not adjusting the baseline rates increases both the risk per
WLM and the etiologic fraction. With previously used mortality data and relative risk
models, an exposure of 0.181 WLM/y and adjustment of the baseline rates, the estimates
of risk would be 2.3 per 10,000 WLM with an etiologic fraction of 6.5%. With the same
44
-------
inputs and relative risk model, but no adjustment of the baseline rate, the risk per WLM
would be 2.5 per 10,000 WLM with an etiologic fraction of 7.0%.
3. Mortality data: The previous EPA estimates were based on 1980 mortality data.
Updating the mortality data to 1990 increases the risk per WLM and causes a slight
decrease in the calculated etiologic fraction. The risk per WLM increases because of the
increases in baseline lung cancer rates. The etiologic fraction most likely decreased
because stationary populations based on 1990 data contain a greater proportion of older
people, and radon-related relative risks decrease with attained age.
4. Relative risk model: In 1992 EPA used the BEIR IV relative risk model:
ERR(a) = 0.0175 y(a) (W1 + Y2 WJ.
The predicted age-specific excess relative risks from the BEIR IV model tend to be only
about half as large as the ERR from the scaled concentration model. As a result,
switching to the concentration relative risk model roughly doubles both the estimates of
risk per WLM and etiologic fraction (see Table 13).
Table 13: Dependence of risk estimates on changes in methodology since 1992.
Exposure
Rate
(WLM /y)
0.242
0.181
0.181
0.181
Adjustment of
Baseline Rates
Yes
Yes
No
No
Mortality
Data
1980
1980
1980
1990
Relative Risk
Model
BEIR IV
BEIR IV
BEIR IV
BEIR IV
Risk per
WLM
2.2
2.3
2.5
3.0
Etiologic
Fraction
8.5%
6.5%
7.0%
6.7%
0.181
No
1990
Scaled
Concentration
5.4
13.4%
G. Effects Other than Fatal Lung Cancers
The estimates above refer only to fatal lung cancers. As cited in BEIR VI, lung
cancer incidence in 1994 was estimated to be about 12% higher than lung cancer
mortality (NAS 1999, DHHS 1995). Assuming that the etiologic fraction would be nearly
the same for lung cancer incidence as mortality, the numerical estimates of risk per WLM
and of radon-induced lung cancers would be about 12% higher than for lung cancer
mortality. Thus, one might project about 23,600 (21,100x1.12) radon-induced lung cancer
cases in 1995.
45
-------
To a limited extent, inhaled radon gas is absorbed into the bloodstream and
transferred to all parts of the body. Radioactive decay of this radon and its daughters
result in a radiation dose—predominantly from alpha-particles—to all potential cancer
sites. The cancer risk associated with this dose is very small compared to the lung cancer
risk due to decay of radon decay products deposited in the bronchial epithelium. Using
dosimetric models and risk factors recommended by the International Commission on
Radiological Protection, James (1992) has estimated that the risk to other organs is about
2% of the lung cancer risk.
H. Current Smokers
The BEIR VI committee provided relative risk models and estimates of radon-
induced lung cancer deaths for both ES and NS but did not provide guidance on how the
risks may differ for current versus former smokers. Most likely, this is because the miner
data provide relatively little information on how to distinguish between the former and
current smoker risks. An approach to this issue is to assume that the radon-related
relative risk given by the concentration model is the same for former and current smokers;
that is, the relative risks for both former and current smokers are assumed to be about 0.9
times the relative risks for the U.S. population. Taking into account the respective
baseline lung cancer death rates in the two groups, this made it possible to provide the
rough estimates, shown in Table 14, of the number of radon-induced lung cancer deaths
in 1995. The same relative risk assumptions were used to produce the risk per WLM
estimate for current smokers described later in this section.
To derive the estimates in Table 14, we first partitioned the number of ES lung
cancer deaths among former and current smokers, and then applied the ES etiologic
fractions of 0.129 or 0.116 given in Table 8 for males and females, respectively. The
partitioning can be accomplished as follows. First, age-specific lung cancer death rates
can be obtained using 1990 vital statistics and assuming that the lung cancer death
relative risks are 27.05 for male current smokers, 10.69 for male former smokers, 13.45
for female current smokers, and 4.47 for female former smokers (Malarcher et al. 2000).
Second, using 1990 census data and prevalence data for current and ever smokers
(DHHS 1997), we can estimate age-specific numbers of current and former smokers for
both males and females. By then applying the age-specific lung cancer death rates to
these numbers of former and current smokers, we have calculated that about 50% of male
and 67% of female ES lung cancer deaths in 1990 were among current smokers.
Assuming these percentages were similar for 1990 and 1995, about 45,300 of the 90,600
male and 37,300 of the 55,800 female ES lung cancer deaths in 1995 were among
current smokers. For males, although the risks of lung cancer are about 2-3 times greater
for current than former smokers, the surprisingly high number of lung cancer deaths
among former smokers is due to a much higher former smoker than current smoker
prevalence at ages at which cancer is most likely to occur.
46
-------
Relative risks for current and former smokers are based on data from the Cancer
Prevention Study II (CPS II). The CPS II is a large cohort study of about 1.2 million
participants, who were recruited by volunteers from the American Cancer Society. The
CPS II participants are not representative of the general U.S. population. However the
Office of Smoking and Health (OSH) of the Centers for Disease Control has carefully
analyzed the CPS II data to ensure the validity of risk estimates that result from the survey
(Malarchar et al. 2000).
Table 14: Estimating radon-induced lung cancer deaths for current and former
smokers.
Gender
Male
Female
Smoking
Category
Ever
Current
Former
Ever
Current
Former
Lung Cancer Deaths
90,600
50% of ES
50% of ES
55,800
67% of ES
33% of ES
45,300
45,300
37,300
18,500
Fraction
Due to
Radon
0.129
0.129
0.129
0.116
0.116
0.116
Radon-
Induced
Lung Cancer
Deaths
11,700
5,850
5,850
6,500
4,300
2,200
Since they have a higher baseline rate, it is likely that the the risk per WLM would
be greater for current smokers than ever smokers. Using the same life-table methods
described in Section VLB, and relative risk values of 27.05 for males and 13.45 for
females it is possible to calculate a crude estimate of risk per WLM for current smokers
(presumed to start smoking at age 18 y and do not quit) equal to 15 x 10"4 (rounded to the
nearest 5 x 10"4). This estimate suggests about a 50% greater risk per WLM for lifelong
smokers than ES.
It should be emphasized that the estimates given in this section may be especially
sensitive to assumptions on smoking, including some that were not needed in BEIR VI
(because the committee confined estimates to NS and ES). The next section offers a
discussion on estimation problems related to smoking.
47
-------
I. Dependence of Lung Cancer Death Rates on Smoking
The validity of the life-table calculations as estimates of risk for present and future
radon exposures depend on several factors, including whether mortality rates —
especially lung cancer death rates — remain reasonably stable. Possible changes in
lung cancer rates must be considered because these rates are extremely sensitive to
changes in smoking prevalence and habits. Smoking patterns have changed and will
continue to evolve, and these changes will undoubtedly affect future risks related to radon
exposure.
For example, in 1990, the proportion of Americans of ages 25 to 44 y who had ever
smoked 100 cigarettes (ES) was about 50%, versus 60% for adults of ages 45 to 64 y
(CDC 1994). To account for the complicating effects of changing smoking patterns on
mortality rates, we must make separate life-table calculations for ES and NS. As in BEIR
VI, our separate life-table calculations have only accounted for the differential mortality
effects of smoking-related lung cancer, but it has been estimated that lung cancer
accounts for only about 28% of current smoking-related deaths in the U.S. (Bergen and
Caporaso, 1999). Other major health effects of smoking include an increased risk of
circulatory disease, benign lung disease such as emphysema, and other cancers. To
determine whether we need to account for the differential mortality effects from all
smoking-related diseases, we have made preliminary risk per WLM calculations for ES
and NS, using life tables (Rogers and Powell-Griner 1991) for heavy, light, former and
never smokers that accounted for the increased smoking-related risks of death from all
causes (not just lung cancer). The tables had been derived from three national surveys,
the 1985 and 1987 National Health Interview Surveys, and the 1986 Mortality Followback
Survey. The overall risk per WLM derived this way differed only slightly from the risk per
WLM already described in Section D and shown in Table 10. We can then conclude that
the risk per WLM is not very sensitive to differential mortality due to smoking-related
causes other than lung cancer. For simplicity, we have decided not to consider this issue
further.
More vexing problems are suggested by results from the Cancer Prevention
Studies, which showed that relative risks of lung cancer death associated with smoking
change overtime (DHHS 1997). For the period 1959-65, the estimated relative risk for
current smokers was 11.9 for males and 2.7 for females. This means that among males,
the lung cancer rate for current smokers was 11.9 times as great as that for never
smokers, and for females the ratio was 2.7. For 1982-88, estimated relative risks were
27.1 for males and 13.5 for females. Factors that influence these relative risks include
cigarette composition, number of cigarettes smoked, and smoking duration. Similarly,
changes among former smokers, including trends in time since cessation, would affect
both former and ever smoker relative risks.
Even if smoking patterns were stable, determining the relationship between
smoking and lung cancer rates would still be complicated. Results from national studies,
48
-------
given in Table 15, indicate that the relative risks may depend on age at expression
(Malarcher et al. 2000). From CPS II, the relative risks for current smokers tend to
decrease with age at expression. A consistent trend in age-specific relative risks is not as
evident from estimates derived by using data from both the National Mortality Feedback
Survey (NMSF) and the National Health Interview Survey (NHIS). The NMSF collected
data on a representative sample of decedents of ages 25 years or older; the NHIS is a
nationally representative household survey. Results based on data from NMFS and NHIS
may not be as reliable as from the CPS II, as indicated by the wider confidence intervals,
and since the NMSF had to rely on proxy respondents for information on smoking. For a
more comprehensive discussion, see Malarcher et al. (2000).
Unfortunately, at this time, one can not reliably predict how smoking-related relative
risks may change over time, or quantify the dependence on age at expression. We have
therefore decided to follow the recommendations of BEIR VI, which assumed that the
relative risk of lung cancer death for ES is 14 for males and 12 for females, independent
of age. Table 16 suggests that our estimates of risk per WLM for the general population
and ES may be somewhat insensitive to assumptions about the relative risk of fatal lung
cancers for ES. For relative risks ranging from 9.33 to 28 for males and 8 to 24 for
females, the risk per WLM would be within 6% of the nominal estimates for either ES or
the general population. In contrast, the estimated risk per WLM for NS would range from
0.9 x 1C'4 to 2.4 x 1Q-4.
49
-------
Table 15: Age-specific relative risks3 and age-adjusted relative risks of fatal lung
cancers for current and former smokers'3 versus never smokers for whites
(from Malarcher et a/. 2000).
Sex
Male
Female
Age (y)
35-59
60-69
70-79
> 80
Age adjusted
35-59
60-69
70-79
> 80
Age adjusted
NMFS/NHISC
Current
82.05
(19.8, 339)
10.73
(4.26, 26.7)
8.78
(3.65, 21.1)
19.25
(4.22, 87.8)
40.65
(15.7, 105)
32.13
(7.56, 137)
11.22
(4.64, 27.1)
21.70
(9.50, 49.5)
27.19
(11.1,66.4)
24.39
(10.2, 58.4)
Former
27.96
(6.38, 122)
3.52
(1.37, 7.03)
3.86
(1.62, 9.19)
8.92
(2.04, 39.0)
14.33
(5.33, 38.6)
12.77
(2.71, 60.1)
5.65
(2.13, 15.0)
7.50
(3.09, 18.2)
3.68
(1.28, 10.5)
9.15
(3.59, 23.3)
CPS lld
Current
27.21
(16.5,44.8)
30.71
(21.4,44.0)
27.23
(19.6, 37.9)
13.40
(8.18,21.9)
27.05
(19.3, 37.9)
14.77
(10.8, 20.2)
14.70
(11.7, 18.5)
11.28
(8.88, 14.3)
7.31
(4.76, 11.2)
13.45
(11.1, 16.3)
Former
11.09
(6.65, 18.5)
11.25
(7.82, 16.2)
9.43
(6.77, 13.1)
6.55
(4.15, 10.3)
10.69
(7.57, 15.08)
4.53
(3.16, 6.50)
5.05
(3.88, 6.55)
4.50
(3.44, 5.90)
2.95
(1.81,4.83)
4.47
(3.58, 5.59)
'Central estimates of relative risks with 95% confidence intervals given in parentheses.
b Malarcher et al. define current smokers as persons who reported they smoked now; former smokers
reported they had ever smoked but did not smoke now.
0 NMFS, National Mortality Feedback Survey; NHIS, National Health Interview Survey
d CPS II, Cancer Prevention Survey II
50
-------
Table 16: Sensitivity of risk perWLM estimates to assumptions about the relative
risk of fatal lung cancers for ES compared to NS.
Ratio of ES
lung
Male
9.33
14
21
28
divided by NS fatal
cancer rates
Female
8
12
18
24
Risk per WLM
ES
9.39
9.68
9.89
10.0
(10"4) by smoking category
NS All
2.41 5.64
1.67 5.38
1.15 5.19
0.88 5.10
J. Summary
We described three lifetime radon-related risk estimates for stationary populations
based on 1990 U.S. mortality rates: the risk per WLM (5.38 x 10'4), EF (about 0.134 in
1995), and YLL (17.2 y). Our estimates of risk per WLM are much larger for ES than NS,
but EF is about twice as large for NS than for ES. These estimates are based upon a
scaled version of the BEIR VI concentration model, assumptions that exposure to radon
are constant, and assumptions about ES prevalence and smoking related health effects.
We have discussed many of the ways these estimates depend on these assumptions and
have shown, for example, that estimates of risk per WLM may be more sensitive to
assumptions about smoking prevalence than estimates of EF. We have also presented
risk estimates for specific ages at exposure with the caveat that these estimates are
subject to considerable uncertainties. These include what might be termed "modeling
uncertainties". Almost all risk estimates, including those described in this document, are
dependent on the modeling framework used for the data analysis. The BEIR VI
committee used relative risk models to analyze the miner data. Alternative models for the
analysis of the miner data will be one of the topics of the next chapter.
51
-------
VII. UNCERTAINTIES
A. Background
The BEIR VI committee identified 13 sources of uncertainty in its estimates of risks
from indoor radon. These were divided into two categories: (1) uncertainties in the
parameter estimates for the exposure-response model derived from the miner data and
(2) uncertainties in specifying the form of the model and in its application to the general
U.S. population. A quantitative uncertainty analysis was performed, but it was limited to
those factors that could be addressed without relying heavily on the subjective judgment
of experts. Thus, the quantitative analysis considered only the statistical variability in the
miner data and the comparative dosimetry between mine and residential exposures.
Employing a "random-effects model," the committee incorporated variation among cohorts
as well as sampling variation within cohorts.
The BEIR VI committee provided quantitative uncertainty estimates for the AR and
for the number of radon-induced lung cancer deaths, based on each of the models
derived from the miner data. For the concentration model, the 95% confidence interval
around the central estimate of AR (14%) ranged from about 10 to 27%. For the duration
model, the central estimate was about 10% and the uncertainty interval ranged from about
8 to 20%. The committee also considered a simple constant relative risk model (CRR),
which was based on an analysis of only those miners receiving an estimated exposure of
less than 50 WLM. Although the CRR model was deemed to have less credibility than the
duration or concentration model for calculating central estimates of risk and lung cancer
deaths, the committee's preferred uncertainty estimates were obtained from the CRR
model. The CRR analysis led to an uncertainty range of 2-21 %, with a central estimate of
about 12%.
The most striking difference among the projections is in the lower bound estimate
for the CRR model, which is much lower than for the other two models. This difference
primarily results from larger sampling errors inherent to the CRR estimate. Limiting the
study population to miners with low exposures sharply reduces both the number and the
attributable risk of radon-induced lung cancers, causing a large increase in the relative
standard error. In stating its preference for the CRR model estimate of uncertainty, the
BEIR VI committee notes that the low radon exposure conditions included in the CRR
analysis are more comparable to those in homes.
We believe that the BEIR VI CRR model-based uncertainty analysis should be
interpreted with caution since it excludes the useful information from miners with
exposures greater than 50 WLM. In particular, the CRR analysis excludes available
information about relationships between the dose response and modifying factors such
as time-since-exposure or attained age. The much wider uncertainty intervals derived
using the CRR approach appear to be a consequence of an arbitrary cutoff leading to a
substantial increase in sampling error. There seems to be insufficient justification to
reduce the lower bound estimate well below what was derived from analyses of the entire
52
-------
data set. Although the BEIR VI CRR-model uncertainty bounds provide a useful indication
of how our estimates depend on data from miners with exposures <50 WLM, we believe
that, at this point, the use of the scaled concentration model for deriving uncertainties is
more consistent with our overall approach.
It should be noted that the uncertainty ranges derived in BEIR VI, based on
application of the preferred models to the entire data set, do not fully reflect the degree of
uncertainty because important sources of uncertainty were not factored into the
quantitative uncertainty analysis. Sections B and C, below, discuss sources of uncertainty
not treated quantitatively in BEIR VI.
As in BEIR VI, we have generally limited our quantitative uncertainty analysis to
factors that can be addressed without relying heavily on subjective expert judgement. The
quantitative uncertainty analysis relies on a Monte Carlo simulation that accounts for
uncertainties in residential exposures (see Section D), uncertainties in parameters values
in the BEIR VI concentration model, the K-factor, and a scaling factor (included because
of differences between results from the BEIR VI duration and concentration model). Major
differences between our Monte Carlo simulation and the simulation used in BEIR VI is in
the way we treat uncertainties in residential exposures, the K-factor, and the fact that the
BEIR VI simulations do not account for the additional scaling factor. Neither our
simulation (see Section E) nor the BEIR VI simulation account for uncertainties in
extrapolating to low exposure rates — see Section F for a discussion. Finally a very
simple sensitivity analysis is given in Section G to indicate how our estimates may depend
on assumptions about risks from exposures to ES, NS, and children, and how relative
risks depend on time-since-exposure.
B. Uncertainties in the Miner Data
1. Errors in exposure estimates: There are two such major issues with respect to
the miner data itself. First, the data on miner exposure is deficient in many ways, which
may bias the estimate of the relative risk coefficient to varying degrees in the individual
cohort studies. Moreover, as stated in BEIR VI (MAS 1999, p. 161): "For most of the
cohorts, exposure measurement errors are likely to be greatest in the earliest periods of
operation, when exposures were largest and fewer measurements were made. For this
reason, measurement errors not only affect the estimates of the overall risk coefficient, but
may also bias estimates of parameters that describe the relationship of risk with other
variables such as exposure rate, time-since-exposure, and age at risk." Since the
magnitude of possible errors in exposure estimates are often extremely difficult or
impossible to quantify, it is very hard to estimate the magnitude of the uncertainty in risk
estimates introduced by this source. The reasonable concordance among the various
miner studies is somewhat reassuring on this point; in particular, removal of any one study
from the analysis has little effect on the overall risk estimate. Nevertheless, differences in
the ERR/WLM estimated from the various miner studies are larger than what could be
expected from sampling errors alone, and it is likely that the exposure errors do contribute
substantially to these differences.
53
-------
2. Confounding by other exposures: Second is the issue of possible
confounding with other mine exposures. Some miners were exposed to arsenic, silica,
and diesel exhaust, all of which may affect lung cancer risk. The BEIR VI report
concludes that diesel exhaust appears to be a weak carcinogen and is "probably not a
strong modifier of the risk of radon progeny." The data on silica are somewhat conflicting,
and silica's role has not been directly assessed, but the scant epidemiological evidence
does not show silicosis to be a strong modifier of radon risk (Samet et al. 1994, MAS
1999). Two of the miner cohorts, China and Ontario, had quantitative data on arsenic
exposure; in addition, Ontario, Colorado, New Mexico, and France had data indicating
whether miners had previous mining experience. Adjusting for arsenic exposure in the
Chinese cohort sharply reduced the estimate of ERR/WLM from 0.61% to 0.16%.
Otherwise, adjustment for arsenic or previous mining made little difference in the
estimated risk coefficient.
In evaluating the possible effect of other exposures on the estimated ERR/WLM, it
is important to consider whether these exposures are correlated with radon progeny
exposures and whether they act synergistically with radon in causing lung cancer. So, for
example, in the case of the Chinese tin miners, arsenic and radon exposures were highly
correlated; moreover, when the data were adjusted for arsenic exposure, the ERR/WLM
was similar across arsenic exposure categories, suggestive of a multiplicative interaction
between the two carcinogenic agents. Failure to adjust for arsenic exposure, in this case,
would have led to an overestimate of the ERR/WLM. Were it to be determined that the
interaction is actually submultiplicative, the estimated ERR/WLM would have to be
increased from 0.16% to a value between 0.16% and 0.61%.
In contrast, if there were a multiplicative interaction but no correlation between
exposures, no bias in the ERR/WLM would result. For example, little correlation between
radon and cigarette smoking is expected; furthermore, the two agents are strongly
synergistic in causing lung cancer. Therefore, confounding by smoking should be small.
Nevertheless, uncertainties in miner smoking may produce considerable uncertainty in
radon risk estimates. This uncertainty results not from confounding, but from the lack of
detailed smoking information.
On the other hand, if there were no correlation and no synergism (i.e., the effect of
the two exposures is additive), then the other exposure will simply produce a uniform
increase in lung cancer rates across radon exposure categories, resulting in an
underestimate of the ERR/WLM.
3. Smoking by miners: Five miner cohorts had useful information on smoking:
China, Colorado, Newfoundland, Malmberget, and New Mexico. From this restricted data
set it was determined that the interaction between radon and smoking was probably
submultiplicative, although a multiplicative interaction could not be excluded. Overall, a
best fit to the data indicated that NS had about 2.1 times the ERR/WLM as ES. There are
wide uncertainty bounds on the risk estimate for NS, and therefore on the ratio of the risk
54
-------
coefficients for the two groups. Some perspective on the uncertainty can be gained by
examining the effect of omitting any one of the cohorts from the analysis (MAS 1999).
Omitting the Chinese cohort data had the largest percentage effect, reducing the ratio
from 2.1 to 1.2. The largest increase was found when the New Mexico cohort was
omitted; in that case the estimated ratio increased to almost 3. While the BEIR VI
committee did not provide any quantitative uncertainty estimates by smoking category, it
would appear that there is roughly an extra factor of two uncertainty in the risk for NS
compared to that for ES (or for the general population).
There is very little information on risks to miners who had ceased smoking, and
BEIR VI does not explicitly quantify the risk to former smokers. Former smokers are
subsumed in the ES category. Using the BEIR VI model and baseline lung cancer rates
for former smokers, risk estimates for this group could be derived. Estimates for former
smokers would be more uncertain than those for ES, as a group; moreover, the relative
risk for individual former smokers may vary greatly with detailed smoking and radon
exposure histories.
C. Uncertainties in Extrapolating to Residential Exposures
1. AC-factor: In performing its quantitative uncertainty analysis, the BEIR VI
committee did consider variability and uncertainty in the K-factor, concomitant with the
uncertainties in modeling the miner data. The variability in Kwas characterized as a
lognormal distribution with a gm=1.0 and a gsd=1.5. The gsd itself was assigned an
uncertainty distribution that was loguniform over the range from 1.2 to 2.2, but no
uncertainty in the gm was assumed. When the variability in K was incorporated into the
quantitative uncertainty analysis, the uncertainty distributions were shifted upwards.
Basically, the reason for the shift is that the mean value of the distribution for K is higher
than the gm, which was used as the nominal estimate. Factoring in the uncertainty in the
gsd had little effect on the respective lower bound estimates but did lead to significant
increases in upper bound and median estimates (MAS 1999: Table A-10).
The BEIR VI committee treated the uncertainty associated with the median
estimate (gm) for K as negligible. While the variability in K is larger than the uncertainty
in its median value, this seems unreasonable in view of the sensitivity of the K-factor to
how the respiratory tract is modeled and our imperfect knowledge of the parameters
affecting estimates of K, such as aerosol size distributions, ultrafine fractions, breathing
rates, nasal deposition of the ultrafine activity, and relative radiosensitivity of lung regions.
In particular, only limited information is available for estimating aerosol conditions in mines
without diesel engines despite the fact that many miners in the epidemiologic studies
worked in mines without such equipment (Cavallo 2002). Correspondingly, the aerosol
size distributions in homes were based on measurements in just 6 homes (MAS 1999).
Cavallo (2000) argued that it is likely that the K-factor is much less than 1, because
in mines "more particles are found at larger or smaller diameters where the deposition and
dose per unit of progeny concentration is substantially higher." James et al. (2003)
55
-------
calculated a value of K of about 1 based on the exposure assumptions described in BEIR
VI. Other results (Porstendorfer and Reineking 1999) indicate that the K-factor could be
somewhat larger than 1. Using a lung dose model "with a structure that is related to the
new ICRP respiratory tract model", their estimate of dose per unit exposure is about 1.1
times as large for homes with "normal" aerosol conditions as compared to mines without
working activities. For homes without smoking, their calculations indicate a K-factor as
large as 1.5, suggesting a sensitivity to aerosol conditions in the home. To be consistent
with the range of K-factors that these results and arguments suggest, we have
subjectively assigned a normal distribution with |j =1.0 and o =0.25 for the median value of
the K-factor.
2. Dependence of risks on gender: The miner cohorts used to develop the BEIR
VI risk models consisted of males only. The BEIR VI committee assumed that the risk
models derived for male ES and NS apply equally to females. This seems reasonable.
The estimated K-factor is almost the same for females and males (MAS 1999). Baseline
lung cancer rates are substantially lower for females than for males, but this reflects
differences in past smoking patterns; overtime, lung cancer rates in females are
approaching those in males. Indeed there is now evidence suggesting that females are
more susceptible to tobacco carcinogens (Zang and Wynder 1996). It is unclear how
such differences in susceptibility would be reflected in radon risk; it would depend on
whether the degree of synergism between smoking and radon was appreciably different
for the two sexes. Given the lack of information on this point, radon risks for females must
be regarded as more uncertain.
3. Dependence of risks on age at exposure: Essentially all the data on
childhood exposures to radon were obtained from the Chinese tin miner cohort, and even
those data are relatively sparse. Consequently, the uncertainty in risks associated with
childhood exposures must be regarded as substantially higher than for adult exposures.
As shown in Table 17, the ERR/WLM observed for the Chinese miners who began mining
as children is generally about a factor of two higher than for others in the cohort, even
after adjustment in the background risk for various other factors (Xuan et al. 1993). Since
any enhanced effect of childhood exposures among these miners would have been
diluted by the effect of additional exposures received as adults, these results suggest that
the relative risk coefficient associated with childhood exposures could be several times
higher than for adult exposures. These findings must be interpreted with caution,
however. Xuan et al., as well as others who have examined these results (Lubin et al.
1994, MAS 1999), concluded that the pattern of risk did not vary consistently with age at
first radon exposure.
56
-------
Table 17: Effects of age at first radon exposure on ERR per WLM for several
analyses of a Chinese tin miner cohort (from Xuan et al. 1993).
ERR
adjustment3
None
Attained age
Time-since-
exposure
Radon rate
<10
1.0b
1.0
1.0
1.0
Age
10-14
1.1
1.1
1.3
1.2
at first radon
15-19
1.1
0.9
1.1
1.2
exposure
20-24
0.2
0.3
0.3
0.4
(y)
25-29
0.4
0.5
0.6
0.5
>30
0.7
0.4
0.5
0.7
a Each analysis adjusted ERR by either attained age, time-since-exposure, radon rate, or none ofthe above.
b Baseline risk fixed at 1.0
c Background risk adjusted for age and arsenic exposure.
There are also the unique features of the Chinese miner cohort that may make the
results on these miners less applicable to the residential exposures of interest than those
obtained from the other miner cohorts. In particular, lung cancer rates in the cohort are
extremely high (38% of all deaths), probably due in large part to arsenic exposure (Xuan
et al. 1993). Since arsenic and radon exposures are strongly correlated, the potential for
confounding by arsenic is great. Moreover, these miners were unique in their smoking
habits, using water pipes in addition to cigarettes. The nature of the interactions between
the smoking, arsenic, and radon in causing lung cancer is problematic. It is also
noteworthy that, after control for arsenic exposure, the estimates of ERR/WLM are
substantially lower for this cohort than for the other miner cohorts. Thus, the estimated
risk coefficients for the Chinese miners who began working as children are still lower than
the estimated risk coefficients derived from the other miner studies.
Epidemiologic follow-up on the atomic bomb survivors fails to show any clear
evidence of an enhanced risk of radiation-induced lung cancer associated with childhood
exposures as is seen with some other cancer sites (Thompson et al. 1994). Moreover,
both the miner data and the atomic bomb survivor data on lung cancer exhibit a falloff in
the ERR with time after exposure, a falloff that is likely to continue beyond the period of
epidemiologic follow-up. Although the radiation exposure was predominantly low-LET
(y-rays) in the case of the atomic bomb survivors rather than high-LET (a-particles), these
considerations also tend to argue against highly elevated lifetime lung cancer risks from
childhood radon exposures.
In conclusion, the uncertainties in risk estimates for childhood radon exposures are
larger than for the general population but are difficult to quantify. Information on these
risks could, in principle, be gained through residential studies, but carrying out such
57
-------
studies would be logistically difficult—and might introduce new uncertainties— because of
the long time elapsing between exposure and onset of the disease.
4. Smoking patterns in the U.S. population: As shown in Figure 7, estimates of
EF (i.e., the fraction of lung cancers attributable to radon) are fairly insensitive to the
baseline lung cancer rates or smoking patterns in the population. In contrast, projections
of risk per unit exposure or of the number of radon-induced lung cancer deaths are
strongly dependent on population lung cancer rates and smoking patterns. As a
consequence, there is additional uncertainty in projecting population risks due to current
and future radon exposures because of uncertainties with respect to future trends in lung
cancer rates and smoking.
Another source of uncertainty relates to how the synergism between radon and
cigarette smoke depends on the temporal pattern of the exposures. Animal data suggest
that the exposures act synergistically only when the radon exposure precedes the
cigarette smoke (Chemaud et al. 1981, Cross 1994). This would suggest that the risks of
childhood radon exposures would be enhanced by adolescent or adult smoking. On the
other hand, radon exposure to former smokers (subsequent to smoking cessation) may
pose only about the same risk as exposures to NS of the same age. At this time,
however, the risks from radon exposures received prior to starting, or subsequent to
quitting, smoking remain highly uncertain.
D. Uncertainty in the Estimate of Average Residential Exposure
As noted in Section VI.F, the average annual residential exposure is estimated to
be:
w = C [F x 0.01 WL/(pCi/L)] [Q x 51.6 WLM (WL-y)'1]
where: C is the average radon concentration in homes, F is the average equilibrium
fraction, and Q is the occupancy factor (average time spent indoors at home ). The
nominal estimates adopted for C, F, and Q are 1.25 pCi/L, 0.4, and 0.7, respectively,
which imply a nominal average exposure rate of w =0.181 WLM/y. To evaluate the
uncertainty in w, we must consider the uncertainty in each of the three parameters, C, F,
and Q .
1. Uncertainty in the average radon concentration (C): EPA's National
Residential Radon Survey (NRRS) determined the average radon concentration to which
people are exposed in their homes is 1.25 pCi/L, with a standard error in measurement of
0.06 (Marcinowski et al. 1994). However, this estimate was based on a simple arithmetic
average of the concentration levels on floors "frequently occupied." This takes no account
of: the fraction of time spent on each floor; the variability of radon level on a particular floor
and how this is correlated with people's location in the house; nor the time spent on floors
not classified as a frequently occupied areas (particularly basements). Floor occupancy
58
-------
data were collected in NRRS, but they primarily reflect summertime activity patterns,
which are likely to be atypical; consequently, they were not used to adjust the estimate of
average indoor radon. If they had, the estimate would have been reduced by about 7%.
Based on these considerations, a normal distribution is assigned to Cwith |j =1.2 and
a =0.08.
2. Uncertainty in equilibrium fraction (F): The BEIR VI committee
recommended a value of 0.4 for F, based on a detailed study of six homes (Hopke et al.
1995). Within these homes the equilibrium fractions were highly variable with time and
dependent on the presence of a smoker. While this study was sensitive to temporal
variations in F, it is unclear to what extent this small sample of homes is typical of U.S.
residences. Based on measurements in 21 homes in New York or New Jersey, George
and Breslin (1980) found that F was, on average, about 50% in basements and about
60% or higher on other floors; in contrast, an average equilibrium fraction of 33% was
determined from measurements of 20 houses in Butte, Montana (Israeli 1985). A larger
survey of livable areas in 200 houses conducted by the state of New Jersey yielded an
average equilibrium factor of 45% (NJDEP 1989). To characterize the uncertainty in F, we
have subjectively assigned a lognormal probability distribution, with gm=0.40 and
gsd=1.15, corresponding to a 90% Cl of 0.32-0.50.
3. Uncertainty in the average occupancy factor (Q): A large survey of human
activity patterns has recently been conducted by the EPA (Tsang and Kleipeis 1996).
Results from this National Human Activity Pattern Survey (NHAPS) are summarized in
EPA's 1997 Exposure Factors Handbook. The NHAPS data are compilations of 24-hour
diary information from a sample selected using a random digit dial method. Results from
the 9,386 respondents were weighted to obtain results representative of the U.S.
population and of specific demographic factors, seasons, etc. It was found that, on
average, Americans spend 67% of their time indoors in residences. The sampling error is
estimated to be only about 0.3%. The response rate to the survey was 63%; of the
remaining 37%, roughly two-thirds were contacted but refused to participate, and the
remainder could not be contacted. The incomplete response may result in some error. In
particular, the survey may have missed people who were away from home on vacation,
etc. As a result, the estimate of Q is likely to be biased high. Other errors may result from
recall bias and imperfections in the sampling methodology. Taking into account these
problems, we have assigned a normal probability distribution to Q, with a mean of 0.65
and a standard deviation of 0.03.
E. Monte Carlo Simulation
We describe here a Monte Carlo simulation for quantifying uncertainties for
estimates of risk per WLM, EF, YLL, and number of (radon-induced) fatal lung cancer
deaths. The simulation is similar to those used by the BEIR VI committee, in that it is
limited to factors that can be addressed without relying heavily on subjective expert
judgement. The simulation accounts for uncertainties in factors for determining the
59
-------
average residential radon exposure, the K-factor, and uncertainties in the fitted parameter
values in the BEIR VI concentration model for calculating age-specific ERR's.
Distributions for these parameters are given in Tables 18 and 19. Parameters for
determining (miner-equivalent) doses, average radon concentration (pCi/L), occupancy
factor, equilibrium factors, and the K-factor, are all assigned normal or lognormal
distributions as detailed in the previous section. The BEIR VI concentration model for
calculating the corresponding ERR's can be written as:
ERR (a)= A P (ws_14 + Q15_24 w15_24 + 625+ w25+)<$>(a) (6)
Here, we have explicitly included the scaling factor, A, which was assigned the nominal
value 0.825 to obtain risk estimates between the concentration and duration model
estimates. In our simulation, A has a lognormal distribution with gm = 0.825 and a gsd =
1.31, the gsd of the duration and concentration model RWLM estimates. This distribution
for A reflects the dependence of the modeled ERRs on the way either exposure duration
or concentration are categorized. For the simulation, we assumed as in BEIR VI that the
attained age function, 4)(a), is constant within age intervals <55 y, 55 through 64 y, 65
through 74 y, and >75 y. As in the BEIR VI report, we assigned lognormal or normal
probability distributions to the parameters p, Q15.24, 625+, $55_64, $65_74, and 4)75+, with median
values equal to the fitted BEIR VI concentration model estimates and the covariances
shown in Table 18.
The simulation of lifetime risk estimates began by repeatedly generating n = 10,000
sets of the exposure factors, K-factor, and relative risk model parameters. For each set of
factors/parameters, we then calculated the average residential exposures and
corresponding age-specific ERRs. The lifetable methods described in Section VLB.4 were
then used to calculate the risk per WLM, EF, and YLL. For all our simulations, we used
1989-91 mortality data and used age-specific ES prevalence data. As in the BEIR VI
report, we multiplied the age-specific relative risks (for radon-induced deaths) by a factor
of 2.0 for NS and 0.9 for ES. The relative risk for all lung cancer deaths between ES and
NS was set to 14 (males) and 12 (females).
Results from the Monte Carlo simulation are summarized at the bottom of Table 19.
The risk per WLM is about 2x10'4 to 12xlQ-4, and the YLL is about 15 y to 20 y. The
average annual residential exposure is between 0.12 WLM and 0.21 WLM, resulting in an
EF greater than 0.05, corresponding to more than 8,000 radon induced fatal lung cancers.
It seems highly unlikely that the EF or the number of radon-induced fatal lung cancers are
as large as the calculated upper bound limits (0.30 and 45,000). Nominal estimates for
the risk per WLM, EF, the number of premature lung cancer deaths, and YLL are all very
close to the respective median values. To separate out the effect of uncertainties in
exposure factors, the simulation was repeated with radon concentration, occupancy factor,
and equilibrium factor set to the nominal values: 1.25 pCi/L, 0.7, and 0.4. Resulting
uncertainty intervals (see Table 20) are 0.06 to 0.3 (EF) and 9,000 to 50,000 (number of
premature lung cancer deaths). It appears that uncertainties in exposure factors are
60
-------
minor compared to uncertainty in factors that determine relative risk. For EF, the ratios of
the endpoints of the uncertainty intervals are about 5, regardless of whether the exposure
factors are simulated or held constant.
Results of the simulation are sensitive to the choice of distribution for A, for which
minimal information was available. The simulation does not account for uncertainties
associated with errors in miner exposure estimates, confounding due to exposures other
than radon in the mines, health effects related to ever-changing smoking habits, risks
associated with childhood radon exposures, or model mis-specification. In our case, no
model mis-specification would mean that the ratios of residential radon induced lung
cancer mortality rates to background rates depend only on the exposures in WLM within
intervals determined by the BEIR VI age-concentration categorizations for attained age
and time-since-exposure. A discussion of alternative risk models that are biologically
motivated is given in the next section. This is followed by a section that describes the
sensitivity of our risk estimates to parameters that characterize and differentiate risks for
subgroups such as ES, NS, and children.
61
-------
Table 18: Parameters for uncertainty distributions for risk factors in the
concentration model (MAS 1999)
I. Estimated Parameter Values
Parameter
Value
log(P)
-2.76
6 15-24
0.77
625+
0.51
log(4)55-64)
-0.56
iog(4)65-74)
-1.23
iog(4)75+)
-2.38
II. Covariance Matrix
log(P)
° 15-24
^25+
iog(4)55-64)
iog(4)65.74)
iog(4)75+)
log(P)
9.47
-0.36
-0.04
-2.87
-3.18
-3.44
6 15-24
0.77
0.24
-0.10
-0.17
-0.19
625+
0.42
-0.15
-0.33
-0.54
iog(4)55-64)
5.71
2.85
2.90
iog(4)65-74)
10.87
3.20
iog(4)75+)
87.65
62
-------
Table 19: Monte Carlo simulation of Risk per WLM, EF, YLL, average residential
exposure, and number of radon-induced fatal cancers.
I. Parameter Assumptions
Radon concentration
(pC/L)
Exposure
factors Occupancy factor
Equilibrium factor
K-factor
Proportion of youth (< 18 y) that will
smoke
Exposure response parameter ratios
Relative risks of lung cancer death
from smoking; (ES vs. NS)
Age-concentration model parameters
Relative risk model scaling parameter
Normal (p =1.2, o = 0.08)
Normal (p = 0.65, o = 0.03)
Lognormal (gm = 0.4, gsd = 1.15)
Normal (p = 1.0, o = 0.25)
0.37 (males); 0.36 (females)
0.9(ESvs. All); 2.0 (NS vs. All)
14.0 (Males), 12.0 (Females)
See Table 18
A~LN (gm = 0.825, gsd = 1.31)
II. Results
Risk per WLM
(10-4)
Etiologic fraction
Years of life lost per
radon-induced death
Number of fatal lung
cancer deaths from
radon exposure
Exposure (WLM/y)
Smoking
Category
ES
ES and NS
ES
ES and NS
ES and NS
ES and NS
All
Nominal
9.7
5.4
0.12
0.136
17.2
21,100
0.18
Median
9.8
5.4
0.11
0.12
17.3
19000
0.16
90% U.I.
(4, 20)
(2, 12)
(0.05, 0.3)
(0.05, 0.3)
(15,20)
(8,000, 45,000)
(0.12,0.21)
63
-------
Table 20: Monte Carlo simulation of EF, YLL, average residential exposure, and
number of radon-induced fatal cancers with exposure factors fixed at nominal
values.3'13
II. Results
Etiologic fraction
Number of fatal lung
cancer deaths from
radon exposure
Smoking
Category
ES
ES and NS
ES and NS
Nominal
0.12
0.136
21,100
Median
0.12
0.14
21,000
90% U.I.
(0.05,0.3)
(0.06, 0.3)
(9,000, 50,000)
Radon concentration = 1.25 pCi/L, occupancy factor = 0.7, and equilibrium factor = 0.4.
b Same non-exposure parameter assumptions as in Table 19.
64
-------
F. Uncertainty in Extrapolating to Low Exposure Rates
The BEIR VI Committee found that the ERR/WLM increased with decreasing
exposure rate over the range of observation in the miner cohorts. The lowest exposure
rate classification considered by the BEIR VI committee was <0.5 WL, which in the
occupational context corresponds to annual exposures below 6 WLM. However, the
average residential exposure rate is estimated to be only 0.181 WLM y"1. Thus, applying
the miner derived models to residential radon exposures necessitates an extrapolation to
exposure rates well below the levels where there is useful epidemiological data upon
which to base those models. This creates a source of uncertainty that is difficult to
quantify, and we limit ourselves here to a qualitative discussion of the issue.
The increasing risk with decreasing exposure rate [inverse dose rate effect (IDRE)]
observed in the miners parallels evidence from radiobiology, indicating that for a given
dose of high-LET radiation, the effect is maximal at low dose rates. Typically, it is found
that for sufficiently low doses the response is independent of dose rate, but that at high
doses the response increases with decreasing dose rate. This characteristic behavior has
been observed for cell transformation, produced by neutron or alpha-particle irradiation
(Hill et al. 1982, Bettega et al. 1992). Moreover, such a dependence on dose rate has
been observed in studies of lung cancer induction by radon decay products in rats
(Chemaud et al. 1981, Cross et al. 1984).
The radiobiological evidence thus suggests that the ERR/WLM would be at least as
high at the low exposure rate conditions prevailing in homes as the ERR/WLM derived
from the miner studies. Indeed, there is no definitive evidence in BEIR VI that a low dose
rate plateau had been reached in the lowest exposure rate category (cf. Table 3-3 in BEIR
VI), so it could be argued that the risk in homes might be substantially underestimated by
the BEIR VI model, which implicitly assumes that the ERR/WLM has already reached its
maximum value at about 0.5 WL.
Biophysical explanations for the dose rate pattern described above for high-LET
radiation generally involve saturation of damage to a radiosensitive population of cells.
Because of this saturation phenomena, the effect of n hits to the same sensitive cell is
less than n times the effect of 1 hit. The response could be increased, however, if the
dose is protracted over a time scale comparable to the replenishment time of the sensitive
cell population. A modified version of this mechanism has been proposed by Brenner and
Sachs (2002) in which a small population of hypersensitive cells can be mutated by hits to
neighboring cells (bystander effect). At higher doses the bystander effect becomes
saturated and the process is dominated by direct hits to non-sensitive cells. It is also
postulated that a direct hit to a sensitive cell usually kills that cell. The competition among
these processes gives rise to a complex dose response relationship, in which the
response rises rapidly to a maximum, then decreases, before beginning a further linear
increase with dose. An IDRE would be present at intermediate dose levels.
65
-------
Alternative biologically based models fit to the epidemiological data can yield very
different predictions regarding the extrapolation to low exposure rates. Moolgavkar and
colleagues have proposed a 2-stage model of carcinogenesis in which cells first undergo
a single mutation that puts them in a precancerous, or intermediate state (Moolgavkar and
Knudson 1981, Moolgavkar and Luebeck 1990). The pool of intermediate cells may then
expand under the influence of cancer "promoters". Finally, a second mutational event can
occur in an intermediate cell, which divides uncontrollably to form a malignant tumor.
Applying this modeling approach to the analysis of the combined effects of cigarette
smoking and radon on lung cancer incidence in the Colorado Plateau miners, Luebeck et
al. (1999) concluded that radon acted mainly as a promoter of lung cancer and that it was
its promoting activity which produced the observed IDRE. Further calculations with the
two-stage model indicated that the ERR/WLM would not plateau with decreasing exposure
rate, as expected from the findings discussed previously, but would peak and then fall off.
It was also projected that the risks from residential radon exposure would be about 2 or 4
times lower than projected by the BEIR VI model, for smokers and never-smokers,
respectively.
Two issues might be raised with respect to the conclusions derived from the two-
stage model. First, there is the problem of basing the analysis solely on the Colorado
Plateau miner data, with all its uncertainties in exposure estimation and the
incompleteness of smoking information required for the analysis. Second, it attributes the
IDRE to a promotional mechanism when there is only sketchy evidence that alpha-particle
radiation acts as a cancer promoter, but there is ample evidence that alpha radiation is a
mutagen and that the mutagenic effect exhibits an inverse dose rate dependence.
Bogen (1997) has proposed a variant of the two-stage model, which projects a
protective effect of radon over a range of exposure rates, as suggested by the ecological
studies of Cohen discussed earlier. At this point, all such models must be regarded as
highly speculative. Only a more complete mechanistic understanding of alpha-particle
induced carcinogenesis or more definitive epidemiologic data on the variation of lung
cancer incidence with radon levels in homes can resolve the issue of exposure rate
extrapolation.
66
-------
G. Sensitivity Analysis of Risk Estimates to Assumptions about Health Effects
from Exposures to Radon
The Monte Carlo simulation quantified uncertainties related to exposure factors and
many of the parameters that were used for modeling excess relative risks. This section
investigates the sensitivity of our risk estimates to assumptions about some factors that
had not been accounted for in the Monte Carlo simulation. Examined here are: first, the
sensitivity of our risk estimates to parameters that would differentiate risks for subgroups
such as ES, NS, and children; and second, the dependency of the estimates on
assumptions about the relationship between relative risks and time-since-exposure.
Let us first assume that ERRs are accurately represented by the submultiplicative
scaled BEIR VI concentration model. Using pws and pES to denote risk coefficients for NS
and ES, we have:
ERR (a)= pws (w5_14 + Q15_24 w15_24 + Q25+ w25+) $(a) for NS
ERR (a)= PES (w5.14 + Q15.24 w15.24 + Q25+ w25+) $(a) for ES
where pES = 0.9p, and pws = 2p, and p = 0.0634 (see equation 5 in Section VLB. 2).
Since most of the miners were ES, it is likely that the ratio (PES/P) is very close to 0.9. In
contrast, there was much less data on NS, and as discussed in Section VII. B. 3, there may
be an extra factor of two uncertainty in the ERR for NS compared to that for the general
population. Table 21 shows the risk per WLM and EFfor pws = 0.0634 and pws = 0.254,
corresponding to 0.5 and 2 times the nominal value for the scaled concentration model
(PNS = 0.127). For these calculations, time-since-exposure and attained age parameters
were set to nominal values. Estimates of risk per WLM and EF for NS are proportional to
the NS risk coefficient. For example, doubling the risk coefficient for NS (from 2p to 4p)
doubles the (NS) estimates of risk per WLM (from 1.7xlQ-4 to 3.3x1 0'4) and EFfrom (0.26
to 0.53). The effect on risk estimates for the entire population would naturally be much
smaller: for NS risk coefficients of 2p to 4p, the risk per WLM would range from 5.4xlO"4 to
6.3x 10'4 and the EF would range from 0.1 4 to 0.1 6. Setting the NS risk coefficient to p
would result in about an 8% reduction in the (overall) risk per WLM and EF estimates.
67
-------
Table 21: Dependence of the risk per WLM and EF estimates on the NS risk
coefficient
Estimate
Risk per WLM
(10-4)
EF
Smoking
Status
NS
All
NS
All
NS Risk Coefficient3
PWS = P
0.8
4.9
0.13
0.12
PNS = 2P
1.7
5.4
0.26
0.14
PNS = 4P
3.3
6.3
0.53
0.16
a(3 = 0.0634 is the risk coefficient for the scaled concentration model. (3NS is the risk coefficient for NS.
Regarding the effect of childhood exposures, we defined pc to be the exposure response
parameter for exposures received before one's 18th birthday. Thus,
ERR (a)= pc (w5_14c + Q15_24
+ P (W5.14,A + Q 15-24
Q25+ w25+J $
+ Q25+ W25+A)
where the subscript c denotes exposures received before the 1 8th birthday, and the
subscript^ denotes exposures received after the 18th birthday. The estimated risk per
WLM from childhood exposures (exposures received before the 18th birthday) is
proportional to pc. For pc = p = 0.0634, the estimated risk per WLM from childhood
exposures is about 5.6xlO"4.
Table 22 shows the risk per WLM and EF (for lifetime exposures) for pc = 0.0317
and pc = 0.127, corresponding to 0.5 and 2.0 times the nominal value for the risk
coefficient p = 0.0634 . For these calculations, time-since-exposure and attained age
parameters were again set to nominal values. Here, doubling the risk coefficient for
children (from p to 2p) would increase the estimates of risk per WLM (from 5.4xlO"4 to
6.7x1 0'4) and EF (from 0.14 to 0.17) by about 24%. Setting the childhood risk coefficient
to 0.5p would result in about a 12% reduction in the (overall) risk per WLM and a similar
reduction in the EF.
68
-------
Table 22: Dependence of the risk per WLM and EFestimates on the childhood risk
coefficient (pc)
Estimate
Risk per WLM
(10-4)
EF
Childhood Risk Coefficient3
pc = 0.5p
4.7
0.12
PC=P
5.4
0.14
PC = 2P
6.7
0.17
(3 = 0.0634 is the risk coefficient for the scaled concentration model. The childhood risk coefficient, (3C, is the risk
coefficient for exposures before the 18th birthday.
Finally, consider the sensitivity of the risk estimates to assumptions about the
dependence of relative risk on time-since-exposure. For the scaled-concentration model,
the relative risk (for a given attained age) plateaus - at 51% of the maximum value - after
25 years from time of exposure (625+= .51). However, the risk model can be generalized
to incorporate the possibility that these relative risks continue to decline for time-since-
exposures greater than 25 y. Suppose
ERR (a)= P (w5_14 + 6
15-24 vv 15-24
'25-34 "25-34
035+ w35+) 4)(a)
where 025.34 and 035+ are time-since-exposure parameters for intervals 25 through 34 years
or 35 years and greater. This is equivalent to the formulation used for our scaled-
concentration model if 025.34 = 035+ . As shown in Table 23, if 035+ were reduced by 50%
so that 0+ = 0.5 x 0 = 0.255, estimated risks would be about 20% smaller. The
35+
2534
results of this sensitivity analysis are well within the range of plausible risk values based
on results from the Monte Carlo simulation. Thus, although the scaled-concentration
model does not incorporate all plausible ways in which risks depend on time-since-
exposure, this particular "model" uncertainty does not seem to dominate other
uncertainties that were quantified.
69
-------
Table 23: Dependence of the estimated risk per WLM and EF estimates on
assumptions on how relative risks fall off with time-since-exposure.
Estimate
Risk per WLM
(10-4)
EF
Smoking
Status
NS
ES
All
NS
ES
All
Time-since-exposure coefficient3
635+ = 0.5 x 62534
1.3
7.8
4.3
0.21
0.10
0.11
°35+ ~ 0 25-34
1.7
9.7
5.4
0.26
0.12
0.14
' 025-34 (equals 0.51) and 035H
greater.
are time-since-exposure coefficients for the intervals 25 through 34 y and 35y or
70
-------
APPENDIX A: AGE-SPECIFIC, EVER-SMOKING PREVALENCE ESTIMATES
The calculation of gender- and age-specific ES estimates for 1990 was
accomplished in three steps. The first step was to extrapolate white male and female
prevalence estimates, obtained from the National Institutes of Health (NIH), for calendar
years 1987 and 1988 to calendar year 1990. These NIH estimates were derived from
results from six NHIS surveys (DHHS 1997), and were calculated for each of 17 different
birth cohorts that range from 1885-89 to 1965-69. The second step adjusted these
estimates using data from OSH and 1990 census data to obtain prevalence estimates for
the entire male and female populations (across all races). Finally, smoothing splines were
used to estimate the prevalence at each age. Details follow.
The first two steps of the process are illustrated in Tables A1 and A2. The third
and fourth columns of the tables give the ES prevalence for whites estimated from six
NHIS surveys for 1987 and 1988. We extrapolated the ES prevalence for each age
group to obtain estimates in calendar year y by assuming a constant rate of change as
follows:
Here p(y) denotes the prevalence for a birth cohort in calendar year y, and y0 denotes the
last year for which we have NIH smoking prevalence estimates specific to that age group.
For our purposes, y = 1990.
For example, the extrapolated white male ES prevalence for ages 20.5 to 25.5 y in
1990, is:
0.3705 = 0.361 1 (0.361 1 / 0.3565)2
To see this, note that for this age group, the last year for which we have NIH smoking
prevalence estimates is 1 988. Then since y0 = 1 988, p(y0 ) = 0.361 1 , p(y0 -1 ) = 0.3565,
and y = 1 990, the result follows from the equation.
Next, OSH and 1 990 census data were used to obtain prevalence estimates for the
entire male and female populations (across all races). From OSH, the ES prevalence in
1990 was 58.7% for males and 42.3% for females. The 1990 census data allowed us to
combine the ES prevalence estimates in the fifth column of Tables A1 and A2. Weighted
averages of the 17 prevalence estimates, equal to 58.77% for males and 46.04% for
females, were obtained using weights equal to the proportion of males and females (from
the 1990 U.S. census) in each of the 17 age groups. Prevalence estimates for each age
group were then obtained using the following formula:
71
-------
0.587}
P> =P^g formates
(0.423} . , .
P< =[04604JP^-for femdleS
Here, pijWhites denotes the ES prevalence for white males or females in the ith age group,
and p, denotes the ES prevalence for the entire population (all races). For example, the
ES prevalence for males of ages 30.5 to 35.5 y is:
05285=
.5877
Finally, smoothing splines were used to obtain the age-specific ES prevalences
given in Table A3. We accomplished this using a Newton-Raphson iterative procedure
(see Hastie et a/. 1990) for fitting the logits of the adjusted prevalence estimates in Tables
A1 and A2. This involved the application of the MATLAB spine toolbox procedure "spaps"
(de Boor 1998) to the logits, where the logits were input as functions of the midpoints (in
years) of the corresponding age intervals; these midpoints are equal to 23, 28, ..., 103.
Other inputs for this procedure were initial weights equal to the proportion of males or
females alive in each age group (from 1990 census data), and tolerances set to 0.001 for
females and 0.0003 for males.
72
-------
Table A1: Ever-smoking prevalence estimates for males by age group.
Cohort
(Birth year)
1965-69
1960-64
1955-59
1950-54
1945-49
1940-44
1935-39
1930-34
1925-29
1920-24
1915-19
1910-14
1905-09
1900-04
1895-99
1890-94
1885-89
Age (years) —
on July 1,
1990
20.5-25.5
25.5-30.5
30.5-35.5
35.5-40.5
40.5-45.5
45.5-50.5
50.5-55.5
55.5-60.5
60.5-65-5
65.5-70.5
70.5-75.5
75.5-80.5
80.5-85.5
85.5-90.5
90.5-95.5
95.5-100.5
100.5-105.5
Ever-smoking prevalence (%)
1987
35.65
45.41
52.91
59.17
66.25
71.46
73.29
74.24
76.77
76.18
74.72
71.72
66.80
59.67
NA
NA
NA
Whites
1988
36.11
45.53
52.91
59.17
66.31
71.20
73.02
73.78
76.39
75.60
74.13
70.87
65.75
NAC
NA
NA
NA
1990a
37.05
45.77
52.91
59.17
66.43
70.68
72.48
72.87
75.64
74.45
72.96
69.20
63.70
56.38
50.56
38.97
39.09
Adjusted13
1990
37.00
45.72
52.85
59.10
66.35
70.60
72.40
72.78
75.55
74.36
72.88
69.62
63.62
56.31
50.50
38.92
39.04
aExtrapolated from 1987 and 1988 data, as discussed in text.
bAdjusted so that weighted average of age-grouped prevalence estimates equals OSH prevalence estimate
of58.7%.
cPrevalence estimates for cohorts born before 1900 were extrapolated using regression on
logarithmically transformed prevalence data.
73
-------
Table A2: Ever-smoking prevalence estimates for females by age group.
Cohort
(Birth year)
1965-69
1960-64
1955-59
1950-54
1945-49
1940-44
1935-39
1930-34
1925-29
1920-24
1915-19
1910-14
1905-09
1900-04
1895-99
1890-94
1885-89
Age (years) —
on July 1,
1990
20.5-25.5
25.5-30.5
30.5-35.5
35.5-40.5
40.5-45.5
45.5-50.5
50.5-55.5
55.5-60.5
60.5-65-5
65.5-70.5
70.5-75.5
75.5-80.5
80.5-85.5
85.5-90.5
90.5-95.5
95.5-100.5
100.5-105.5
Ever-smoking prevalence (%)
1987
37.78
46.51
50.37
47.87
51.78
55.77
54.39
53.00
50.24
46.34
42.69
34.86
26.31
16.64
NA
NA
NA
Whites
1988
38.23
46.60
50.41
48.04
51.83
55.60
54.19
52.78
49.93
45.92
42.08
33.96
25.55
NAC
NA
NA
NA
1990a
39.15
46.78
50.49
48.38
51.93
55.26
53.79
52.34
49.32
45.09
40.89
32.23
24.10
14.98
10.62
7.00
6.70
Adjusted13
1990
35.97
42.98
46.39
44.45
47.71
50.77
49.42
48.09
45.31
41.43
37.56
29.61
22.14
13.76
9.75
6.43
6.15
aExtrapolated from 1987 and 1988 data, as discussed in text.
bAdjusted so that weighted average of age-grouped prevalence estimates equals OSH prevalence estimate
of42.3%.
cPrevalence estimates for cohorts born before 1900 were extrapolated using regression on
logarithmically transformed prevalence data.
74
-------
Table A3: Smoothed age-specific ES prevalence estimates for males and females.
Age
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
Males
0.2966
0.3105
0.3252
0.3406
0.3567
0.3732
0.3900
0.4070
0.4239
0.4405
0.4568
0.4725
0.4876
0.5023
0.5165
0.5304
0.5441
0.5575
0.5709
0.5844
0.5979
0.6117
0.6253
0.6386
0.6513
0.6632
0.6739
0.6836
0.6922
0.6998
Females
0.3075
0.3169
0.3276
0.3392
0.3516
0.3644
0.3774
0.3902
0.4025
0.4141
0.4245
0.4334
0.4409
0.4469
0.4513
0.4542
0.4556
0.4561
0.4561
0.4563
0.4573
0.4595
0.4627
0.4668
0.4714
0.4764
0.4815
0.4864
0.4909
0.4946
Age
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
Males
0.7064
0.7120
0.7167
0.7208
0.7242
0.7272
0.7299
0.7324
0.7348
0.7372
0.7398
0.7425
0.7453
0.7478
0.7500
0.7515
0.7524
0.7525
0.7518
0.7506
0.7487
0.7462
0.7431
0.7394
0.7350
0.7298
0.7240
0.7174
0.7101
Females
0.4974
0.4991
0.4997
0.4994
0.4983
0.4966
0.4943
0.4915
0.4882
0.4844
0.4803
0.4757
0.4707
0.4653
0.4595
0.4533
0.4468
0.4399
0.4327
0.4250
0.4168
0.4082
0.3990
0.3891
0.3783
0.3667
0.3541
0.3406
0.3265
Age
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
Males
0.7021
0.6934
0.6841
0.6741
0.6634
0.6523
0.6406
0.6285
0.6159
0.6030
0.5898
0.5763
0.5626
0.5487
0.5346
0.5205
0.5062
0.4919
0.4776
0.4632
0.4490
0.4348
0.4207
0.4067
0.3929
0.3793
0.3659
0.3527
0.3397
Females
0.3118
0.2966
0.2813
0.2658
0.2504
0.2351
0.2201
0.2055
0.1914
0.1778
0.1649
0.1526
0.1411
0.1303
0.1202
0.1107
0.1019
0.0938
0.0862
0.0792
0.0727
0.0667
0.0611
0.0560
0.0513
0.0470
0.0430
0.0394
0.0360
75
-------
APPENDIX B: SMOOTHING THE BEIR VI RELATIVE RISK FUNCTIONS
As described in Part III, the BEIR VI preferred models specify that the excess
relative risk (relative risk -1) depends on time-since-exposure, attained-age, and either
rate of exposure (concentration) or duration according to the formula:
ERR = p (w5.14 + Q15.24 w15.24 + 6
25+
The 6-parameters detail how relative risk depends on time-since-exposure, and 4)age
describes the dependency on attained age. For almost all residential exposures, \z is
equal to 1 in the concentration model; \z is equal to 13.6 in the duration model for attained
ages greater than 40 y. p is constant for either model, and the effective exposure,
weff = w5_14 + 615_24 w15_24 + 625+ wzs+. The effective exposure is a continuous function of
attained age. However in the BEIR VI models, ERR is discontinuous at attained ages 55
y, 65 y, and 75 y because of discontinuities in the attained age function 4). For attained
age categories <55 y, 55-65 y, 65-75 y, and > 75 y, corresponding values for 4) are 1 ,
0.57, 0.34, and 0.28 for the duration model and 1 , 0.65, 0.38, and 0.22 for the
concentration model.
We smoothed the modeled ERR by using splines (Fritsch and Carlson, 1980) to
smooth the attained age component 4). We did this by finding a monotonic spline with
nodes at ages 40 y, 50 y, 55 y, 65 y, 75 y, 80 y, and 90 y for which the integral of 4) was
preserved for intervals 50-80 y, 55-65 y, and 65-75y. Results are given in Table B1 .
76
-------
Table B1: Spline smoothed values for cj)a, from the BEIR VI concentration and
duration models.
Age (y)
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Duration
model
1.0000
0.9882
0.9574
0.9147
0.8672
0.8219
0.7731
0.7130
0.6455
0.5743
0.5033
0.4364
0.3772
0.3297
0.2976
0.2847
Concentration
model
1.0000
0.9888
0.9599
0.9197
0.8752
0.8329
0.7892
0.7377
0.6808
0.6211
0.5610
0.5031
0.4498
0.4038
0.3674
0.3432
Age (y)
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
Duration
model
0.2831
0.2819
0.2811
0.2805
0.2801
0.2796
0.2790
0.2781
0.2768
0.2749
0.2585
0.2224
0.1800
0.1447
0.1300
Concentration
model
0.3277
0.3154
0.3055
0.2971
0.2896
0.2822
0.2742
0.2647
0.2530
0.2383
0.2124
0.1743
0.1344
0.1028
0.0900
' This parameter describes the dependency of the modeled excess relative risks on attained age.
77
-------
APPENDIX C: NOTATION AND FORMULAS
1) Summary risk measures
RWLM = risk per WLM (working level month)
EF = etiologic fraction
YLL = average years of life lost per radon-induced lung cancer death
2) Basic quantities and functions
h(x) = baseline lung cancer death rate at age x - the probability that a person,
exposed to baseline levels of radon and alive at age x, will die from lung
cancer before attaining age (x + dx) is equal to h(x)dx
S(x) = baseline survival function - the fraction of live-born individuals in a
population exposed to baseline radon levels that is expected to survive to
agex.
g(x) = radon exposure rate (WLM / y) at age x.
gb(x) = background radon exposure rate set to 0.181 WLM / y.
ge(x) = radon exposure rate in excess of background
g(x) - gb(x)
r = lifetime risk of a premature lung cancer death due to excess radon exposure
w(x) = cumulative radon exposure (WLM) at age x.
e(x, 9e) = excess relative risk (ERR) at age x due to an excess radon exposure - the
probability that a person, exposed to excess levels of radon and alive at age
x, will die from lung cancer before attaining age (x + dx) is equal to
h(x) [1 + e(x, ge)] dx.
S(x, 9e) = fraction of live-born individuals in a population that are expected to survive to
age x with excess radon exposure = ge(a) at ages a < x
1.05 = the presumed sex ratio at birth (male-to-female)
3) Subscripts
Subscripts for h(x), S(x), e(x), and S(x, g) are used to denote specific populations or
subpopulations:
pop nonstationary U.S. population that includes males, females, ES, and NS.
sta stationary population with fixed percentages of male ES, male NS, female ES and
female NS
m subpopulation of males; r subpopulation of females
ES subpopulation of ES; ws subpopulation of NS
subpopulation of male ES, etc.
78
-------
4) Baseline lung cancer death rates used in the text
h(x) for a (sometimes generic) population at age x.
hpop(x) for the U.S. population that includes males, females, ES, and NS.
hsta(x) for a stationary population that includes males, females, ES, and NS.
A)m(x) for males; h/^x) for females
hES(x) for ES; hNS(x) for NS
hmES(x) for male ES, etc.
5) Baseline survival functions used in the text
S(x), Ssta(x), Sm(x), S/x), SES(x), SNS(x),SmfS(x), etc.
6) Survival functions modified by an excess radon exposure rate (gj
S(x, ge), Ssta(x, ge), Sm(x, ge), S/x, ge), SES(x, ge), SNS(x, ge), SmES(x, ge), etc.
We often assume that the excess radon exposure rate is equal to a very small
constant, A, so that ge = A. The modified survival function, omitting subscripts,
would be S(x, A).
7) Smoking prevalence
The fraction of live-born individuals that is expected to smoke at least 100
cigarettes during their lifetime is denoted by:
pbjrth for the U.S. population (both males and females).
pbirth, m for males; pbirth, f for females
The fraction of individuals at age x that have smoked at least 100 cigarattes is
denoted by:
p(x) for the population
pm(x) for males; p/x) for females
8) Excess relative risk (ERR)
eES(x, ge) for ES
ews(x, ge) for NS
eES(x, ge) = (0.9/2.0) ews(x, ge)
79
-------
9) Survival function formulas
S(x) = (1 .05 Sm(x) + SXx)) / 2.05
Sm(x) = pbirth, m Sm_ES(x) + (1 - pwrt/lf J Sm>NS(x)
SES(x) = [1 .05 pwrt/lf m Sm_ES(x) + pbirth_ f Sf_ ES(x)] I pbirth
10) Adjustments to survival functions for smoking status
Notation:
AES(x) = SES(x) I S(x); ANS(x) = SNS(x) I S(x)
AmES(x) = Sffl> ES(x) / Sm(x); A,ES(x) = S, ES(x) / S/x)
Am,NS(x) = Sm, NS(x) / Sm(x); AfiNS(x) = Sf: NS(x)
Equations:
11) Adjustments to survival functions due to radon exposure in excess of
background
Notation:
B(x, g) = S(x, g) I S(x)
6m,ES(x, ge) = Sm,ES(x, gg) I Smfs(x)
Formulas:
B«< .ES (X>9e)= GXP
12) Exposure to radon
gb(x) = (1.25 pCi/L) [(0.4) (1Q-2WL(pCi/L)-1)] [(0.7)((365.25)(24)/170 WLM/(WL-y))]
80
-------
13) Life expectancy
Notation:
L = Life expectancy (note: this is similar to the notation used in BEIR IV).
Lm, Lf for males and females
LES, LNs for ES and NS
/ for male ES.
*-m, ES
Formulas:
(' ~ Ptxrihw / ' ^~m JUS
(' Pbitthf f ' L-fNS
LE$ = > • vz> ' P tMhtf ' Lm _£S + p tMfjf • ^-f£S
Lflls=1.05'(1- p^, ; • Lw ^ + (1 - p^hf) • Lms
L = C/.05'Lm +Lf)/2.05
14) Combining summary measures from different populations
Notation: Subscripts denote subpopulations based on gender and smoking.
RWLMf =
-(1.05- p^f • L^s • RWLM^£S + p^f • L,fs • RWLMff5) / (2. 05 LBS)
= [1.05(1 -p^JL^s- RWLM^^ +(1- p^f)Lf,ls • RWLMfJts]/ (2.05
RWLM = (1. 05 • Lm • RWLMm + L, • RWLMf) / (2. 05 L)
EF = (1.05- R^^ -EFm + Rba^lifnf • EFf) / (2.05
81
-------
APPENDIX D: LUNG CANCER RISKS BY RADON LEVEL AND SMOKING STATUS
Table D1 presents estimates of the risk of lung cancer death by radon level for NS,
current smokers and the general population. Estimates are subject to considerable
uncertainties as discussed in Sections VI.H, VI.I and Chapter VII. In particular, note that
the risk models in the BEIR VI report did not specify excess relative risks for current
smokers. Because the excess mortality rates are in some instances very large, baseline
mortality rates were adjusted for the excess risk due to radon exposure.
Table D1: Lifetime risk of lung cancer death by radon level for never smokers,
current smokers, and the general population.
Radon Level3
(pCi/L)
20
10
8
4
2
1.25
0.4
Lifetime Risk of Lung Cancer Death
from Radon Exposure in Homes
Never Smokers Current Smokers General Population
3.6%
1.8%
1.5%
0.7%
0.4%
0.2%
0.1%
26.3%
15.0%
12.0%
6.2%
3.2%
2.0%
0.6%
10.5%
5.6%
4.5%
2.3%
1 .2%
0.7%
0.2%
Assumes constant lifetime exposure in homes at these levels.
b Estimates are rounded to the nearest tenth of a percent. No indication of uncertainty should be inferred from this
practice.
82
-------
REFERENCES
Bergen, AW, N Caporaso. Cigarette smoking. J. Natl. Cancer Inst. 91, 1365-1375, 1999.
Bettega, D, P Calzolari, GN. Chiorda and L Tallone-Lombardi. A transformation of C3H
10T% cells with 4.3 MeV alpha particles at low doses: effects of single and fractionated
doses. Radial Res. 131, 66-71, 1992.
Bogen, KT. Do U.S. county data disprove linear no-threshold predictions of lung cancer
risk for residential radon? — A preliminary assessment of biological plausibility. Hum.
Eco/. Risk Assess. 3, 157-186, 1997.
Brenner, DJ and RK Sachs. Do low dose-rate bystander effects influence radon risks?
Inter. J. Radial Biol., in press, 2002.
Cavallo, A. The radon equilibrium factor and comparative dosimetry in homes and mines.
Radial Prot. Dosim. 92, 295-298, 2000.
Cavallo, A. Reply to D. Krewski et al. Comment on protection of residential radon lung
cancer risks: the BEIR VI risk models - (letter to the editor). Radial Prot. Dosim. 102,
373, 2002.
(CDC) Centers for Disease Control, Morbidity and Mortality Weekly Report, Atlanta, Ga. -
CDC Surveillance Summaries, 43 (SS-3). Surveillance for selected tobacco use behaviors
- United States, 1900-94. Nov. 18, 1994. Also available at
http://www.cdc.gov/nccdphp/osh/adstat1.htm
(CDC) Centers for Disease Control. Morbidity and Mortality Weekly Report, Atlanta, Ga.
Cigarette smoking among adults—United States 1993. 1995.
(CDC) Centers for Disease Control. Morbidity and Mortality Weekly Report, Atlanta, Ga.-
CDC Surveillance Summaries, 49 (SS-10) Youth tobacco surveillance - United States
1998-1999. Oct. 13,2000.
Chemaud, J, R Perraud, R Masse and J Lafuma. Contribution of animal experimentation
to the interpretation of human epidemiological data, pp. 551-557 in: Proceedings of the
International Conference on Radiation Hazards in Mining: Control, Measurement and
Medical Aspects. M. Gomez, ed., Kingsport Press Inc., Kingsport, TN, 1981.
Cohen, BL. A test of the linear no-threshold theory of radiation carcinogenesis.
Environmental Res. 53, 193-220, 1990.
Cohen, BL. Test of the linear no-threshold theory of radiation carcinogenesis for inhaled
radon decay products. Health Phys. 68, 157-174, 1995.
83
-------
Cohen, BL. Response to Lubin's proposed explanations of our discrepancy. Health Phys.
75, 18-22, 1998.
Cohen, BL. Response to criticisms of Smith et al. Health Phys. 75, 23-28, 1998a.
Cross, FT. Invited commentary: residential radon risks from perspective of experimental
animal studies. Am. J. Epidemiol. 140, 333-339, 1994.
Cross, FT, RF Palmer, GE Dagle, RE Busch and RL Buschbom. Influence of radon
daughter exposure rate, unattachment fraction, and disequilibrium on occurrence of lung
tumors. Radial Prot. Dosimetry7, 381-384, 1984.
de Boor, C. Spline Toolbox User's Guide. The MathWorks, Inc., Natick, MA, 1998.
(DHHS) Department of Health and Human Services. A report of the Surgeon General:
Reducing the Health Consequences of Smoking. 25 Years of Progress. U.S. Government
Printing Office, Washington DC, 1989.
(DHHS) Department of Health and Human Services. SEER Cancer Statistics Review
1973-1990. National Institutes of Health 92-2789. 1995.
(DHHS) Department of Health and Human Services. Public Health Service and National
Cancer Institute. Burns, DM , L Garfinkel, and JM Samet, editors. Changes in Cigarette-
Related Disease Risks and Their Implication for Prevention and Control. NIH 97-4213.
Smoking and Tobacco Control Monograph 8. U.S. GPO, Bethesda, MD, 1997.
Elkind, MM. Radon-induced cancer: a cell-based model of tumorigenesis due to
protracted exposures. Int. J. Radial Biol. 66, 649-653, 1994.
(EPA) Environmental Protection Agency. Risk Assessments Methodology, Environmental
Impact Statement: NESHAPS for Radionuclides. Background Information Document,
Volume I. EPA 520/1-89-005. U.S. Environmetal Protection Agency, Washington, DC,
1989.
(EPA) Environmental Protection Agency. Technical Support Document for the 1992
Citizen's Guide to Radon. EPA 400-R-92-011. U.S. Environmental Protection Agency,
Washington, DC, 1992.
(EPA) Environmental Protection Agency. Exposure Factors Handbook. EPA/600/P-
95/002/a-c. U.S. Environmental Protection Agency, Washington, DC, 1997.
Field, RW, BJ Smith and CF Lynch. Ecologic bias revisited, a rejoinder to Cohen's
response to "Residential 222Rn exposure and lung cancer: Testing the linear no-threshold
theory with ecologic data." Health Phys. 75, 31-33, 1998.
84
-------
Fritsch, FN and RE Carlson. Monotonic piecewise cubic interpolation. SIAM J. Numer.
Anal. 17,238-246, 1980.
George, AC and AJ Breslin. The distribution of ambient radon and radon daughters in
residential buildings in the New Jersey-New York area, pp. 1272-1292 in: Natural
Radiation Environment III, TF Gessell and WM Lowder, eds., U.S. DOE, 1980.
Goldsmith, JR. The residential radon-lung cancer association in U.S. Counties: a
commentary. Health Physics. 76,553-557,1999.
Greenland, S and J Robins. Review and commentary: Conceptual problems in the
definition and interpretation of attributable fraction. Am. J. Epidemiol. 128, 1185-97, 1988.
Greenland, S and J Robins. Invited commentary: Ecologic studies - biases,
misconceptions and counterexamples. Am. J. Epidemiol. 139, 747-760, 1994.
Hastie, TJ and RJ Tibshirani. Generalized Additive Models. Chapman and Hall, New
York, 1990, p.96-98.
Hill, CK, FJ Buonoguro, CP Myers, A Han and MM Elkind. Fission-spectrum neutrons at
reduced dose rate enhance neoplastic transformation. Nature 298, 67-68, 1992.
Hopke, PK, B Jensen, CS Li, N Montassier, P Wasiolek, AJ Cavallo, K Gatsby, RH
Socolow, AC James. Assessment of the exposure to and dose from radon decay products
in normally occupied homes. Environ. Sci. Technol. 29, 1359-1364, 1995.
Hornung, RWand TJ Meinhardt. Quantitative risk assessment of lung cancer in U.S.
uranium miners. Health Phys. 52, 417-430, 1987.
(ICRP) International Commission on Radiological Protection. Protection Against
Radon-222 at Home and at Work. ICRP Publication 65. Pergamon, Tarrytown, NY, 1993.
Israeli, M. Deposition rates of Rn progeny in houses. Health Phys. 49, 1069-1083, 1985.
James, AC. Dosimetry of radon and thoron exposures: Implications for risks from indoor
exposures. Pp. 167-198 in: Indoor Radon and Lung Cancer: Reality or Myth, FT Cross
ed., Battelle Press, Columbus, Ohio, 1992.
James, AC, A Birchall, and GH Akabani. Comparative Dosimetry of BEIR VI Revisited.
Radiation Protection Dosimetry. Submitted, 2003.
Krewski, D, JH Lubin, JM Samet, PK Hopke, AC James, KP Brand. Projection of
residential radon lung cancer risks: The BEIR VI risk models - (Letter to the editor).
Radial Prot. Dosim. 102, 371-3, 2002.
85
-------
Lubin, JH. On the discrepancy between epidemiologic studies in individuals of lung cancer
and residential radon and Cohen's ecologic regression. Health Phys. 75, 4-10, 1998.
Lubin, JH. Rejoinder: Cohen's response to "On the discrepancy between epidemiologic
studies in individuals of lung cancer and residential radon and Cohen's ecologic
regression." Health Phys. 75, 29-30, 1998a.
Lubin, JH. The influence of residential radon exposure on the estimation of exposure-
response trends for lung cancer in underground miners exposed to radon. Radiat. Res.
150,259-261, 1998.
Lubin, J. DISCUSSION: Indoor radon and risk of lung cancer. Radiat. Res. 151, 105-106,
1999.
Lubin, JH and JD Boice, Jr. Lung cancer risk from residential radon: meta-analysis of eight
epidemiologic studies. J. Natl. Cancer Inst. 89, 49-57, 1997.
Lubin, JH, JD Boice Jr., C Edling, RW Hornung, G Howe, E Kunz, RA Kusiak, HI
Morrison, EP Radford, JM Samet, M Tirmarche, A Woodward, SX Yao and DA Pierce.
Radon and Lung Cancer Risk: A Joint Analysis of 11 Underground Miners Studies.
National Institutes of Health, National Cancer Institute. NIH Publication No. 94-3644. U.S.
Department of Health and Human Services, Washington, DC, 1994.
Luebeck, EG, WF Heidenreich, WD Hazelton, HG Paretzke and SH Moolgavkar.
Biologically based analysis of the data for the Colorado uranium miners cohort: Age, dose
and dose-rate effects. Radiat. Res. 152, 339-351, 1999.
Malarcher, AM, J Schulman, LA Epstein, MJ Thun, P Mowery, B Pierce, L Escobedo, and
GA Giovino. Methological Issues in Estimating Smoking-attributable Mortality in the
United States. Amer. J. Epidem. 152, 573-584, 2000.
Moolgavkar, SH and A Knudson. Mutation and cancer: a model for human
carcinogenesis. J. Natl. Cancer Inst. 66, 1037-1052, 1981.
Moolgavkar, SH and EG Luebeck. Two-event model for carcinogenesis: Biological,
mathematical, and statistical considerations. Risk Analysis 10, 323-341, 1990.
Marcinowski, F, RM Lucas and WM Yeager. National and regional distributions of airborne
radon concentrations in U.S. homes. Health Phys. 66, 699-706, 1994.
(NAS) National Academy of Sciences. Health Risks of Radon and Other Internally
Deposited Alpha-Emitters: BEIRIV. National Academy Press, Washington, DC, 1988.
(NAS) National Academy of Sciences. Comparative Dosimetry of Radon in Mines and
Homes. National Academy Press, Washington, DC, 1991.
86
-------
(MAS) National Academy of Sciences. Health Effects of Exposure to Radon: BEIR VI
National Academy Press, Washington, DC 1999.
(NCHS) National Center for Health Statistics. Vital Statistics Mortality Data, Detail, 1989.
NTIS order number of datafile tapes: PB92-504554. U.S. Department of Health and
Human Services, Public Health Service, Hyattsville, Md., 1992.
(NCHS) National Center for Health Statistics. Vital Statistics Mortality Data, Detail, 1990.
NTIS order number of datafile tapes: PB93-504777. U.S. Department of Health and
Human Services, Public Health Service, Hyattsville, Md., 1993a.
(NCHS) National Center for Health Statistics. Vital Statistics Mortality Data, Detail, 1991.
NTIS order number of datafile tapes: PB93-506889. U.S. Department of Health and
Human Services, Public Health Service, Hyattsville, Md., 1993b.
(NCHS) National Center for Health Statistics. U.S. Decennial Life Tables for 1989-91. Vol.
1, No. 1. DHHS, PHS-98-1150-1. United States Life Tables. Public Health Service,
Washington, DC, 1997.
Nelson, CB, JS Puskin and DJ Pawel. Adjustments to the baseline lung cancer mortality
for radon-induced lung cancers in the BEIR VI risk models. Radiation Research, 156,
220-221,2001.
(NJDEP) New Jersey Department of Environmental Protection. Highlights of the Statewide
Scientific Study of Radon, New Jersey Department of Environmental Protection, Trenton,
1989.
Porstendorfer, J and A Reineking. Radon: Characteristics in Air and Dose Conversion
Factors. Health Phys. 76, 300-305, 1999.
Puskin, JS. Mortality rates with average county radon levels. Health Phys. (in press),
2003.
Rogers, RG and Powell-Griner EP. Life expectancies of cigarette smokers and
nonsmokers in the United States. Social Science in Medicine 32, 1151-59, 1991.
(SAB) Science Advisory Board. An SAB Advisory: Assessing Risks from Indoor Radon.
An advisory prepared by the Radiation Advisory Committee (RAC) on Proposed EPA
Methodology for Assessing Risks from Indoor Radon (EPA-SAB-RAC-ADV-99-010),
Washington, DC, 1999
(SAB) Science Advisory Board. An SAB Report: Assessment of Risks in Homes. Review
of the Draft Assessment of Risks from Radon in Homes by the Radiation Advisory
Committee (RAC) of the Science Advisory Board (EPA-SAB-EC-00-010), Washington,
DC, 2000
87
-------
Samet, JM, DR Pathak, MV Morgan, DB Coultas, DS James and WC Hunt. Silicosis and
lung cancer risk in underground uranium miners. Health Phys. 61, 745-752, 1994.
Smith, BJ, RW Field and CF Lynch. Residential 222Rn exposure and lung cancer: Testing
the linear no-threshold theory with ecologic data. Health Phys. 75, 11-17, 1998.
Thompson, DE, K Mabuchi, E Ron, M Soda, M Tokunaga, S Ochikubo, S Sugimoto,
T Ikeda, M Terasaki, S Izumi and DL Preston. Cancer incidence in atomic bomb survivors.
Part II: Solid Tumors, 1958-1987. Radial Res. 137, S17-S67, 1994.
Tsang, AM and NE Klepeis. Results Tables from a Detailed Analysis of the National
Human Activity Pattern Survey (NHAPS) Response. Draft report prepared for the U.S.
Environmental Protection Agency by Lockheed Martin, 1996.
U.S. Census. 1990 Census data: Database C905TF3C1 from:
http://venus.census.goV/cdrom/lookup/1040151632
Xuan, XZ, JH Lubin, JY Li and WJ Blot. A cohort study in southern China of workers
exposed to radon and radon decay products. Health Phys. 64, 120-131, 1993.
Zang , EA and EL Wynder. Differences in lung cancer risk between men and women:
examination of the evidence. J. Natl. Cancer Inst 88, 183-192, 1996.
88
------- |