EPA Radiogenic Cancer Risk
Models and Projections for
the U.S. Population
radiation P
LAR(D,e) = I M(D,e,a)*S(a)/S(e)da
United States
Environmental Protection Agency
Office of Radiation and Indoor Air
Radiation Protection Program (6608J)
EPA 402-R-11-001
April 2011
-------
EPA 402-R-11-001
EPA Radiogenic Cancer Risk Models and
Projections for the U.S. Population
April 2011
U.S. Environmental Protection Agency
Office of Radiation and Indoor Air
1200 Pennsylvania Ave., NW
Washington, DC 20460
-------
-------
PREFACE
This document presents new U.S. Environmental Protection Agency
(EPA) estimates of cancer incidence and mortality risks due to low doses of
ionizing radiation for the U.S. population, as well as their scientific basis. It
replaces the 1994 EPA report, Estimating Radiogenic Cancer Risks, often
referred to as the "Blue Book." In 1999, the Agency applied the 1994 Blue Book
contents, metabolic models, and usage patterns to publish Federal Guidance
Report 13 (FGR-13), Cancer Risk Coefficients for Environmental Exposure to
Radionuclides. FGR-13 includes coefficients for calculating estimates of cancer
risk for over 800 radionuclides. It is anticipated that results presented here will be
applied to update the radionuclide risk coefficients in the next revision of FGR-13.
For the most part, estimates of radiogenic risk in this document are
calculated using models recommended in the National Academy of Sciences
report: Health Risks from Exposure to Low Levels of Ionizing Radiation, BEIR VII
Phase 2 (MAS 2006). The MAS report, often referred to as BEIR VII, was
sponsored by EPA and several other federal agencies. As in BEIR VII, models
are provided here for estimating risk as a function of age at exposure, age at risk,
gender, and cancer site, but a number of extensions and modifications to the
BEIR VII approach have been implemented.
In response to requests by the Office of Radiation and Indoor Air (ORIA),
the Radiation Advisory Committee (RAC) of the Science Advisory Board (SAB)
has formally reviewed the scientific basis and methodology for this report. In
2008, the SAB completed an Advisory in response to the draft White Paper:
Modifying EPA Radiation Risk Models Based on BEIR VII. In the "White Paper,"
ORIA proposed many of the methods for calculating risks which were eventually
adopted for this report. Then in December, 2008, ORIA submitted for SAB review
the draft EPA Radiogenic Cancer Risk Models and Projections for the U.S.
Population. The RAC review was released on January 5, 2010. In the cover letter
to Administrator Jackson, Dr. Deborah Swackhamer, Chair, SAB, and Dr. Bernd
Kahn, Chair, RAC, wrote that the 2008 draft was "impressively researched [and]
based on carefully considered concepts" and "scientifically defensible and
appropriate." They also provided comments and suggestions. In her letter of
March 3, 2010, Lisa P. Jackson provided responses to the RAC comments and
suggestions.
This report was prepared by David J. Pawel and Jerome S. Puskin of
EPA's Office of Radiation and Indoor Air (ORIA). The authors gratefully
acknowledge reviews by: Owen Hoffman, lulian Apostoaei, John Trabalka, and
David Kocher of SENES Oak Ridge, Inc.; Mary Clark, Neal Nelson, and Lowell
Ralston of ORIA.
-------
Contact information for the authors is:
U.S. Environmental Protection Agency
Office of Radiation and Indoor Air (6608J)
Washington, DC 20460
Email address: pawel.david(S)epa.gov
-------
ABSTRACT
Background. This document presents new U.S. Environmental Protection
Agency (EPA) estimates of cancer incidence and mortality risks due to low doses
of ionizing radiation for the U.S. population, as well as their scientific basis. It
replaces the 1994 EPA report Estimating Radiogenic Cancer Risks, often
referred to as the "Blue Book." In 1999, the Agency applied the 1994 Blue Book
contents, metabolic models, and usage patterns to publish Federal Guidance
Report 13 (FGR-13), Cancer Risk Coefficients for Environmental Exposure to
Radionuclides. FGR-13 includes coefficients for calculating estimates of cancer
risk for over 800 radionuclides. It is anticipated that results presented here will be
applied to update the radionuclide risk coefficients in the next revision of FGR-13.
For the most part, estimates of radiogenic risk in this document are calculated
using models recommended in the National Academy of Sciences' report: Health
Risks from Exposure to Low Levels of Ionizing Radiation, BEIR VII Phase 2 (MAS
2006). The MAS report, often referred to as BEIR VII, was sponsored by EPA
and several other federal agencies. As in BEIR VII, models are provided here for
estimating risk as a function of age at exposure, age at risk, gender, and cancer
site.
A number of extensions and modifications to the BEIR VII approach have
been implemented. First, BEIR VII focused on the risk from low-LET radiation
only, whereas risks from high-LET radiations are also addressed here. Second,
this document goes beyond BEIR VII in providing estimates of risk for basal cell
carcinomas, kidney cancer, bone sarcomas, and cancers from prenatal
exposures. Third, a modified method is employed for estimating breast cancer
mortality risk, which corrects for temporal changes in breast cancer incidence
and survival. Fourth, an alternative model is employed for estimating thyroid
cancer risk, based primarily on a report from the National Council on Radiation
Protection and Measurements (NCRP). Fifth, EPA's central estimate of risk for
many cancer sites is a weighted arithmetic mean of values obtained from the two
preferred BEIR VII models for projecting risk in the U.S. population, rather than a
weighted geometric mean, as employed in BEIR VII. Finally, this report provides
a somewhat altered and expanded analysis of the uncertainties in the cancer risk
estimates, focusing especially on estimates of risk for whole-body irradiation and
for specific target organs.
Results. Summary risk coefficients are calculated for a stationary population
(defined by 2000 U.S. vital statistics). Numerically, the same coefficients apply
for a cohort exposed throughout life to a constant dose rate. For uniform whole-
body exposures of low-dose gamma radiation to the entire population, the cancer
incidence risk coefficient (Gy"1) is 1.16x10"' (5.6x10"2 to 2.1x10"1), where the
numbers in parentheses represent an estimated 90% confidence interval. The
corresponding coefficient for cancer mortality (Gy"1) is about one-half that for
incidence: 5.8x10"2 (2.8x10"2 to 1.0x10"1).
IV
-------
CONTENTS
Preface ii
Abstract iv
List of Tables vii
List of Figures ix
Acronyms and Abbreviations x
Executive Summary 1
1. Introduction 5
2. Scientific Basis for Cancer Risk Models 6
2.1 Biological Mechanisms 6
2.1.1 Biophysical interactions 6
2.1.2 Carcinogenesis 7
2.1.3 Radiogenic carcinogenesis 8
2.1.4 Extrapolation of low-LET risks to low doses
and low dose rates 10
2.1.5 Low dose phenomena 11
2.2 Epidem iology 13
3. EPA Risk Projections for Low-LET Radiation 16
3.1 Introduction 16
3.2 BEIR VII Risk Models 16
3.3 Risk Models for Kidney, Central Nervous System, Skin, and
Other "Residual Site" Cancers 24
3.4 Risk Model for Thyroid Cancer 31
3.5 Calculating Lifetime Attributable Risk 33
3.6 Dose and Dose Rate Effectiveness Factor 35
3.7 EAR and ERR LAR Projections for Cancer Incidence 35
3.8 ERR and EAR Projections for Cancer Mortality 36
3.9 Data on Baseline Rates for Cancer and All-Cause Mortality 38
3.10 Combining Results from ERR and EAR Models 40
3.10.1 BEIR VII approach 40
3.10.2 EPA approach 41
3.10.3 A justification for the weighted AM 42
3.11 Calculating Radiogenic Breast Cancer Mortality Risk 45
3.12 LAR by Age at Exposure 47
3.13 Summary of Main Results 58
3.14 Comparison with Risk Projections from ICRP and UNSCEAR 67
3.14.1 ICRP risk models 67
3.14.2 UNSCEAR risk models 69
-------
4. Uncertainties in Projections of LAR for Low-LET Radiation 72
4.1 Introduction 72
4.2 Sources of Uncertainty Quantified in this Report 73
4.3 "One at a Time" Uncertainty Analysis 75
4.4 Monte Carlo Approach for Quantifying Uncertainties in LAR 85
4.4.1 Monte Carlo method 85
4.4.2 Non-sampling sources of uncertainty 86
4.4.3 Bayesian analysis for sampling variability 93
4.4.4 Approach for other cancers 96
4.5 Results 98
4.6 Comparison with BEIRVII 104
4.6.1 Quantitative uncertainty analysis in BEIRVII 104
4.6.2 Comparison of results 105
4.7 Conclusions 106
5. Risks from Higher LET Radiation 108
5.1 Alpha Particles 108
5.1.1 Laboratory studies 108
5.1.2 Human data 109
5.1.3 Nominal risk estimates for alpha radiation 119
5.1.4 Uncertainties in risk estimates for alpha radiation 120
5.2 Lower Energy Beta Particles and Photons 120
6. Risks from Prenatal Exposures 125
7. Radionuclide Risk Coefficients 127
8. Noncancer Effects at Low Doses 128
Appendix A: Baseline rates for Cancer and All-Cause Mortality
and Computational Details for Approximating LAR 130
Appendix B: Details of Bayesian Analysis 135
References 142
Glossary 159
VI
-------
LIST OF TABLES
Table Page
3-1 BEIRVII risk model cancer sites 18
3-2 Summary of BEIRVII preferred risk models 19
3-3 Parameter values for preferred risk models in BEIRVII 21
3-4 Projection of LAR (Gy~1) for brain and CMS cancers for three alternative
ERR models 31
3-5 Estimated ERR/Gy and effect modifiers for age at exposure and TSE 32
3-6 Summary of SEER thyroid relative and period survival rates 33
3-7 EAR and ERR model projections of LAR for cancer incidence for a
stationary population and a population based on 2000 census data 36
3-8 Age-averaged LAR for cancer mortality based on a stationary population 38
3-9 Changes in age-averaged cancer rates for the SEER 13 registries 40
3-10 Comparison of EPA and weighted GM method for combining EAR and
ERR LAR projections for incidence 42
3-11 Female breast cancer cases and 5-y relative survival rates by age of
diagnosis for 12 SEER areas, 1988-2001 46
3-12 LAR for cancer incidence by age at exposure 54
3-13 LAR for cancer mortality by age at exposure 56
3-14 LAR for cancer incidence for lifelong and childhood exposures rates 58
3-15 LAR projections for incidence 59
3-16 LAR projections for mortality 60
3-17 Comparison of EPA and FGR-13 LAR projections for incidence 62
3-18 Comparison of EPA and FGR-13 LAR projections for mortality 63
3-19 Sex-averaged LAR projections for incidence and mortality 64
3-20 Comparison of EPA and BEIR VII LAR calculations 65
3-21 LAR incidence and mortality projections for a population based on 2000
census data 66
3-22 Comparison of ICRP (2007) and EPA risk model parameter values for
solid cancers 68
3-23 Comparison of EPA and ICRP (2007) sex-averaged projections of
incidence for chronic exposures to the U.S. population 69
VII
-------
LIST OF TABLES, continued
Table Page
3-24 Summary of UNSCEAR (2008) risk models for solid cancer incidence
and leukemia mortality 70
3-25 EPA and UNSCEAR (2008) cancer incidence risk projections from
chronic exposures to the U.S. population 71
4-1 Uncertainty factors for non-sampling sources of uncertainty 86
4-2 EPA projections and uncertainty distributions for cancer incidence LAR 99
4-3 Percentage of uncertainty in LAR for cancer incidence attributable to
sampling, risk transport, and DDREF 102
4-4 EPA projection and uncertainty distributions for cancer incidence for
childhood exposures for selected sites 103
4-5 EPA and BEIR VII uncertainty intervals for LAR of solid cancer Incidence... 106
5-1 Lung cancer mortality and RBE 118
B-1 Prior distributions for ERR model parameters 136
B-2 Comparison of posterior distributions for ERR linear dose response
parameter with estimates in BEIR VII 138
VIM
-------
LIST OF FIGURES
Figure Page
2-1 Dose response for low-LET y-rays and high-LET neutrons or a-particles 9
3-1 Age-time patterns in radiation-associated risks for solid cancer incidence
excluding thyroid and nonmelanoma skin cancer 20
3-2 ERR for leukemia for age-at-exposure = 20 and TSE = 10 23
3-3 ERR and EAR for exposures at low doses and/or dose rates by TSE for
three different ages at exposure 24
3-4 Comparison of two ERR models for brain and CMS cancers with the
residual site ERR model 30
3-5 Examples of distributions which might be used for the risk transport
weight parameter 44
3-6 LAR for incidence by age at exposure for solid cancer sites 49
3-7 LAR for mortality by age at exposure for solid cancer sites 51
3-8 LAR by age at exposure for leukemia for incidence and mortality 53
3-9 LAR for all cancers combined by age at exposure for exposures at low
doses and/or dose rates for incidence and mortality 53
4-1 Dependence of LAR for selected cancer sites for both lifelong and
childhood exposures (age < 15) on ERR model parameter values 78
4-2 Dependence of LAR for selected cancer sites for lifelong exposures on
the linear dose response parameter (ft), DDREF, and the ERR model
weight parameter (w) 82
4-3 Subjective probability density function for DDREF 88
5-1 Cumulative fraction of total dose vs. secondary electron kinetic energies
for a variety of low-LET radiations calculated using the method of Burch .... 123
5-2 Cumulative fraction of total dose vs. secondary electron energies for a
variety of slow and fast initial electron energies calculated by the
Monte Carlo track structure method 124
A-1 Baseline incidence and mortality rates for specific cancer sites 131
B-1 Posterior distributions for ERR for selected cancer sites 139
B-2 Posterior distributions for the age-at-exposure parameter 140
B-3 Posterior distributions for the attained age parameter 141
IX
-------
LIST OF ACRONYMS AND ABBREVIATIONS
ATB At the Time of the Bombings
BCC Basal Cell Carcinoma
BEIR VII Health Risks from Exposure to Low Levels of Ionizing Radiation
BEIR VII Phase 2
Cl Confidence Interval
DDREF Dose and Dose Rate Effectiveness Factor
DEF Dose Effectiveness Factor
DREF Dose Rate Effectiveness Factor
DSB Double Strand Break
EAR Excess Absolute Risk
EPA Environmental Protection Agency
ERR Excess Relative Risk
eV Electron Volt
FGR-13 Federal Guidance Report 13
GM Geometric Mean
GSD Geometric Standard Deviation
Gy Gray
ICRP International Commission on Radiological Protection
IREP Interactive RadioEpidemiological Program
LAR Lifetime Attributable Risk
LET Linear Energy Transfer
LNT Linear No -Threshold
LQ Linear-Quadratic
LSS Life Span Study
MAS National Academy of Sciences
NCHS National Center for Health Statistics
NCI National Cancer Institute
NCRP National Council on Radiation Protection and Measurements
NIOSH National Institute for Occupational Safety and Health
ORIA Office of Radiation and Indoor Air
RBE Relative Biological Effectiveness
REF Radiation Effectiveness Factor
RERF Radiation Effects Research Foundation
RR Relative Risk
SCC Squamous Cell Carcinoma
SEER Surveillance, Epidemiology, and End Results
Sv Sievert
TSE Time Since Exposure
UF Uncertainty Factor
Ul Uncertainty Interval
UNSCEAR United Nations Scientific Committee on the Effects of Atomic
Radiation
WLM Working Level Months
-------
EXECUTIVE SUMMARY
The U.S. Environmental Protection Agency, as part of its responsibilities
for regulating environmental exposures and its Federal Guidance role in radiation
protection, develops estimates of risk from low-level ionizing radiation.1
This document presents new EPA estimates of cancer incidence and
mortality risk coefficients pertaining to low dose exposures to ionizing radiation
for the U.S. population, as well as their scientific basis. The "dose" refers to the
amount of energy deposited by the radiation in a unit mass of tissue, expressed
in units of gray (Gy). The "risk" is generally defined to be the probability of a
health effect (i.e., a cancer or a cancer death), and the risk per unit dose is called
a "risk coefficient." Where there is no possible confusion, "risk coefficients" and
"ionizing radiation" will usually be referred to here, simply, as "risks" and
"radiation." For the most part, risk estimates are calculated using models
recommended in the National Academy of Sciences' BEIR VII Report (MAS
2006), which was sponsored by EPA and several other federal agencies. The
models and risk estimates presented here replace those published in a 1994
report, Estimating Radiogenic Cancer Risks, with some modifications in 1999
(EPA 1994, 1999a, 1999b).
As in BEIR VII, models are provided for estimating risk as a function of
age at exposure, age at risk, gender, and cancer site, but a number of extensions
and modifications to the BEIR VII approach have been implemented. First, BEIR
VII focused on the risk from low-LET radiation only, whereas risks from high-LET
radiations are also addressed here. Second, this document goes beyond BEIR
VII in providing estimates of risk for basal cell carcinomas, kidney cancer, bone
sarcomas, and cancers from prenatal exposures. Third, a modified method is
employed for estimating breast cancer mortality risk, which corrects for temporal
changes in breast cancer incidence and survival. Fourth, an alternative model is
employed for estimating thyroid cancer risk, based primarily on a report from the
National Council on Radiation Protection and Measurements (NCRP). Fifth,
EPA's central estimate of risk for many cancer sites is a weighted arithmetic
mean of values obtained from the two preferred BEIR VII models for projecting
risk in the U.S. population, rather than a weighted geometric mean, as employed
in BEIR VII. Finally, this report provides a somewhat altered and expanded
analysis of the uncertainties in the cancer risk estimates, focusing especially on
estimates of risk for whole-body irradiation and for specific target organs.
Underlying the risk models is a large body of epidemiological and radio-
biological data. In general, results from both lines of research are consistent with
a linear, no-threshold dose (LNT) response model in which the risk of inducing a
See http://www.epa.gov/radiation for further information on EPA's radiation protection activities
and Federal Guidance function.
-------
cancer in an irradiated tissue by low doses of radiation is proportional to the dose
to that tissue.
The most important source of epidemiological data is the Life Span Study
(LSS) of the Japanese atomic bomb survivors, who received an acute dose of
radiation, mostly in the form of y-rays, with a small admixture of neutrons. The
LSS study has important strengths, including: a nearly instantaneous exposure,
which can be pinpointed in time; a large, relatively healthy exposed population
encompassing both genders and all ages; a wide range of radiation doses to all
organs of the body, which can be estimated reasonably accurately; and detailed
epidemiological follow-up for about 50 years. The precision of the derived risk
estimates is higher than all other studies for most cancer sites; nevertheless it is
limited by errors in dosimetry and sampling errors. The sampling errors are often
quite large for specific cancer types, and the uncertainties are even larger if one
focuses on a specific gender, age at exposure, or time after exposure. Another
important uncertainty is the transfer of site-specific cancer risk estimates to the
U.S. population, based on results obtained on the LSS population, for sites with
substantially different baseline incidence rates.
In addition to the LSS, other epidemiological studies provide important
information about radiogenic cancer risks. These include studies of medically
irradiated patients and groups receiving occupational or environmental expo-
sures. For thyroid and breast cancers, risk estimates are based on pooled
analyses of the LSS and medically irradiated cohorts. While studies on popu-
lations exposed occupationally or environmentally have, so far, been of limited
use in quantifying radiation risks, they can provide valuable insight into the risks
from chronic exposures.
Summary risk coefficients are calculated for a stationary population
(defined by 2000 U.S. vital statistics) rather than a population with an age
distribution of the actual U.S. population. Numerically, the same coefficients
apply for a cohort exposed throughout life to a constant dose rate. This puts the
radiation risk estimates derived here on a comparable basis to risk estimates for
chemicals derived from lifetime animal exposure experiments. For uniform
whole-body exposures of low-dose gamma radiation to the entire population, the
cancer incidence risk coefficient (Gy~1) is 1.16x10"1 (5.6 x10"2 to 2.1x10"1), where
the numbers in parentheses represent an estimated 90% confidence interval.
The corresponding coefficient for cancer mortality (Gy~1) is about half that for
incidence: 5.8x10 (2.8x10"2 to 1.0x10"1). For perspective, the average individual
receives about 1 mGy each year from low-LET natural background radiation, or
about 75 mGy, lifetime. The average cancer incidence and mortality risks from
natural background radiation are then estimated to be about 0.87% and 0.44%,
respectively.
The estimated risks are significantly higher for females than for males:
1.35x10'1 Gy1 vs. 9.55x10'2 Gy'1 (incidence) and 6.9x10'2 Gy1 vs. 4.7x10'2 Gy'1
-------
(mortality), respectively. Estimates of risk per unit dose differ widely among
cancer sites. For females, these are largest for lung and breast cancers, which
together account for about 44% (incidence) and 50% (mortality) of the risk from
uniform whole-body radiation. For males, risks per unit dose are largest for colon
and lung cancers, accounting for about 29% (incidence) and 40% (mortality) of
the risk for all cancer sites.
Radiogenic risks for childhood exposures are of special interest. Doses
received from ingestion or inhalation are often larger for children than adults, and
the risks per unit dose are substantially larger for exposures during childhood
(here defined as the time period ending at the 15th birthday) than from exposures
later in life. For children, the estimated risks from uniform whole-body radiation
for cancer incidence are 2.0x10"1 Gy"1 (males) and 3.3x10"1 Gy"1 (females) with
90% uncertainty intervals: 7.7x10"2 to 3.6x10"1 Gy"1 (males) and 1.2x10"1 to
5.5x10"1 Gy"1 (females). The corresponding estimated risks for mortality are
8.5x10"2 Gy"1 (males) and 1.5x10"1 Gy"1 (females). There is generally much more
uncertainty in the estimated risks from childhood exposures than in the risks for
the entire population. A-bomb survivors who were children at the time of the
bombings (ATB) still have substantial years of life remaining in which cancers are
to be expressed. Further follow-up will provide more statistical precision and
greater clarity as to how these risks vary many decades after the exposure.
For ingestion or inhalation of radionuclides that concentrate in individual
organs, the risk for those specific sites may predominate. In this context, it is
important to recognize that the percent uncertainties for site-specific risk
coefficients are generally greater than the coefficient for uniform, whole-body
irradiation; this is largely due to the smaller number of cancers for specific sites in
the LSS and to uncertainties in how radiogenic risks for specific cancer sites in
the U.S. might differ from those in a Japanese population of A-bomb survivors.
Cancer sites with large relative changes in the calculated lifetime risk
(about 2-fold or more) from EPA's previous estimates published in Federal
Guidance Report 13 (FGR-13) (EPA 1999b) include: kidney, liver, female lung,
and female bladder (increased); and female colon (decreased). For both males
and females, the estimated risk for all cancers combined increased by about
35%. For mortality, there was a notable change in estimated risk for cancers of
the female colon (decreased), and female lung (increased). In general, the new
EPA mortality estimates do not differ greatly from those in FGR-13; remarkably,
for all sites combined, the estimates changed by less than 2% for both males and
females.
One issue in radiation risk assessment is how to extrapolate risk estimates
derived from data on relatively high acute exposures in case of the LSS cohort to
low dose, or chronic exposure situations, which are of greatest interest to EPA.
Many subjects in the LSS cohort did receive very low doses, but there is
inadequate statistical power to quantify risk below about 0.1 Gy. This is about
100 times the annual whole-body, low-LET dose to an average individual from
-------
natural background. Thus, the question is how to extrapolate from an observed
risk due to an instantaneous dose of 0.1 Gy or more to an extrapolated risk from
a chronic exposure of = 1 mGy per year.
Efforts have been made to integrate information gathered from radiation
biology and epidemiology into a theoretical framework that would allow reliable
risk projections at dose rates approaching natural background. Radiation is
known to induce mutagenic damage to the cell's DMA. Due to clustering of
ionizations produced by low-LET as well as high-LET radiation, this damage is
often complex, involving two or more breaks with concomitant base damage all
within a few nanometers in the DMA molecule. This argues against a threshold
for radiation-induced carcinogenesis and in favor of a linear dose-response
relationship at low doses. Experimental studies have uncovered novel low-dose
phenomena, which may modulate the dose-response relationship at low doses.
However, the relevance of these findings to human carcinogenesis remains
unclear, and epidemiological studies of cancer induction in cohorts receiving
fractionated or chronic exposures have so far been broadly consistent with LNT
predictions. The BEIR VII Committee unequivocally recommended continuing
adherence to the LNT approach. EPA also finds strong scientific support for LNT,
while acknowledging that new research might conceivably lead to revisions in the
future.
Aside from the case of radon (which is not in the scope of this report),
human data on risks from high-LET radiation (a-particles) are much more limited
than for low-LET. For most cancer sites, risk coefficients for a-particles are based
on a relative biological effectiveness (RBE) factor of 20 estimated from laboratory
experiments; i.e., the organ-specific risk coefficients are set equal to 20 times
that for y-rays. Epidemiological results on patients injected with an a-emitting
radionuclide are consistent with an RBE of 20 for liver cancer, relative to the
LSS, but an RBE of only about 2 for leukemia. An analysis of data on plutonium
workers at the Mayak plant in the former Soviet Union also yielded an estimated
a-particle RBE of roughly 20 for lung cancer (relative to the LSS), but there is
considerable uncertainty in the doses delivered to sensitive cells in the lung for
the Mayak worker cohort. In the case of bone cancer, low-LET data on humans is
very sparse, and the bone cancer risk model employed here is derived from data
on patients injected with 224Ra.
Radiation is known to induce mutations in animal germ cells, but
hereditary effects in humans have not been demonstrated. Nevertheless, genetic
risks from low dose radiation exposure can be estimated based on animal
studies. These estimates are generally lower than for cancer induction. There is
also evidence that radiation at moderate doses can induce health effects such as
cataracts and cardiovascular disease, and these effects may not have a
threshold. However, unlike the case of radiogenic cancer and hereditary effects,
there is, at present, no direct evidence nor a strong theoretical basis for such
effects at lower/chronic exposures.
-------
1. Introduction
The 1994 report, Estimating Radiogenic Cancer Risks (EPA 1994),
presented EPA estimates of site-specific risks cancer incidence and mortality
associated with low doses of ionizing radiation. (For brevity, the modifier
"ionizing" will usually be omitted in the remainder of this report.) Primarily, the
calculated risks were derived from models recommended by the International
Commission on Radiological Protection (Land and Sinclair 1991), based on
analysis of epidemiological data on Japanese atomic bomb survivors. While
focusing mainly on a quantitative assessment of uncertainties in these estimates,
a subsequent report also made minor adjustments in EPA's cancer risk
estimates, reflecting changes in U.S. vital statistics (EPA 1999a). Finally, the
methodology developed in the above reports was used in Federal Guidance
Report No. 13 (FGR-13) to derive cancer risk coefficients for low level internal
and external exposures to a set of over 800 radionuclides (EPA 1999b).
In 2006, the National Research Council of the National Academy of
Sciences (NAS) released the BEIR VII report (NAS 2006), which reviewed recent
evidence pertaining to the health risks from low-level, low linear energy transfer
(LET) radiation. The BEIR VII Committee developed models for calculating the
risks of radiogenic cancers, based on updated information on the A-bomb
survivors, as well as other data. In this report, we employ the BEIR VII models to
arrive at revised estimates of radiogenic risks for most cancer sites. BEIR VII risk
estimates were derived for low doses of y-rays with typical energies between
about 0.1 and 10 MeV, with a brief discussion of possible enhancement of risk for
more densely ionizing electrons and photons. Although the main focus here is, as
in BEIR VII, on low-LET risks, we extend the evaluation of cancer risks to high-
LET radiation (a-particles) and outline a biophysical approach to estimating risks
from low energy photons and electrons. We also present risk models and
estimates for prenatal exposures, and for kidney, bone, and non-melanoma skin
cancers, which are not covered in BEIR VII.
Deviations from the BEIR VII approach are made for averaging the two
types of models used to project risk from the A-bomb survivors to the U.S.
population and for generating estimates of the risks of thyroid cancer and breast
cancer mortality. Finally, a quantitative uncertainty analysis is presented, which is
based on a different approach from that in BEIR VII and which incorporates some
additional sources of uncertainty.
This report is not intended to provide an exhaustive review of the scientific
basis for our risk models. For the most part, the reader is referred to BEIR VII
and other sources in the literature. We have attempted to highlight major sources
of uncertainty and, where pertinent, to include recently published information not
considered by the BEIR VII Committee.
-------
2. Scientific Basis for Cancer Risk Models
2.1 Biological Mechanisms
2.1.1 Biophysical interactions. By definition, ionizing radiation passing
through matter has sufficient energy to break chemical bonds and to remove
electrons from molecules. When this chemical damage occurs in the DMA of a
somatic cell, a mutation in the genetic material can result, ultimately leading to a
malignancy. The damage can be produced directly, when an ionizing particle
impacts the DMA, or indirectly, through the creation of free radicals in the cellular
medium, which diffuse and interact with the genetic material.
Only a tiny fraction of the free radicals produced in cells each day arise
from radiation; nevertheless, DMA damage by low-level radiation is not negligible.
This is because energy deposition events are often produced in clusters, which
can, in turn, produce double strand breaks (DSBs) and more complex damage in
DMA, involving multiple breaks and chemical modifications within a very
restricted portion of the double helix. Cellular repair processes are less capable
of repairing DSBs and complex damage than the simpler types of damage almost
always induced by isolated free radicals. This makes ionizing radiation unique
among environmental carcinogens. Even a single track of the radiation is capable
of producing complex damage sites, which, if misrepaired, can leave the cell with
a mutated gene that can be passed on to the cell's progeny. Depending on the
nature of the mutation, this may be one step in the formation of a malignancy. At
reasonably low doses the number of DSBs and sites of complex damage is
expected to be strictly proportional to dose (UNSCEAR 2000b, NCRP 2001, MAS
2006); this is the primary basis for the linear no-threshold (LNT) theory in which
the probability of inducing a cancer by radiation is proportional to dose with no
threshold below which there is no risk.
Some recent research has cast doubt on the LNT assumption, but the
BEIR VII Report concluded that these results in no way constituted compelling
evidence against LNT. Additional discussion of the issue will be found in sections
below.
The degree of clustering of ionizations, and therefore of the DNA damage,
depends on the type of radiation and its energy. This is reflected in the linear
energy transfer of charged particle radiation (LET), which is a measure of the
amount of energy deposited, per unit path length, as the particle passes through
a medium. Alpha particles emitted by the decay of unstable atomic nuclei have a
relatively high LET (= 100-200 keV/um) in aqueous media, producing a high
density of ionizations, leading to a high frequency of DSBs and clustered damage
sites in the DNA. Since this type of damage is more likely to be misrepaired,
high-LET radiation is more effective at causing mutations, cell transformation,
and cell death (NCRP 2001). This higher effectiveness per unit dose, relative to
some standard radiation (e.g., 60Co Y-raYs), is expressed in terms of a factor
-------
called the relative biological effectiveness2 (RBE) (see Section 5). Initially, 200
kVp X-rays were used as the reference; however, since current radiogenic
cancer risk estimates largely rest on studies of the Japanese atomic bomb
survivors, whose predominant exposure was from Y-ravs, it is now common to
use 60Co y-rays as the reference radiation.
Compared to a-particles, (3-particles and secondary electrons produced by
incident y-rays or medical X-rays typically have much lower linear energy transfer
(0.1-10 keV/um). The ionizations produced by energetic electrons are more
widely spaced, on average, but their production is a stochastic process in which
several ionizations can be created separated by a distance no greater than the
characteristic distance between adjacent DMA bases or between DMA strands.
Moreover, as electrons lose energy, the LET increases and closely spaced
ionizations become more frequent. Hence, clustered DMA damage is more likely
to be produced near the ends of the electron tracks.
X-rays and y-rays can travel appreciable distances through matter without
producing ionizations; however, they interact with atoms to produce energetic
secondary electrons, which behave identically to incident electrons of the same
energy. In aqueous media, over the incident photon energy range 0.1-10 MeV,
the predominant photon interaction is Compton scattering, a process in which an
incident photon transfers part of its energy to an atomic electron, creating a free
electron and a lower energy photon. The energy of a Compton electron is
positively correlated with the incident photon energy. Consequently, as the
incident photon energy is reduced within this energy range, a higher fraction of
the energy is dissipated in the form of lower energy (higher LET) electrons,
resulting in more complex DMA damage and, therefore, perhaps an increased
RBE. As the incident photon energy is reduced further, below 0.1 MeV,
photoelectric absorption becomes increasingly important compared to Compton
scattering, and the variation of LET with photon energy is no longer monotonic.
2.1.2 Carcinogenesis. Carcinogenesis is thought to be a multi-staged
process "initiated" by a mutation in a single cell. Before a malignancy can result,
however, additional mutations must accumulate. This process may be enhanced
by enlarging the pool of initiated cells (clonal expansion), which might be
triggered by the presence of a "promoter." After clonal expansion, more initiated
cells are available to undergo additional mutations, a process referred to as
"cancer progression." Particularly important may be those mutations that
2
Kocher et al. (2005) have introduced a quantity called the "radiation effectiveness factor" (REF)
to compare the cancer causing potency in humans of a specified type of radiation relative to
some standard. According to their definition, the REF is to be distinguished from measured RBEs
that may be used as a basis for estimating the REF, although the RBEs themselves may have
been measured for a different end-point or in a different species. Although it is important to keep
in mind that RBEs used for human risk estimation are generally extrapolated, and not directly
measured, we follow common practice here in applying the term RBE more broadly to include the
estimation of human radiogenic cancer risk.
-------
increase the probability of further mutations - e.g., those impairing DMA repair
processes. Eventually, a set of mutations may remove the essential controls over
cell division, resulting in a malignancy.
2.1.3 Radiogenic carcinogenesis. Over a period of decades, a con-
ceptual model of radiation carcinogenesis was built up from numerous studies
conducted at the molecular, cellular, tissue, and whole organism levels. In this
picture an ionizing track produces DMA damage through direct interaction with
the double helix or through the interaction of free radicals diffusing to the DMA
damage site, after being produced nearby. Misrepair of the DMA damage can
then lead to an initiated cell and, eventually, to a malignancy as outlined above.
The dose response for radiation carcinogenesis is then expected to have the
same mathematical form as that for radiation-induced mutations.
As shown in Figure 2-1, the dose response for the induction of mutations,
cell transformation, or carcinogenesis by low-LET radiation appeared to be linear
at low doses, curvilinear upward at higher doses until eventually becoming
concave downward at still higher doses. Mathematically, the initial portions of the
curve is expressed as a "linear-quadratic" (LQ) function of effect (E) vs dose (D).
E = 0.1 D + a2 D2 (2-1)
At low dose rates, the effect was found to increase linearly, with the same slope,
di, observed initially at high dose rates. The expected response at high doses is
therefore reduced by lowering the dose rate, which effectively removes the
quadratic term in Eq. 2-1.
As also shown in Figure 2-1, the dose-response for high-LET radiation,
appeared to be linear and independent of dose rate, except at rather high doses,
where the function flattens or even turns over. At the high doses, moreover, an
"inverse dose rate effect" may be observed in which the response is increased
when the dose rate is reduced.
Thus, at low doses and dose rates the dose-response for either low- or
high-LET radiation appears to be linear with no evidence of a threshold.
In the case of low-LET radiation, it was inferred that the passage of two
tracks close together in space and time increases the probability of misrepaired
damage, either because the damage produced is more complex or because the
repair machinery becomes partially saturated, reducing its effectiveness. It was
presumed that, at either low doses or low dose rates, only the damage produced
by single tracks is significant, and the response is simply proportional to dose. At
high dose rates, however, repair efficiency will decrease with increasing dose,
leading to the quadratic term in Eq. 2-1.
8
-------
120
AJphti particles
Neutrons
, ABSORBED DOSE (Gy)
Figure 2-1: Solid curves depict the classical dose-response curves for low-LET
y-rays and high-LET neutrons or a-particles. The dashed lines show the
expected response at low dose rates for each type of radiation. From UNSCEAR
1993, p. 698.
At low or moderate doses of high-LET radiation, the production of multiply-
damaged sites in DMA is dominated by single track events. The flattening or
downturn observed at high acute doses may reflect cell killing (NCRP 1980). An
alternative explanation has been proposed in which at any given time a
subpopulation of cells exists in a sensitive time window; spreading the dose out
more in time allows more cells to be hit while they are in that time window,
resulting in an enhanced response (Rossi and Kellerer 1986, Elkind 1994).
Downward curvature and an inverse dose rate effect can also result from the
"bystander effect" (Brenner and Sachs 2003), which will be discussed below.
Conclusions: Traversal of a cell nucleus by radiation can induce damage
to the cell's DMA, initiating the carcinogenic process. Since the damage pro-
duced by even a single track of ionizing radiation can sometimes be misrepaired,
a threshold for cancer induction would appear improbable unless there is a
mechanism for eliminating essentially all dividing cells with damaged DMA (e.g.,
through some kind of immune surveillance). A nearly foolproof screening
mechanism of this sort would seem to be ruled out, however, by the significant
rate of cancer incidence among people not exposed to high levels of radiation.
Under conditions of low doses or low dose rates, the effect of multiple
tracks is expected to be negligible, so the probability of a cell becoming initiated
is simply proportional to dose. This provides a mechanistic basis for the linear
no-threshold (LNT) model of carcinogenesis in which the probability of radiation
causing a cancer is proportional to dose, even at very low doses for which there
-------
is insufficient statistical power to detect any excess incidence of the disease in a
human population.
2.1.4 Extrapolation of low-LET risks to low doses and dose rates. As
discussed above, radiobiological data suggest that the probability of mutational
damage in a cell's DMA from an acute exposure to low-LET radiation can be
expressed as a linear-quadratic (LQ) function of dose (D)\
E = cclD+cc2D2 (2-1)
The linear term is assumed to reflect the effect of single tracks, the quadratic
term the added effect of two tracks traversing the cell close together in space and
time, or perhaps the saturation of repair mechanisms at higher doses. If doses
are delivered in a widely spaced temporal series of acute dose fractions, it is
expected that each dose fraction, Df, will produce an incremental effect,
£1/=a1D/+or2D^ (2-2)
If each fraction is made very small, the quadratic terms will be negligible, and the
overall summed effect will be linear with dose; i.e., E = a\D, where D = ££>/. A
chronic exposure can be thought of as a sequence of very small fractionated
exposures. It follows that if the dose rate from a chronic exposure is low enough
so that the interaction of multiple tracks can be neglected, then the effect will
again be simply given by E = a^D, where D is the total dose.
The effect per unit dose will be reduced in going from a large acute dose,
D, where the quadratic term is significant, to a low dose, where only the linear
term contributes. Overall the effect will be reduced by a Dose Effectiveness
Factor (DEF) = (a1+a2D}la1 = l+OD, where 6=a2la1. Likewise the estimated effect
per unit dose will be reduced by a Dose Rate Effectiveness Factor (DREF), when
a large acute dose is delivered chronically. Since the slope is the same (a/) at
low doses or dose rates, the DREF and the DEF are equal. Thus, according to
the LQ model, the extrapolation from a high acute dose to either a low dose or to
a low dose rate can be embodied into a single correction factor, the Dose/Dose
Rate Effectiveness Factor (DDREF).
It is presumed that the probability of carcinogenesis induced in an
organism from an exposure to radiation is proportional to the number of induced
mutations remaining after repair is complete. This has led scientists to model the
excess risk as a LQ function of dose for a relatively high acute dose, with a
reduction by a DDREF factor for low doses and dose rates. The DDREF for
carcinogenesis would be equal to that for the underlying process of radiation-
induced mutagenesis.
10
-------
Based on its review of radiobiological and epidemiological data, the
UNSCEAR Committee (UNSCEAR 1993; 2000b) concluded that any dose below
200 mGy, or any dose rate below 0.1 mGy/min (when averaged over about an
hour), should be regarded as low. Thus, according to the linear-quadratic model,
for these doses and dose rates, the risk per unit dose would be approximately
equal to the linear coefficient, cti.
2.1.5 Low dose phenomena. Much recent research in radiobiology has
focused on several new phenomena relating to the effects of low dose radiation,
including: (1) the adaptive response, (2) genomic instability, and (3) bystander
effects. These phenomena have raised questions about the reliability of the LNT
model for radiation carcinogenesis. They indicate that, at least under some
conditions, radiation may induce DMA damage, indirectly, by affecting non-
targeted cells, and that the processing of DMA damage by cells may be strongly
dependent on dose, even at very low doses.
Adaptive response. Under some conditions, it has been found that pre-
irradiating cells with an "adapting dose" of low-LET radiation (~10 mGy) reduces
the effects (e.g., chromosome damage, mutations, or cell transformation) of a
subsequent "challenge dose" of ~1 Gy. This has provided some support for the
suggestion that low-dose radiation may stimulate defense mechanisms, which
could be beneficial in preventing cancer or other diseases. Supporting this view
also have been studies in which the spontaneous transformation rates of certain
cells in culture have been reduced by exposure to very low level radiation
(Azzam et al. 1996, Redpath and Antoniono 1998). A subsequent study has,
however, shown a threshold for this "beneficial effect": suppression of trans-
formation disappeared when the dose rate was reduced below 1 mGy/day
(Elmore et al. 2008). Thus, even if this phenomenon occurs in vivo, it may not be
operative at environmental exposure levels.
Genomic instability. It has been found that irradiation of a cell can
produce some kind of change in that cell, not yet characterized, which increases
the probability of a mutation one or more cell divisions later (Morgan et al. 1996).
The relatively high frequency of inducing genomic instability implies that the
relevant target is much larger than a single gene, and there is evidence that, at
least in some cases, the phenomenon is mediated by radiation-induced
epigenetic changes rather than DMA damage (Kadhim et al. 1992, Morgan et al.
1996). The delayed mutations are typically simple point mutations, unlike other
mutations caused by radiation, which are typically deletions or other types of
chromosomal changes resulting from DSBs and more complex DMA damage
(Little et al. 1997).
Bystander effects. Contrary to the conventional picture, DMA damage in
a (bystander) cell can be induced by passage of an ionizing track through a
neighboring cell. The bystander effect can apparently be triggered by passage of
a signal through gap junctions (Azzam et al. 1998). Media transfer experiments
11
-------
have demonstrated that it can also be induced - although probably less
effectively (Mitchell et al. 2004) - by molecules leaking out into the extracellular
fluid (Mothersill and Seymour 1998, Lehnert and Goodwin 1998). It also appears
that the adaptive response and genomic instability may be induced in bystander
cells under some conditions (Coates et al. 2004, Kadhim et al. 2004, Tapio and
Jacob 2007). Recent evidence has also been found of bystander signals from
irradiated cells inducing apoptosis in neighboring transformed cells (Portess et al.
2007).
The preponderance of data regarding these effects has been obtained
from experiments on isolated cells. There is limited information on the occur-
rence of these effects in vivo, and no understanding of how they might modulate
risks at low doses. At first sight, it would appear that the adaptive response
should be protective, whereas bystander effects and genomic instability might
increase risk. Interpretation may be complicated, however, by the possibility for
triggering protective mechanisms in bystander cells, such as an adaptive
response or apoptosis of precancerous cells (Lyng et al. 2000, Portess et al.
2007, Tapio and Jacob 2007).
The BEIR VII Committee was not convinced that these effects would
operate in vivo in such a way as to significantly modify risks at low doses. It was
a consensus of the Committee that:
the balance of evidence from epidemiologic, animal and mechanistic
studies tend to favor a simple proportionate relationship at low doses
between radiation dose and cancer risk (BEIR VII, p. 14).
A similar conclusion was reached by another group of experts assembled by the
International Commission on Radiological Protection (ICRP 2005).
In contrast, the French Academy of Sciences issued a report that strongly
questioned the validity of the LNT hypothesis (Tubiana et al. 2005). The French
Academy report cited a paper by Rothkamm and Lobrich (2003) showing that
repair of DSBs, as measured by the disappearance of y-H2AX foci, was absent
or minimal at low doses, presumably leading to apoptosis of cells with DSBs. The
French Academy report claimed that this finding indicated that risks were greatly
overestimated at low doses. Recent studies have cast doubt on the significance
of this finding, however (Lobrich et al. 2005, Markova et al. 2007).
Conclusion. EPA accepts the recommendations in the BEIR VII and
ICRP Reports to the effect that there is strong scientific support for LNT and that
there is no plausible alternative at this point. However, research on low dose
effects continues and the issue of low dose extrapolation remains unsettled.
12
-------
2.2 Epidemiology
There is overwhelming evidence from epidemiological studies of irradiated
human populations that radiation increases the risk of cancer. Most important
from the standpoint of quantifying radiation risks is the Lifespan Study (LSS) of
atomic bomb survivors in Hiroshima and Nagasaki, Japan. These survivors
constitute a relatively healthy population at the time of exposure, including both
genders and all ages, with detailed medical follow-up for about half a century.
Extremely significant, also, is the wide range of fairly accurately known individual
radiation doses.
The LSS cohort shows an excess in various types of cancer, with the rates
increasing with increasing dose to the target organ. The data from the LSS are
adequate to serve as a basis for developing detailed mathematical models for
estimating risk as a function of cancer site, dose, age, and gender. However, due
to limitations in statistical power, it has not been possible to demonstrate and
quantify risk in the LSS at doses below about 100 mGy.
Epidemiological studies of medically irradiated cohorts provide strong
confirmation for the carcinogenic effects of radiation and some additional
information for generating risk estimates - in particular, for the bone, thyroid,
liver, and breast. Radiation risks have also been extensively studied in
occupationally exposed cohorts, but so far such studies - aside from those on
radon-induced lung cancers in underground miners - have not proved very
useful for actually quantifying risk. Major reasons for this failure have been: poor
dosimetry; low doses, leading to low statistical power; and potential confounding
by life-style factors or other occupational exposures. As discussed in a later
section, however, recent data on workers at the Mayak plutonium production
plant in the former Soviet Union may provide an improved basis for estimating
risks from inhaled a-emitters.
Although the epidemiological data on radiation-induced carcinogenesis
are extensive, calculated risks to members of the U.S. population from doses of
radiation typically received environmentally, occupationally, or from diagnostic
medical procedures suffer from significant sources of uncertainty. Among these
sources are: (1) errors in the epidemiological data underlying the risk models,
including sampling errors, errors in dosimetry, and errors in disease ascertain-
ment; (2) uncertainties in how risks vary over times longer than the period of
epidemiological follow-up; (3) uncertainties in "transporting" risk estimates to the
U.S. population from a study population (e.g., the LSS cohort), which may differ
in its sensitivity to radiation; (4) differences in the type of radiation, or its energy,
between the epidemiological cohort and the target U.S. population; and (5)
uncertainty in how to extrapolate from moderate doses (> 0.1 Gy), for which there
are good data upon which to quantify risk, to lower closes, and from acute to
chronic exposure conditions.
13
-------
Especially contentious is the extrapolation to low doses and dose rates.
Generally speaking, epidemiology cannot be used to detect and quantify the
carcinogenic effects of radiation at doses below about 100 mGy of low-LET
radiation because of limitations on statistical power (Land 1980, Brenner et al.
2003). Most cells in the body receive a radiation dose of about 1 mGy/y -
predominantly y-rays from cosmic, terrestrial and internal sources. Given the
typical energies of these background y-rays (0.1-3 MeV) this corresponds to
roughly 1 ionizing track traversing each cell nucleus, on average, annually. Thus,
during the estimated typical time for DMA repair to be completed (a few hours),
roughly 1 out of 1,000 cell nuclei will be hit, and the probability of multiple hits to
the same nucleus will be very low. By way of comparison, at the lowest doses for
which risk can be quantified in the A-bomb survivors, each nucleus was
instantaneously impacted by ~ 100 tracks.
A notable exception to this 100-mGy limit on the sensitivity of epidemio-
logical studies appears to be for studies of childhood cancers induced by
prenatal exposure to diagnostic X-rays, where an excess risk has been observed
at a dose level of about 6-10 mGy (see Section 6). In this case, statistical power
is magnified by the apparent heightened sensitivity of the fetus, combined with a
low background rate of childhood cancers. Typically, the X-rays employed in
these examinations were 80 kVp, and the estimated mean dose was 6 mGy; this
corresponds to only about 1 incident photon per cell nucleus (Brenner and Sachs
2006). Thus, this finding argues against a threshold for radiation carcinogenesis.
Although epidemiology otherwise lacks the power to detect risks from
acute doses of radiation below about 100 mGy, it can provide information on
risks from smaller doses through studies of populations receiving fractionated or
chronic radiation doses that cumulatively add up to about 100 mGy or more. For
example, it was found that multiple fluoroscopic examinations, each delivering an
average dose of approximately 8 mGy, produced a similar increase in breast
cancer, per unit dose, as a single acute dose to the breast (Howe and
Mclaughlin 1996). Likewise, female scoliosis patients under 20 years of age,
who received repeated X-ray examinations, each with a mean breast dose of
approximately 4 mGy, had a higher breast cancer mortality compared to controls
and an increasing mortality with an increasing number of examinations (Doody et
al. 2000). In both these studies, breast cell nuclei received at most a few nuclear
hits from each dose fraction. Finally, based on a revised analysis of the Israeli
tinea capitis study first published by Ron et al. (1989), but incorporating
uncertainties in dosimetry, Lubin et al. (2004) found that children receiving a
mean total thyroid dose of 75 mGy in 5 fractions had a statistically significant
increase in thyroid cancer compared to unirradiated controls.
Epidemiological studies have also been conducted on cohorts of
individuals who received cumulative doses of 100 mGy or more, but where the
dose is spread out over months or years. Radiologists (Lewis 1963, Smith and
Doll 1981, Berrington et al. 2001) and radiological technicians (Wang et al. 1988,
14
-------
Doody et al. 2006), working before modern radiation protection standards had
been implemented, show increased risks of leukemia and breast cancer,
respectively. However, individual dose estimates are generally lacking in these
studies, and they are not very useful for obtaining quantitative risk estimates. A
number of cohort studies are underway, however, which may better demonstrate
and quantify risks from protracted doses of low-LET radiation.
Among the most important of these studies are: nuclear workers in various
countries (Cardis et al. 2005a, 2007, Muirhead et al. 2009); Chernobyl cleanup
workers ("liquidators") (Hatch et al. 2005, Kesminiene et al. 2008, Romanenko et
al. 2008); children exposed to radioiodine releases from the Chernobyl accident
(Cardis et al. 2005b, Tronko et al. 2006); residents downriver from the Mayak
nuclear plant in Russia (Ostroumova et al. 2006, Krestinina et al. 2005);
residents downwind from the Semipalatinsk nuclear test site in Kazakhstan
(Bauer et al. 2005); and inhabitants of Taiwanese apartments constructed with
steel beams contaminated with 60Co (Hwang et al. 2008). Studies on these
populations are ongoing and suffer from various shortcomings, including
incomplete follow-up, dosimetric uncertainties, limited statistical power and
confounding. Nevertheless, results from several of them suggest that radiation
risks can be detected and quantified, even in cases where the average dose rate
is well below 1 mGy/day, corresponding to less than 1 ionizing track per cell
nucleus per day (Puskin 2008).
Jacob et al. (2009) performed a meta-analysis on 12 epidemiological
studies of cancer risks from moderate doses (50-500 mGy) of low dose rate, low-
LET radiation. The ERR/Gy derived from the meta-analysis was a factor of 1.21
times that derived for the LSS cohort (90% Cl: 0.51-1.90). This would correspond
to a DREF of 0.83, with 90% Cl of approximately 0.5 to 2.
15
-------
3. EPA Risk Projections for Low-LET Radiation
3.1 Introduction
For cancer sites other than thyroid, bone, kidney, and skin cancers, the
new EPA risk projections for low-LET radiation are based on the risk models
recommended in BEIR VII and are described in the next section. As in BEIR VII,
the risk models form the basis for calculating estimates of lifetime attributable risk
(LAR), which approximate the premature probability of a cancer or cancer death
that can be attributed to radiation exposure. Relatively minor modifications were
made to the approach used in BEIR VII to the methodology for calculating LAR;
details are given in Section 3.2 and subse-quent sections. Although the main
results are the new EPA estimates of LAR associated with a constant lifetime
dose rate, we also provide estimates to indicate how radiogenic risks might
depend on age at exposure. A detailed discussion of the uncertainties associated
with these risks is given in Section 4.
The main focus of the BEIR VII Report was to develop estimates of risk for
low-dose, low-LET radiation. However, the BEIR VII models are predominantly
based on analyses of the A-bomb survivor data, where the exposure included
high-LET neutrons, as well as y-rays. A recently completed reappraisal of the A-
bomb dosimetry, referred to as DS02, was used as a basis for the BEIR VII
analysis. In BEIR VII, it was assumed that neutrons had a constant RBE of 10
compared to y-rays, implying a "dose equivalent," D, to each survivor (in Sv)
given by:
D = D7+lODn,
where Dy and Dn are, respectively, the y-ray and neutron absorbed doses (in
Gy). The BEIR VII approach then yields models for calculating the risk per Sv,
which can be directly applied to estimate the risk per Gy from a y-ray exposure.
With a constant RBE of 10, the estimated contribution of neutrons is
relatively minor, although not negligible. A recent publication (Sasaki et al. 2008)
presented radiobiological data supporting an RBE for neutrons that was highly
dose dependent, approaching a value of nearly 100 in the limit of low doses. The
authors found that applying their estimates for the RBE brought about better
agreement between Hiroshima and Nagasaki chromosome aberration data and
reduced the estimate of y-ray risk by about 30%.
3.2 BEIR VII Risk Models
The BEIR VII Committee used excess relative risk (ERR) and excess
absolute risk (EAR) to project radiogenic cancer risks to the U.S. population for
each of the cancer sites given in Table 3-1. ERR represents the ratio of the age-
specific increase in cancer rate attributable to a radiation dose divided by the
16
-------
baseline rate, i.e., the rate associated with the background radiation level,
whereas EAR is simply the difference in rates attributable to radiation. In the
models preferred by the BEIR VII Committee for solid cancer sites, ERR and
EAR are functions of age-at-exposure, attained age (the age at which a cancer
might occur), and sex. For leukemia, the "BEIR VII models" also explicitly allow
for dependence of ERR or EAR on time-since-exposure (TSE).
For each cancer site, the BEIR VII risk models were based, at least partly,
on analyses of data from atomic bomb survivors. ERR and EAR models of the
form given in Eq. 3-1 and 3-2 were fit to LSS data on incidence and mortality:
ERR model: Mc,s,a,b,D) = ^(c,s,a,b)[l+ERR(s,e,a,D)]
= ^(c,s,a,by[l+DERR(s,e,a,Dj\ (3-1)
EAR model: Mc,s,a,b,D')--
= JQ(c,s,a,b)+DEAR(s,e,a,D) (3-2)
Here, ERR(s,e,a,D)and EAR(s,e,a,D)are, respectively, the ERR and EAR for a
given sex (s), age at exposure (e), attained age (a), and absorbed dose (D).
ERR(s,e,a,D) and EAR(s,e,a,D) denote the ERR and EAR per unit of dose
expressed in Gy (for low-LET radiation), and ^(c,s,a,b) is the baseline rate,
which depends on city (c, Hiroshima or Nagasaki), sex, attained age, and year of
birth (b). For all solid cancer sites, an LNT model was fit to the LSS data. In
other words, increases in solid cancer rates were assumed to be approximately
equal to the product of a linear-dose parameter that depends on sex, the
absorbed dose, and a function that depends on age-at-exposure and attained-
age, so that ERR and EAR does not depend on dose.
The BEIR VII committee used very similar models to project risks to the
U.S. population. Their ERR and EAR preferred risk models are of the form,
= ^ (s, a)[l +DERR(s, e, a, D)] (3-3)
= ^ (s, a) +DEAR(s, e, a, D) (3-4)
The only difference in the BEIR VII models for projecting risk to the U.S.
compared to the models fit to the LSS data is that in Eq. 3-3 and 3-4, I0(s,a)
represents the baseline rate for the U.S. population, which depends only on sex
and attained age. Otherwise, the two set of models are identical, i.e.,
ERR(s,e,a,D) and EAR(s, e, a, D) represent the same function in Eq. 3-3 and 3-4
as in Eq. 3-1 and 3-2. For example, the BEIR VII committee found that the ERR
decreased by about 25% per decade of age at exposure (for ages under 30) in
the model that "best" fit the LSS data for most cancer sites; consequently, the
17
-------
ERR decreases by the same 25% per decade in their models used to project risk
to the U.S.
Table 3-1: BEIR VII risk model cancer sites
Cancer site(s)
ICD-O-2 codes
Stomach
Colon
Liver
Lung
Breast (female only)
Prostate
Uterus
Ovary
Bladder
Thyroid
"Remainder category"
Solid cancers of the oral cavity, esophagus, small
intestine, rectum, gall bladder, pancreas, digestive
system*, nasal cavity, larynx, other respiratory
system*, thymus, kidney, and central nervous
system. Also includes renal pelvis, ureter cancers,
melanoma, bone, connective tissue, other genital
cancers*, and other solid cancers*
Leukemia (other than chronic lymphatic leukemia)
C16/3
C18/3
C22/3
C33, 34/3
C50/3
C61/3
C53-54, C559/3
C56.C57 (0,1,2,3,4,8)73
C67/3
C739/3
COO-C15/3, C17/3, C19-21/3, C 23-25/3,
C26/3, C422 / 3, C37-39/3, C379/3,
C649/3, C70-72/(2,3), C40/3, C41/3,
C47/3, C49/3, C44/3, M8270-8279,
C659/3, C 669/3, C51/3, C52/3,
C57(7,8,9)/3, C58/ 3, C60/3, C63/3, C42
(0,1,3,4)73, C69/3, C74-76/3, C77/3,
C809/3
Revised ICD 9: 204-208
* Refers to sites not specified elsewhere in this table. Does not include lymphoma.
Of the two types of risk models, ERR models are more appropriate for
cancer sites for which the age-specific excess in cancer incidence rates
attributable to radiation might be roughly proportional to the baseline rate -
independent of the population. In contrast, EAR models are appropriate when the
excess in cancer rates is independent of the baseline risks. The BEIR VII
Committee used each type of risk model (EAR and ERR) to calculate site-
specific risk projections for a U.S. population. For cancers for which the baseline
rates are higher in the U.S. than in the LSS, the ERR models tend to yield larger
projections of radiogenic risk than the projections from EAR models. For other
cancer sites, the projections from EAR models tend to be larger.
A compromise between the two approaches was used for most cancer
sites. Based on the assumption that, for most cancer sites, radiogenic risks for
18
-------
the U.S. population are within the ranges defined by the ERR and EAR
projections, a reasonable approach would be to calculate an "average" of the
projections based on the two types of risk models, e.g., a weighted arithmetic or
geometric mean. This is the approach used by BEIR VII and other compre-
hensive reports on radiation risks and is described in more detail in Section 3.10.
Table 3-2 provides a summary of the BEIR VII ERR and EAR risk models.
For all solid cancer sites except breast and thyroid, the BEIR VII models were
based exclusively on analyses of the A-bomb survivor incidence data. This differs
from EPA's previous risk models (EPA 1994), which for most cancer sites were
derived from LSS mortality data. In general, the LSS incidence data is preferred
as a basis for the risk models because "site-specific cancer incidence data are
based on diagnostic information that is more detailed and accurate than death
certificate data and because, for several sites, the number of incident cases is
larger than the number of deaths (MAS 2006)." For breast and thyroid cancers,
the BEIR VII models were based on previously conducted pooled analyses of
both A-bomb survivor and medical cohort data (Preston et al. 2002b, Ron et al.
1995). The risk model for leukemia was based on an analysis of mortality within
the LSS cohort. In contrast to some other cancer types, "the quality of diagnostic
information for the non-type-specific leukemia mortality used in these analyses is
thought to be high" (MAS 2006).
Table 3-2: Summary of BEIR VII preferred risk models
Cancer site
Description
Data sources
Solid cancers ERR and EAR increase linearly with
except breast, dose; depends also on sex(s), age
thyroid at exposure (e), attained age (a)
Breast EAR increases linearly with dose.
Effect modifiers: (e ,a). Based on
analysis of pooled data (Preston et
al. 2002b). ERR model not used.
1958-1998 LSS cancer incidence
1958-1993 LSS breast cancer
incidence; Massachusetts TB
fluoroscopy cohorts (Boice et al.
1991); Rochester infant thymic
irradiation cohort (Hildreth et a/.1989)
Thyroid
ERR increases linearly with dose.
Effect modifiers (s ,e). Based on
analysis of pooled data (Ron et al.
1995). EAR model not used.
1958-1987 LSS thyroid cancer
incidence (Thompson et al. 1994).
Medical cohort studies: Rochester
thymus (Shore et al. 1993), Israel
tinea capitis (Ron et al. 1989),
Chicago tonsils (Schneider et al.
1993), Boston tonsils (Pottern et al.
1990).
Leukemia ERR and EAR are linear-quadratic
functions of dose. Effect modifiers:
(s ,e ,ct), time since exposure (t).
1950-2000 LSS cancer mortality
(Preston et al. 2004).
19
-------
Solid cancer sites other than breast and thyroid. For most solid cancer
sites, the preferred BEIR VII EAR and ERR models are functions of sex, age at
exposure, and attained age, and are of the following form:
EAR(D,s,e,a)or ERR(D,s,e,a) =
min(e,30)-30
where e* = -
10
(3-5)
(3-6)
As seen in Table 3-3, the values for the parameters, /3S, y, and /7, depend on the
type of model (EAR or ERR). For ERR models, for most sites:
/?, the ERR per Sv at age-at-exposure 30 and attained age 60,
tends to be larger for females than males;
Y = -0.3 implies the radiogenic risk of cancer at age e falls by about
25% for every decade increase in age-at-exposure up to age 30;
and
77 = -1.4 implies the ERR is almost 20% smaller at attained age 70
than at age 60.
As a consequence, ERR decreases with age-at-exposure (up to age 30) and
attained age. In contrast, for EAR models, y = -0.41 and 77 = 2.8 for most sites.
Thus EAR decreases with age-at-exposure, but increases with attained age.
These patterns are illustrated in Figure 3-1.
2.4
22
£*°
« 1$
I'6
I'4
J"
"
°8
as
04
02
s
v? 60
I"
40 SO 00 70 00 90
Attamwlag*
Ag« M *ipo»u<» 10
30
50 90 70 90
Figure 3-1: Age-time patterns in radiation-associated risks for solid cancer incidence excluding
thyroid and nonmelanoma skin cancer. Curves are sex-averaged estimates of the risk at 1 Svfor
people exposed at age 10 (solid lines), age 20 (dashed lines), and age 30 or more (dotted lines).
(BEIR VII: Figure 12-1A, p. 270).
20
-------
Thyroid. For thyroid cancer, the BEIR VII Committee used only an ERR
model to quantify risk. It was of slightly different form than for other solid cancers
in that ERR continues to decrease exponentially with age-at-exposure for ages
greater than 30 y, and ERR is independent of attained age. The BEIR VII ERR
model for thyroid cancer is given in Eq. 3-7:
ERR(D,s,e) =
(3-7)
The BEIR VII thyroid model is a modified version of an ERR model for childhood
exposures (ages < 15) from a pooled analysis of thyroid cancer incidence studies
(Ron et a/. 1995). NIH (2003) later extended the results to all ages of exposure.
The NIH model is a stochastic model in which probability distributions are
assigned to ERR, and the probability distributions depend only on dose and age-
at-exposure. More speci-fically, the geometric means of these distributions are
assumed to be linear with dose and decline in an exponential fashion with age-
at-exposure. The BEIR VII model is very similar to the NIH model, except that it
is not stochastic, and, consistent with findings of Ron et a/., the ERR/Gy is
assumed to be two times larger for females than males.
Neither model accounts for the dependency of ERR on TSE. Analyses of
the pooled thyroid study data indicate that ERR peaks around 15-19 years after
exposure but is still elevated for TSE > 40 (NCRP 2008).
Table 3-3: Parameter values for preferred risk models in BEIR VII1
Cancer
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Other solid
Thyroid2
Leukemia
PM
0.21
0.63
0.32
0.32
0.12
0.5
0.27
0.53
1.1
5 = -C
ERR
PF
0.48
0.43
0.32
1.4
Not
0.055
0.38
1.65
0.45
1.05
1.2
.48,
9 = 0.87 Sv"1,
model
Y
-0.3
-0.3
-0.3
-0.3
used
-0.3
-0.3
-0.3
-0.3
-0.3
-0.83
-0.4
= 0.42
n
-1.4
-1.4
-1.4
-1.4
-1.4
-1.4
-1.4
-1.4
-2.8
0
None
EAR model
PM
4.9
3.2
2.2
2.3
0.11
1.2
6.2
PF
4.9
1.6
1
3.4
See
1.2
0.7
0.75
4.8
Y
-0.41
-0.41
-0.41
-0.41
text
-0.41
-0.41
-0.41
-0.41
-0.41
n
2.8
2.8
4.1
5.2
2.8
2.8
2.8
6
2.8
Not used
1.62
S = 0,
9 = 0.88
0.93
Sv1, (/) = Q
0.29
56
None
1 Adapted from Tables 12-2 and 12-3 of BEIR VII.
Unlike for other sites, the dependence of ERR on age-at-exposure is not limited to ages<30.
21
-------
Breast. For breast cancer, the BEIR VII Committee used only an EAR
model to quantify risk. The model was based on a pooled analysis (Preston et al.
2002b) of eight cohorts: the LSS cohort, and seven cohorts in which subjects
were given radiation treatment for various diseases and/or conditions -
tuberculosis, an "enlarged" thymus, mastitis, benign breast disease, and skin
hemangioma. The cohorts included Asians, Europeans, and North Americans,
who received either single acute, fractionated, or protracted exposures. Although
there was no simple unified ERR or EAR model that "adequately describes the
excess risks in all cohorts," the BEIR VII EAR model provides a reasonable fit to
data from four of the cohorts: the LSS, two cohorts of U.S. tuberculosis patients,
and one of the "enlarged" thymus infant cohorts. No ERR model was found to
provide an adequate fit to the LSS and tuberculosis cohorts because excess
rates attributable to radiation after adjusting for age-at-exposure are similar in all
the three cohorts, despite much larger baseline breast cancer rates in the U.S.
than Japan. For two of the remaining cohorts - Swedish patients treated for
benign breast disease and a N.Y. cohort of mastitis patients - the authors
suggested that effects of predisposition may have accounted for differences in
excess rates, e.g., mastitis patients may be more sensitive to radiation than other
women. The other two cohorts not used in fitting the BEIR VII model were the
cohorts of hemangioma patients and were of limited size.
In the BEIR VII model, the EAR depends on both age at exposure and
attained age (Eq. 3.8), where the parameter estimates are from Preston et al.
(2002b, Table 12). Unlike for other cancers, the EAR continues to decrease
exponentially with age-at-exposure throughout one's lifetime, and the EAR
increases with attained age less rapidly after age 50 (about the time of
menopause).
EAR(D,s,e,d) = j3Dexp[r(e-25)/lO](a/50)" (3-8)
where p = 9.9; 7 = -0.51; rj= 3.5 for a < 50 and 1.1 for a>50.
Leukemia. BEIR VII provided both EAR and ERR risk models for
leukemia (see Eq. 3-9). These differ from models for most other cancer sites. In
the leukemia models, both the ERR and EAR depend on TSE (/), and risk is a
linear-quadratic function of dose. As shown in Figure 3-2, the EAR and ERR per
unit dose both increase with dose (the fitted value for 9 in Eq. 3-9 is positive).
EAR(D, e, 0 or ERR(D, e, t) = (3SD(\ + OD) exp[/ e * + 5 log(f / 25) + (/>e * log(f / 25)],
for t > 5, and
EAR(D, e, 0 = EAR(D, e, 5), for 2 < t < 5 ,
ERR(D, e, 0 = ERR(D, e, 5) ^ ^ & + 5) , for 2 < t < 5 , and
4,0,e + 2)
EAR(D,e,t} = ERR(D,e,f) = 0 for t < 2 . (3-9a,b)
22
-------
The dependence of EAR and ERR on age and TSE is illustrated in Figure
3-3. Both EAR and ERR decrease with TSE for t > 5, and the rate of decrease is
larger for younger ages at exposure. For the time period 2 to 5 y after exposure,
the EAR is constant. The EAR that would be calculated using the ERR model
(note that excess absolute risk is equal to the product of the ERR and the
baseline cancer rate) is also constant for this time period (2
-------
Males
rr
rr
LU
^u
20
15
10
5
1
-
\
\
\
~\ \
\ \
\\.
Females
20
40
60
80
25
20
15
10
5
0
x 10
-4
0 20
x 10"4
40
60
80
LU
20 40 60
Time since exposure
80
20 40 60
Time since exposure
80
Figure 3-3: ERR (per person-Gy) and EAR (cases per person-Gy) for exposures at
low doses and/or dose rates by TSE for three different ages at exposure: 10 (solid),
20 (long dashes), and 30 and above (short dashes).
3.3. Risk Models for Kidney, Central Nervous System, Skin, and Other
"Residual Site" Cancers
BEIR Vll's risk model for what are often termed "residual site" cancers
deserves special mention. The residual category generally includes cancers for
which there were insufficient data from the LSS cohort or other epidemiological
studies to reliably quantify radiogenic site-specific risks. For these sites, results
from the LSS cohort were pooled to obtain stable estimates of risk. With five
exceptions (cancers of the esophagus, bone, kidney, prostate and uterus) the
BEIR VII Report included the same cancers in this category as EPA did in its
previous risk assessment (EPA 1994, 1999b). This section also describes risk
models for cancers of the skin and of the brain and central nervous system
(CMS).
Esophagus. EPA (1994, 1999b) employed a separate risk model for
esophageal cancer, whereas in BEIR VII the esophagus is one of the "residual"
sites. In part, this is because the risk models for the previous assessment were
based on LSS mortality data, for which there was a significant dose-response for
esophageal cancer. In contrast, the BEIR VII models are based on LSS
24
-------
incidence data, for which there was insufficient evidence of a dose-response.
Consistent with BEIR VII, we include esophageal cancer as one of the residual
sites. This decision is expected to have only a minor impact on EPA's risk
coefficients for intake of radionuclides.
Kidney. EPA (1999b) uses a separate model for cancer of the kidney, but
BEIR VII includes kidney as one of the residual sites. In contrast to esophageal
cancer, a separate risk model is needed for this cancer site because the kidney
is an important target for several radionuclides, including isotopes of uranium.
There is little direct evidence upon which to base an estimate for kidney cancer
LAR. In a recent analysis of LSS incidence data (Preston et al. 2007), there were
only 115 kidney cancers, 70% of which were renal cell cancers. The authors
estimated only 6 excess renal cell cancers from radiation exposure. Furthermore,
whatever the association might be between kidney cancer and radiation, it is
complicated by the fact that the etiology for the various kidney cancer types
differ. The estimated dose-response in the LSS appears to be sensitive to the
type of model being fit. Within the LSS cohort, no indication of a positive dose
response was found (p > 0.5) when a constant ERR model was fit, but results
were significant when fit to a constant EAR model. Confidence intervals for linear
dose response parameters are wide for both models, and there is insufficient
evidence to conclude that the dose response in LSS is substantially different for
kidney cancers than other residual site cancers. It was therefore concluded that a
reasonable approach would be to use the BEIR VII residual site ERR model for
kidney cancers. For the kidney EAR model, an adjustment factor was applied,
equal to the ratio of the age-specific kidney cancer baseline rates divided by the
rates for the residual site cancers. EPA's new kidney cancer EAR model is given
in Eq. 3-10:
r^ A n / x I,kidney \ ' / -^ , n f -, /I /I r\\
EARudney (S, 6, d) = EARresidual (S, C, d) (3~1 0)
Al,residual (X a)
Bone. A new EPA model for a-particle-induced bone cancer risks is based
on an analysis of data on radium dial painters exposed to 226Ra and 228Ra and
patients injected with the shorter-lived isotope 22 Ra (Nekolla et al. 2000). The
risk per Gy for low-LET radiation is assumed to be 1/10 that estimated for a-
particle radiation. Details about the EPA bone cancer risk model and its
derivation are provided in Section 5.1.2 (on human data on risks from higher-LET
radiation).
The new risk projections for bone cancer incidence from low-LET radiation
are 2.4x10"4 Gy1 (males), 2.3x10"4 Gy1 (females), and 2.4x10"4 Gy"1 (sex-
averaged). About 35% of all bone cancers are fatal (SEER Fast Stats), and it is
assumed here that the same lethality holds for radiogenic cases. The projected
mortality risk estimates are then 8.6x10"5 Gy"1 (males), 8.2x10"5 Gy"1 (females),
and 8.4x10"5 Gy"1 (sex-averaged).
25
-------
Prostate and uterus. In contrast to EPA (1999b), BEIR VII provides
separate risk models for these two cancer sites, and these BEIR VII models form
the basis for new EPA projections. This is in contrast to EPA (1994, 1999b), in
which these two cancer sites were included in the residual category. The A-bomb
survivor data now provide sufficient information on radiogenic uterine cancer to
formulate a risk projection of reasonable precision. BEIR VII cited the vastly
differing baseline rates for the U.S. compared to Japan as a reason for providing
a separate prostate estimate.
Skin. Previously, EPA risk estimates for radiation-induced skin cancer
mortality (EPA 1994) were taken from ICRP Publication 59 (ICRP 1991). The one
modification made by EPA was to apply a DDREF of 2 at low doses and dose
rates. Recognizing that the great majority of nonmelanoma skin cancers are not
life threatening or seriously disfiguring, EPA included only the fatal cases in its
estimates of radiogenic skin cancer incidence. The contribution of skin cancers to
the risk from whole-body irradiation was then minor: about 0.2% and 0.13% of
the total mortality and incidence, respectively.
ICRP's calculation of skin cancer incidence risk employed an ERR of 55%
per Sv, along with U.S. baseline skin cancer incidence rates from the 1970s. The
ICRP mortality estimate was also based on conservative assumptions that: (1)
1/6 of radiogenic skin cancers would be squamous cell carcinomas (SCC), the
remainder basal cell carcinomas (BCC) and (2) essentially all of the BCC would
be curable, whereas about 1% of SCC would be fatal. Based on these
considerations, ICRP Publication 59 estimated that 0.2% of the cases would be
fatal.
The ICRP risk estimates closely mirror those previously published by
Shore (1990), who also served on the committee that drafted ICRP Publication
59. Shore (2001) reviewed the subject again in light of additional information and
concluded that essentially all of the radiation-induced skin cancers at low to
moderate doses would be BCC. Therefore, it is assumed here that only BCC are
radiogenic at low doses. He maintained that the fatality rate for BCC is "virtually
nil" but cited a study indicating a rate of 0.05% (Weinstock 1994). Shore also
noted that there was no persuasive evidence that radiation-induced BCC would
be more fatal than sporadic cases.
At the same time, there is evidence that the baseline rates for BCC have
increased dramatically since the 1970s, which might also result in a higher
(absolute) risk per unit dose of inducing a radiogenic skin cancer.
There are 3 major cohort studies of radiation-induced skin cancer with
thorough dosimetry and long-term follow-up (Shore 2001): (1) the LSS cohort,
including both children and adults, exposed to a wide range of doses of y-rays
from the atomic bomb (Ron et al. 1998, Preston et al. 2007); (2) a cohort of 2,224
children in New York City treated for tinea capitis (ringworm of the scalp) with an
26
-------
average dose of 4.75 Gy of 100 kVp X-rays (Shore et a/. 2002); and (3) a cohort
of 10,834 children in Israel treated for tinea capitis with an average dose of 6.8
Gy of 70-1 00 kVp X-rays (Ron et a/. 1 991 ).
The ERR/Gy for the two tinea capitis cohorts were found to be very
similar: 0.6 Gy"1 (NY) and 0.7 Gy"1 (Israel). Both studies showed a decline in risk
with age at exposure: 12% per y in the New York study, and 13% per y in the
Israeli study (Shore 2001). The average age at exposure in the New York study
was 7.8, compared to 7.1 in the Israeli. Overall, the results of the two studies
then indicate a risk coefficient of « 0.7 Gy"1 for exposure at age 7, with about a
12% per year decrease in risk with age at exposure. Both the LSS and the Israeli
tinea capitis study appear to show some decline in the ERR at longer times since
exposure, but the declines were not statistically significant; the New York tinea
capitis study showed no indication of a decline, even after 45-50 years after
irradiation (Shore 2001). Based on this information, the tinea capitis data can be
reasonably described by the equation below:
(3-11)
Where D is dose (Gy) and e is the age at exposure.
Skin cancer incidence exhibited a nonlinear dose response in the LSS
(Preston et a/. 2007). Fitted to a spline function with a knot at 1 Gy, the ERR/Gy
for BCC was estimated to be about 5.5 times higher above 1 Gy than below (7
times for all nonmelanoma skin cancers). Similarly to the tinea capitis results, the
risk was found to decrease by about 12.3% per year of age at exposure, the fall-
off extending into adult age groups (Ron et a/. 1998). Normalized to the same
dose and age at exposure, the ERR was considerably higher in the Japanese A-
bomb survivor population than in the mainly Caucasian populations irradiated for
tinea capitis. In contrast to the tinea capitis cohorts, there was no evidence of a
higher radiation risk to UV shielded parts of the body. This suggests that there
may be a synergism between ionizing and ultraviolet radiation for Caucasians,
but not for the Japanese. Quite possibly, this relates to differences in skin
pigmentation (Ron et a/. 1998). For this reason, we are primarily basing our skin
cancer risk estimates on the tinea capitis data, which is probably more applicable
to the U.S. population.
As discussed in Sections 2.14 and 3.6, the low-LET risk model in BEIR VII
for all solid cancers is consistent with a LDEF of approximately 1+0. 5D, where D
is the dose in Gy. Assuming that this relationship holds for BCC induction, and
given the magnitude of the average therapeutic doses received by the New York
and Israeli tinea capitis patients, a LDEF of about 3.4 or 4.4 would be inferred for
extrapolating the risks estimates derived from these studies to low doses, but this
neglects the possible influence of cell killing at the high therapeutic doses
administered to these patients, which may tend to flatten the dose-response and
reduce the LDEF. On the other hand, a further reduction factor might be
27
-------
appropriate for estimating risks from typical y-rays with energies around 100 keV
or higher (see Section 5.2). The LSS data on skin cancer suggest an even larger
LDEF of about 5.5. The UNSCEAR 2006 Report (UNSCEAR 2008) fit various
dose-response models to the LSS skin cancer incidence data and found a best fit
for models in which either ERR or EAR are "quadratic-exponential" in dose and
include an adjustment for attained age and age-at-exposure:
ERR or EAR = J32D exp(aD)f(a, e). (3-12)
The UNSCEAR models project a negligible risk at very low doses.
Based on the above considerations, we adopt a low-dose/low-dose-rate
y-ray relative risk coefficient about one-third that inferred from a linear fit to the
tinea capitis data:
0.2D(0.88)e-7 (3-13)
For life-table calculations, baseline incidence rates are needed, but SEER
does not include nonmelanoma skin cancers in its database. BCC incidence
rates have increased dramatically over the last 3 decades (Karagas ef a/. 1999),
and it has been estimated that there are 900,000 incident cases of BCC annually
in the U.S. (550,000 in men, 350,000 in women), the great majority of these in
whites (Ramsey 2006). The estimated lifetime risk of BCC in the white population
is very high: 33-39% in men and 23-28% in women. Overall, the age-adjusted
incidence per 100,000 white individuals is 475 cases in men and 250 cases in
women. To calculate age-specific baseline incidence rates, we applied these
age-adjusted numbers and assumed that the rates increase with age to the
power of 4.5, which is the roughly the pattern observed for many cancers
(Breslow and Day 1987).
The age-adjusted fatality rate has recently been estimated to be 0.08 per
100,000 individuals, based on only 12 BCC deaths in the state of Rhode Island
between 1988 and 2000 (Lewis and Weinstock 2004). The case fatality rate for
BCC can then be roughly estimated to be: 0.08 / 0.5(475+250) « 0.03%, which
we have adopted for making skin cancer mortality projections.
The derived risk projections for skin cancer incidence are: 1.8x10"2 Gy"1
(males), 9.6x10"3 Gy"1 (females), and 1.4x10"2 Gy"1 (sex-averaged). The mortality
risk projections are: 5.4x10"6 Gy"1 (males), 2.9x10"6 Gy"1 (females), and 4.1x10"6
Gy"1 (sex-averaged).
As noted above, the great majority of non-melanoma skin cancers are not
serious, in the sense that they are not life threatening or significantly disfiguring.
This is particularly true for BCC. We believe that it is reasonable to omit these
cancers from our cancer incidence risk estimates rather than including them
28
-------
along with much more serious types of cancers. Were we to include all the
estimated radiogenic BCC cases, the numerical estimate of risk from uniform,
whole-body radiation would be increased by about 9%. Serious cases of BCC,
involving invasion of the cancer into underlying tissues can arise, however, if the
problem is neglected for a long time. It would be reasonable to include all these
cases in our whole-body risk estimates. Unfortunately, however, there appear to
be no reliable data on the fraction of BCC cases that turn out to be significantly
disfiguring or to require extensive surgery. For this reason, EPA is following its
previous practice of including only the estimated radiogenic BCCs in its official
estimates of radiogenic cancer incidence (Table 3-16). By way of illustration, if
one were to assume that 5% of the radiogenic BCC cases are "serious" enough
to be included in the cancer incidence estimates, the resulting average skin
cancer risk coefficient would be =5x10~4 per Gy, and the age-averaged whole-
body risk coefficient for incidence would be increased by about 0.5%
Brain and central nervous system. As in BEIR VII, EPA has no formal
separate risk model for brain and central nervous system (CMS) cancers.
Instead, these cancers are included as part of the residual site category.
Nevertheless, it is possible to compare BEIR Vll's ERR model for residual site
cancers to alternative ERR models that have been derived from LSS data on
brain and CMS cancers. Preston et al. (2002a) found a nearly statistically
significant (p=0.06) dose-related excess of CMS tumors other than schwannomas
that were diagnosed between 1958 and 1995 among 80,160 A-bomb survivors.
There was a "marked decrease" in excess risk with age at exposure, but no clear
pattern associated with attained age. A model for ERR, based on their analysis,
is given in Eq. 3-14. Based on essentially the same data, UNSCEAR (2008)
obtained the model given in Eq. 3-15. As shown in Figure 3-4, the UNSCEAR
model features an even steeper decrease in ERR with age-at-exposure (for ages
< 10) than the model by Preston and others. It can be seen in the same Figure
that a more gradual decrease in ERR with age-at-exposure is predicted by the
BEIR VII risk model for residual cancers. However, in the BEIR VII model, the
ERR is highly dependent on attained age; e.g., for exposures before age 10, the
ERR per Gy at attained age 15 ranges from about 15 to 22, whereas the ERR
per Gy for attained age 60 is always less than 1.
ERR(D,e) = 0.15£>exp(-0.97(>-30)/10) (3-14)
ERR(D,e) = 7.43145 £>exp[-0.98971og(e)] (3-15)
Unfortunately, it is not clear which of the three alternative ERR models
shown in Figure 3-4 is closer to the "truth." For example, it is not clear whether
the ERR for CMS cancers depends on attained age. In the analysis by Preston et
al., no decrease was found in ERR with attained age, but this may be due to the
small number of excess CMS tumors that were associated with radiation (12,
excluding schwannomas). Some evidence of an attained age effect is suggested
in the Israeli tinea capitis (ringworm) study (Ron et al., 1988). The authors found
29
-------
a significantly elevated risk for attained ages up to 35, "when the risk appeared to
decline." However, other studies provide no conclusive evidence of an attained
age effect, and it is not clear why the age pattern in radiogenic risk for CMS
should be similar to that for other "residual site" cancers.
Table 3-4 indicates that the projected LAR for CMS cancers based on the
three alternative models are within a factor of about 2; sex-averaged LAR range
from 0.0013 (Preston et al. model) to 0.0029 (UNSCEAR model) for lifelong
exposures, and from about 0.005 (Preston et al.) to 0.010 (BEIR VII residual
ERR model) for childhood exposures.
Exposure at ages < 10
Attained age = 15
UNSCEAR
Attained age=25
/ Attained age = 60
/
RERF (2002)
Age at exposure
Exposure at ages > 10
0.3
0.2
0.1
10
20 40 60 80
Age at exposure
Figure 3-4: Comparison of two different ERR models for brain and CMS cancer with the residual
site ERR model (dashed-and-dotted lines). For UNSCEAR (2006) and RERF (Preston et al.
2002a), ERR depends on age at exposure. ERR for the residual site cancer (shown here for
males) depends on sex, age at exposure and attained age.
30
-------
Table 3-4: Projection of LAR (Gy~1) for brain and CMS cancers for three
alternative ERR models
Lifelong exposures1 Childhood exposures2
ERR model Males Females Males Females
BEIRVII "Residual"
Preston et a/. (2002a)
UNSCEAR (2008)
0.0023
0.0014
0.0017
0.0029
0.0011
0.0013
0.0085
0.0056
0.0061
0.0119
0.0045
0.0049
1 Risks for exposures during the first year of life are omitted from these calculations; ERR for the
UNSCEAR model approaches infinity for ages near zero.
2 Risk from exposures between 1st and 15th birthday.
3.4 Risk Model for Thyroid Cancer
EPA's new risk model for thyroid cancer incidence is very similar to a
model recommended by NCRP (2008), which explicitly accounts for the
dependence of ERR on both age-at-exposure and time-since-exposure. Both the
NCRP and new EPA thyroid risk models are primarily based on a model derived
by Lubin and Ron (1998) from a subset of the pooled data of thyroid incidence
studies described in previous section. The models are all of the form:
ERR(D, e, t) = 0DA(e)T(f), (3-16)
and the multiplicative factors for age-at-exposure, A(e), and time-since-exposure,
T(t), are given in Table 3-5.
As is apparent in Table 3-5, the models are very similar. However, in
contrast to the model derived by Lubin and Ron, EPA uses a single coefficient for
TSE between 5 and 14, and TSE > 30. (There is insufficient data to detect
differences in ERR for some of the subcategories for TSE used by Lubin and
Ron). The Lubin and Ron model does not provide estimates of ERR for age-at-
exposure > 15. For these "non-childhood" exposures, the EPA model borrows
from the BEIR VII model, which stipulates an 8% y"1 decrease in ERR with age-
at-exposure. We chose not to use the NCRP model because, although their
model appears reasonable, the report did not provide an explanation for the
minor discrepancies with Lubin and Ron (1998) or how results were extended for
age-at-exposure > 15.
For calculating LAR for mortality, NCRP (2008) used the sex-averaged
estimates of 5-y cancer fatality rates from the SEER program for the period 1998-
2002 (see Table 3-6), and then doubled these to account for further mortality
more than 5 y after diagnosis. Thus, 2(100-99.3)% = 1.4% of radiogenic cancers
diagnosed before age 45 were assumed to be fatal, compared to 50% for
cancers diagnosed after age 75.
31
-------
Table 3-5: Estimated ERR/Gy and effect modifiers for age at exposure and
time since exposure (TSE)
ERR/Gy (P)
Lubin & Ron (1998)
10.7
Models
EPA1
10.7
NCRP (2008)
11.7
Age-at-exposure: A(e)
<5
5-9
10-14
15-19
20+
1.0
0.6
0.2
None given
None given
1.0
0.6
0.2
0.2 exp[-0. 083(e-1 5)]
0.2exp[-0.083(e-15)]
1.0
0.7
0.2
0.2
0.09 (e<30), 0.03 (e>30)
TSE: T(t)
<5
5-14
15-19
20-24
25-29
30-40
40+
0
1.3(iS10);1.0(f>10)
1.9
1.2
1.6
0.5 (f<35); 0.2 (f>35)
0.7
0
1.15
1.9
1.2
1.6
0.47
0.47
0
1
1.6
1
1.4
0.394
0.394
1 For age-at-exposure > 15, the ERR per Gy decreases 8% y"
Based on the EPA thyroid incidence model, and the NCRP approach for
mortality, the LAR for mortality would be about 2.7xlO"4 (males) and 5.7xlO"4
(females). Dividing these by EPA projections for incidence (see Section 3.13)
yields overall fatality rates of 13% (males) and 9% (females). However, 10-y
relative survival rates for thyroid cancer have been about 95% since 1993 (see
Table 3-6), and few deaths are found to occur more than 10 y after diagnosis.
Furthermore, the fatality rate for radiogenic thyroid cancer is unlikely to be
greater than for sporadic cancers (Bucci et a/. 2001). Based on these
considerations, EPA conservatively assumes a simple 5% fatality rate for all
radiogenic thyroid cancers.
32
-------
Table 3-6: Summary of SEER thyroid relative and period survival rates
Type of Statistic
5-year relative survival
5-year period survival
Data Sex
1998-20021 Both
Both
1999-20062 Male
Female
Age at
diagnosis
<45
45-54
55-64
65-74
75+
All
All
All
Percent
99.3
98.1
93.7
90.2
75.0
97.4
94.2
98.3
(95 1 95 2 96 3
10-year relative survival 1999-20062 Both All v ' ' ' ' 3' '
1 From NCRP (2009)
2 From Altekruse et al. (2010)
3 Year of diagnosis: 1993-1997
3.5 Calculating Lifetime Attributable Risk
As in BEIR VII, lifetime attributable risk (LAR) is our primary risk measure.
As discussed in Section 3.2, separate evaluations of LAR were made for most
cancer sites using both an excess absolute risk (EAR) model and an excess
relative risk (ERR) model. For a person exposed to dose (D) at age (e), the LAR
is:
110
LAR(D,e)= ^M(D,e,o)-S(d)/S(e)da, (3-17)
e+L
where M(D, e, a} is the excess absolute risk at attained age a from an exposure
at age e, S(a) is the probability of surviving to age a, and L is the minimum latency
period (2 y for leukemia, 5 y for solid cancers). (Note: In Eq. 3-17 and
subsequent equations, dependence of these quantities on sex is to be
understood). The LAR approximates the probability of a premature cancer death
from radiation exposure and can be most easily thought of as weighted sums
(over attained ages up to 110) of the age-specific excess probabilities of
radiation-induced cancer incidence or death, M(D, e, a}.
For any set of LAR calculations (Eq. 3-17), the quantities M(D, e, a} were
obtained using either an EAR or ERR model. For cancer incidence, these were
calculated using either:
MI(D,e,d) = EARI(D,e,d) (EAR model) (3-18)
or MI(D,e,a) = ERRI(D,e,a)-AI(a) (ERR model) (3-19)
33
-------
where A/(a) is the U.S. baseline cancer incidence rate at age a. Datasets used to
derive baseline incidence rates are described in Section 3.8.
For mortality, the approach is very similar, but adjustments needed to be
made to the equations since both ERR and EAR models were derived using
incidence data. In BEIR VII, it was assumed that the age-specific ERR is the
same for both incidence and mortality, and the ERR model-based excess risks
were calculated using:
MM (D,e,a) = ERR, (D,e,a) • /1M (a) . (3-20)
i.e., the age- and sex-specific mortality risks is the excess relative incidence risk
times the baseline mortality rate. For EAR models, BEIR VII used essentially the
same approach by assuming:
Mu(D,e,a)= '''(a). (3-21)
^ (a)
Note that in Eq. 3-21 , the ratio of the age-specific EAR to the incidence rate is
the ERR for incidence that would be derived from the EAR model. Eq. 3-20 was
used for all cancer sites other than skin and thyroid cancers, for which a constant
fatality rate (0.03% for skin cancer and 5% for thyroid cancer) was applied to the
projections for incidence. Eq. 3-21 was used for all sites except bone (fatality rate
= 35%) and breast cancer. A description of the approach for estimating breast
cancer mortality risk, and its rationale, is given in Section 3.1 1 .
The LAR for a population is calculated as a weighted average of the age-
at-exposure specific LAR. The weights are proportional to the number of people,
N(e), who would be exposed at age e. The population-averaged LAR is given by:
-. 110-Z,
LAR(D,pop) = f N(e)-LAR(D,e)-de. (3-22)
AT" * J
For the BEIR VII approach, N(e) is the number of people, based on
census data, in the U.S. population at age e for a reference year (1999 in BEIR
VII), and TV* is the total number summed over all ages. In contrast, for our
primary projection, we used a hypothetical stationary population for which N(e) is
proportional to S(e), based on observed 2000 mortality rates. In this case,
110-L
J S(e)-LAR(D,e)-de
LAR(D, stationary) = -2 — . (3-23)
J S(e)de
34
-------
Eq. 3-23 represents the radiogenic risk per person-Gy from a lifetime chronic
exposure. Note that the equations above do not account for changes in future
mortality rates. For a stationary population, Eq. 3-23 is equivalent to Eq. 3-
22, so that the risk coefficient for a chronic exposure is equal to the (age-
averaged) risk coefficient for an acute exposure.
Computational details on how the integrals in Eq. 3-17, 3-22 and 3-23
were approximated are given in Appendix A.
3.6 Dose and Dose Rate Effectiveness Factor
To project risk at low or chronic doses of low-LET radiation, the BEIR VII
Committee recommended the application of a Dose and Dose Rate Effectiveness
Factor (DDREF), as described in Section 2.1.4. Effectively, this assumes that at
high acute doses, the risk is given by a linear-quadratic (LQ) expression,
ajD+a2D2, whereas at low doses and dose rates, the risk is simply
In the case of leukemia, LSS data shows upward curvature with increasing
dose. The BEIR VII fit to the LQ model yielded a value of B=a2/ai = 0.88 Sv"1.
For solid tumors, the upward curvature in the LSS data appears to be
lower and is not statistically significant (i.e., 6 is not significantly different from 0).
While BEIR VII did not explicitly recommend a LQ model for solid cancer risk, it
nevertheless concluded that some reduction in risk at low doses and dose rates
was warranted. It adopted a Bayesian approach, developing separate estimates
of the DDREF from radiobiological data and a statistical analysis of the LSS data.
The estimate for the DDREF obtained in this way was 1.5, somewhat lower than
values that had been commonly cited in the past. The BEIR VII Report notes that
the discrepancy can largely be attributed to the fact that the DDREF is dependent
on the reference acute dose from which one is extrapolating. According to BEIR
VII, the appropriate dose should be about 1 Sv because data centered at about
this value drives the LSS analysis. In contrast, much of the radiobiological data
refers to effects observed at somewhat higher doses, for which the DDREF
would be higher. Assuming that the extrapolation is indeed from an acute dose of
1 Sv, the DDREF of 1 .5 corresponds to a LQ model in which 6 = 0.5 Sv"1.
3.7 EAR and ERR LAR Projections for Cancer Incidence
EAR and ERR model-based LAR projections for a stationary population
based on 2000 mortality data are given in bold typeface in Table 3-7. These are
compared to EAR and ERR projections based on census data, with weights
proportional to the number of people of each age in the year 2000. The results
indicate that our primary risk projections are about 5-10% lower than they would
be if based on a census population. Results in Table 3-7 reflect the DDREF
adjustment of 1 .5 for all cancer sites except leukemia, bone and skin.
35
-------
Table 3-7: EAR and ERR model projections of LAR for cancer incidence1'2
for a stationary population3 and a population based on 2000 census data4
Risk Model
Population Weighting
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Leukemia
Skin
Sex
M
F
M
F
M
F
M
F
F
M
F
F
M
F
M
F
M
F
M
F
M
F
M
F
M
F
ERR
Stationary
15
19
160
103
17
7.4
155
485
Not Used
126
11
34
107
105
22
65
275
292
26
24
No model
No model
109
86
182
96
Projection
Census
16
21
171
110
19
8.0
166
520
Not Used
135
12
37
113
111
24
70
300
315
28
26
No model
No model
109
87
199
103
EAR
Stationary
171
204
112
67
92
53
120
233
289
3.8
50
29
75
63
No model
No model
196
184
21
16
2.4
2.3
53
32
No model
No model
Projection
Census
184
217
120
71
98
56
126
244
316
4.1
53
31
79
66
No model
No model
210
196
23
17
2.7
2.6
57
34
No model
No model
1 Number of cases per 10,000 person-Gy.
2 Uses DDREF of 1.5 for all sites except leukemia, bone, and skin
3 Based on 2000 decennial life tables (Arias 2008)
4 NCHS (2004)
3.8 ERR and EAR Projections for Cancer Mortality
We adopt the BEIR VII approach for ERR and EAR projections of LAR for
mortality for all cancer sites except breast cancer. As noted previously, for its
ERR model-based projection, BEIR VII used:
36
-------
D,e,d).AM(d) , (3-24)
and for its EAR based projections,
^ EARj(D,e,d)
MM(D,e,a) =
A, (a)
(3-25)
In Eq. 3-25, the ratio in square brackets is equal to the ERR for incidence
that would be calculated using the EAR model. In both Eq. 3-24 and 3-25, the
BEIR VII approach assumes that the ERR for incidence and mortality are equal.
However, this ignores the "lag" between incidence and mortality, which could
lead to bias in the estimate of mortality risk in at least two different ways.
First, there would be a corresponding lag between the ERR for incidence
and mortality, which might result in an underestimate of mortality risk. For
purposes of illustration, suppose that: (a) a particular cancer is either cured
without any potential life-shortening effects or results in death exactly 10 y after
diagnosis and (b) survival does not depend on whether or not it was radiation-
induced. Then,
ERRM(e,a) = ERR,(e,a-lG)> ERR,(e,a) . (3-26)
The relationship would also hold for the EAR if the baseline cancer rate has the
same age-dependence for A-bomb survivors as for the U.S. population.
Second, since current cancer deaths often occur because of cancers that
developed years ago, application of the EAR-based ERR for incidence can result
in a substantial bias due to birth cohort effects. If age-specific incidence rates
increase (decrease) over time, the denominator in Eq. 3-25 would be too large
(small). This could result in an underestimate (overestimate) of the LAR.
The BEIR VII approach is reasonable for most cancers, because the time
between diagnosis and a resulting cancer death is typically short. An exception is
breast cancer, for which our approach is presented in Section 3.11.
Results of LAR calculations using the BEIR VII approach are given in
Table 3-8. Although not shown, LAR for mortality tends to be about 5% larger for
census-based weights than for weights based on a stationary population.
Mortality and incidence data used for the calculations are described in the next
section.
37
-------
Table 3-8: Age-averaged LAR1'2'3 for cancer mortality based on a
stationary population4
Risk Model
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Leukemia
Skin
Sex
M
F
M
F
M
F
M
F
F
M
F
F
M
F
M
F
M
F
M
F
M
F
M
F
M
F
ERR
7.5
11
74
45
12
6.1
140
384
Not used
19
2.5
22
21
27
1.1
3.2
112
132
8.4
7.4
No model
No model
80
63
0.05
0.03
EAR
88
111
51
29
75
46
111
200
955
0.8
16
22
19
23
No model
No model
103
108
8.0
6.3
0.9
0.8
31
20
No model
No model
1 Cases per 10,000 person-Gy
2 Except for skin, bone, kidney, and thyroid cancers, projections based on BEIR VII risk models.
3 Based on DDREF of 1.5 except for leukemia, bone, and skin
^Arias (2008).
5 See Section 3.11
3.9 Data on Baseline Rates for Cancer and All-Cause Mortality
Cancer specific incidence and mortality rates are based on data from the
Surveillance, Epidemiology, and End Results (SEER) program of the National
Cancer Institute (NCI). Begun in the early 1970s, SEER collects incidence and
survival data from several, mostly statewide and metropolitan, cancer registries
38
-------
within the U.S. The SEER program has expanded several times - most notably,
from 9 registries (SEER 9) to 13 registries (SEER 13) in the early 1990s and,
more recently, from 13 registries to 17 registries (SEER 17). The program also
obtains mortality data from the National Center for Health Statistics (NCHS).
Cancer incidence (SEER 2009a,b) and mortality (SEER 2010) rates for
this report were obtained using the software package SEER-Stat, available from
the SEER website (http://seer.cancer.gov). For this report, the cancer and sex-
and age-specific baseline incidence rates were obtained as a weighted average
of the smoothed 1998-1999 rates based on data from the SEER 13 registries and
2000-2002 rates from SEER 17 registries. This contrasts with BEIR VII, which
used (a previous version) of public-use SEER 13 data for the years 1995-99.
Graphs of the baseline rates and details on how the data were smoothed are
given in Appendix A.
SEER areas currently comprise about 26% of the U.S. population and are
not a random sample of areas within the U.S. Nevertheless the cancer rates
observed in the combined SEER areas are thought to be reasonably similar to
rates for the U.S. population. Sampling errors for these baseline rates are
relatively small, and contribute only negligibly to uncertainties in projections of
(radiogenic) LAR. However, it is anticipated that risk projections might occa-
sionally be updated to reflect changes in rates for both incidence and mortality.
Table 3-9 gives estimates of the annual rate of change in incidence rates
for the SEER 13 registries for the years 1992-2007. During this time period,
incidence rates for most cancers changed by less than 2% per year. Notable
exceptions include liver cancer (> 6% per year increase from 1992-96), thyroid
cancer (almost 6% per year increase from 1997-2007), and prostate cancer
(about 11% per year decrease from 1992-1995). Thus, if these past trends are
any indication, it is conceivable that after about 10 years, an update in baseline
incidence rates alone could be responsible for a 50% or greater change in the
LAR projection for one or more cancers. (It is beyond the scope of the report to
speculate on changes in baseline mortality rates).
For calculating survival probabilities, 2000 decennial life tables (Arias
2008) were used instead of 1999 life tables as in BEIR VII.
39
-------
Table 3-9: Changes in age-averaged cancer rates from 1992-2007 for the
SEER 13 registries1
Cancer site Average annual percent increase
Stomach -1.4
Colon and Rectum -2.2 (1992-95), 1.9 (1995-98), -2.6 (1998-2007)
Liver & Intrahepatic Bile Duct 6.4 (1992-96), 2.6 (1996-2007)
Lung (male) -2.1
Lung (female) 0.6 (1992-98), -0.6 (1998-2007)
Bladder 0.1 (1992-2004), -1.5 (2004-2007)
Breast 1.1 (1992-99),-1.8 (1999-2007)
Prostate -11.1 (1992-95), 2.0 (1995-2000), -2.3(2000-07)
Corpus and Uterus, NOS 0.6 (1992-97), -0.6 (1997-2007)
Cervix Uteri -2.9
Ovary -0.6 (1992-2001), -2.0 (2001-07)
Thyroid 3.0 (1992-97), 5.7 (1997-2007)
Leukemia -0.1 (1992-2004), 2.1 (2004-2007)
1 Abstracted from SEER Fast Stats (NCI 2011)
3.10 Combining Results from ERR and EAR Models
3.10.1 BEIR VII approach. BEIR VII calculates LAR values separately
based on preferred EAR and ERR models and then combines results using a
weighted geometric mean. More specifically,
LAR(B1) = (LAR(R) y* (LAR(A) }l-v' (3-27)
where w* is the weight for the ERR model and depends on cancer site. If the
weight (w*) equals 0.5, a simple GM would be calculated. Instead for most
cancer sites, BEIR VII recommended a weight (w*) equal to 0.7 - placing
somewhat more emphasis on results from ERR models. (A notable exception is
lung cancer, for which the EAR model was given more weight. BEIR VII cited
Pierce et al. (2003), who found a submultiplicative interaction between smoking
and radiation in the A-bomb survivor data. Subsequently, Furukawa et al. (2010)
reported that the submultiplicative interaction may be restricted to only heavy
smokers.)
There are at least two problems with BEIR Vll's use of the weighted GM.
First, it is difficult to explain how a projection based on the GM should be
interpreted. Second, the GM is not additive in the sense that: the GM of two risk
projections for the combined effect of separate exposures is generally not equal
40
-------
to the sum of the GM projections for the exposures. For these reasons, EPA has
instead employed a weighted AM to combine ERR and EAR projections, which
has a relatively straightforward interpretation and is additive.
3.10.2 EPA approach. We calculate the combined age-specific risk (at
high dose rates) using a weighted arithmetic mean, so that:
(A) (D, e, a)] , (3-28)
and the LAR at exposure age e is calculated as before:
110
LAR(D, e) = J M(EPA} (D, e, a) -S(d)l S(e)da . (3-29)
e+L
In Eq. 3-28, JvfA} and A/R) represent the age-specific EARs derived from the EAR
and ERR models, respectively; e.g., for incidence: M\A)(D,e,d) = EARI(e,a)D,
and M\R)(D,e,d) = ERRI(e,a)D-^I(d). It can be easily shown that:
LAR(EPA) (D, e) = w* LAR(R} (D, e} + (l- w*}LAR(A} (D, e) (3-30)
In general, the weighted arithmetic mean approach (Eq. 3-30) will always
result in larger LAR projections than the BEIR VII approach based on the GM.
However, as can be seen in Table 3-10, the difference is substantial only for sites
such as stomach, liver, prostate, and uterine cancers, for which the LAR
projection is sensitive to the model type (ERR vs. EAR). For all cancers
combined (excluding nonfatal skin cancers), use of the weighted AM results in an
LAR projection about 12% (males) or 6% (females) greater than the BEIR VII
approach based on the GM.
41
-------
Table 3-10: Comparison of EPA and weighted geometric mean (GM) method
for combining EAR and ERR LAR projections for incidence1'2
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Leukemia
Bone
Total
(excluding skin)
Sex
M
F
M
F
M
F
M
F
F
M
F
F
M
F
M
F
M
F
M
F
M
F
M
F
M
F
ERR
Projection
(A)
15
19
160
103
17
7.4
155
485
Not used
126
11
34
107
105
22
65
275
292
26
24
109
86
No model
No model
EAR EPA
Projection Projection
(B) (C)
171
204
112
67
92
53
120
233
289
3.8
50
29
75
63
No model
No model
196
184
21
16
53
32
2.4
2.3
62
75
146
92
40
21
130
308
289
89
23
33
97
92
22
65
251
259
24
22
92
69
2.4
2.3
955
1350
Weighted
GMof
A and B:
(D)
31
39
144
91
28
13
129
290
289
44
18
33
96
90
22
65
248
254
24
21
87
63
2.4
2.3
856
1270
Ratio:
D/C
2.01
1.90
1.01
1.02
1.40
1.57
1.01
1.06
1.00
2.02
1.30
1.00
1.01
1.02
1.00
1.00
1.01
1.02
1.00
1.02
1.05
1.09
1.00
1.00
1.12
1.06
1 Cases per 10,000 person-Gy.
2 Based on DDREF of 1.5 for sites other than leukemia and bone
3.10.3 A justification for the weighted AM. The weighted arithmetic
mean approach can be justified by first expressing the age-specific lifetime
excess risk for the U.S. as a weighted arithmetic mean of the relative risk and
absolute risk model projections. One then assigns a subjective probability
distribution to the weight (w), for which the expected value of the probability
distribution is approximated by the BEIR VII nominal value (E[w] = w*). For any
42
-------
such subjective distribution, the weighted arithmetic mean will be an unbiased
estimate of the "true" excess risk.
More specifically, let w be an (unknown) parameter such that the (true)
excess risk A/frue) in the U.S. population is given by:
(3-31)
It follows from Eq. 3-31 that:
M(tme}-M(A}
and if 0 < w < 1 , then h^tme) is bounded by A^A) and A/R). A subjective probability
distribution might be then assigned to the parameter (w) to reflect one's state of
knowledge about the relationship between h^tme\ A^A) and A/R). For example, if
one believes that either the ERR or EAR model is correct AND each model is
equally plausible, then one would assign subjective probabilities of 0.5 to the
corresponding values for w.
P(w=0) = 0.5;P(w=l) = 0.5
Alternatively, if the ERR model is more plausible than the EAR model, a larger
probability would be assigned to the former: e.g.,
P(w=0) = 0.3;P(w=l) = 0.7.
On the other hand, A/fr"e) may actually be intermediate between the excess rates
calculated using the EAR and ERR models. If any such value is "equally likely,"
then the uniform distribution 11(0,1) can be assigned to the parameter w.
However, if the excess rates are more likely to be close to the rates predicted by,
say, some type of average of the two risk models, then other choices, such as a
trapezoidal distribution, Tr(a,b,c,d), might be more appropriate (see Figure 3-5).
For both distributions shown in Figure 3-5, neither of the two risk models is "on
average" closer to the truth, E[w] = 0.5, and the simple unweighted average (w*=
0.5) would arguably still be the most reasonable approach.
43
-------
Uniform
1.5
0.5
1.5
Trapezoidal
0.5
0.5 1
weight parameter
0.5
weight parameter
Figure 3-5: Examples of uniform [U(0,1)] and trapezoidal [Tr(0, 0.25, 0.75, 1 .0)]
distributions, which might be used for the risk transport weight parameter.
Probabilities for the weight parameter are equal to areas under the curve.
The BEIR VII report stated that the choice of weight of 0.7, "which clearly
involves subjective judgment, was made because mechanistic considerations...
suggest somewhat greater support for relative risk transport projection, partic-
ularly for cancer sites (such as stomach, liver, and female breast) for which
known risk factors act mainly on the promotion or progression of tumors."
Although the BEIR VII committee did not explicitly specify a subjective
distribution, any subjective distribution for the weight parameter for which E[w] is
approximately 0.7 is arguably consistent with their conclusion. The simplest
distribution with this property is the one for which:
P(w=0) = 0.3;P(w=l) = 0.7.
Another distribution for which E[w] = 0.7 is one that is U(0,1) with probability 0.5,
P(w=0) = 0.05 and P(w=\) = 0.45. The latter distribution implies that there is a
substantial probability (= 50%) that one of the two (ERR or EAR) methods for
transport would yield a very close approximation to the truth, and, if so, the ERR
is far more likely to be "correct." However, if neither model represents a good
approximation, any LAR value within the interval bounded by the two projections
would be equally plausible.
Note that for any subjective probability distribution for the parameter w,
and if w*=E[w], then the "true" value for excess risk will "on average" be equal to
the weighted arithmetic mean. That is,
= *
(3-34)
44
-------
3.11 Calculating Radiogenic Breast Cancer Mortality Risk
This section details our method for calculating radiogenic breast cancer
mortality risk and compares results with calculations based on the BEIR VII
method.
Let Mj(D,e,aj) denote the EAR for incidence at attained age a, from an
exposure at age e. If da represents an infinitesimally small age increment, the
probability of a radiogenic cancer between ages a/ and (a/ + da) would be:
fDe(aI)da=MI(D,e,aI)S(aI)/S(e)da. (3-35)
For the cancer to result in a death at age aM>a:, the patient would have
to survive the interval (a/;aM) , and then die from the cancer at age aM . This and
the concept of the relative survival rate form the basis for the method. The
relative survival rate for a breast cancer patient would be the ratio of the survival
rate for the patient divided by the expected survival rate (without breast cancer).
Assume the relative survival depends only on the length of the time interval and
the age of diagnosis. Let t = au-alt and let R(t,a,} be the relative survival
function. Then the probability of survival with breast cancer for the interval
(a,,aM) is S(aM)IS(aI)R(t,aI).
Suppose the breast cancer mortality rate (h) among those with breast
cancer depends on the age of diagnosis but does not depend on other factors,
such as whether the cancer is radiogenic, or on attained age. Then the proba-
bility of a radiogenic breast cancer death between ages aM and (aM + da) can be
shown to equal:
( i °M
fDe(aM)da = \ — f h(aM)M1(D,e,a1)S(aM)R(t,a1}da1
°M
da. (3-36)
The LAR for breast cancer mortality for an exposure at age e is:
110
LAR(D,e) = j fD,(aM}daM , (3-37)
e+L
and Eq. 3-38 is applied as before to calculate the LAR for the U.S. population.
45
-------
110-L
J S(e)-LAR(D,e)-de
LAR(D, stationary) = -2 — (3-38)
J S(e}de
For these calculations, we used the 5-y relative survival rates given in
Table 3-11 (Ries and Eisner, 2003) and assumed that breast cancer mortality
rates (for those with breast cancer) depend only on age at diagnosis and are
equal to:
/i(a/) = -(0.2)log/Z(5,a/) (3-39)
It should be noted that results from several studies indicate that, for most stages,
breast cancer mortality rates are not highly dependent on time since diagnosis -
at least for the first 10 years (Bland et a/. 1998, Cronin et a/. 2003). Thus, for
these calculations, we assumed that relative survival rates depend on time since
diagnosis as in Eq. 3-40.
R(t, a,) = exp[-f • h(a,)] (3-40)
Table 3-11: Female breast cancer cases and 5-y relative survival rates by
age of diagnosis for 12 SEER areas, 1988-20011
... ~ Relative Survival
Age (y) Cases Rates (%)
20-342
35-39
40-44
45-49
50-54
55-59
60-64
65-69
70-74
75-79
80-84
85+
Total
6,802
12,827
24,914
33,784
34,868
32,701
32,680
34,435
32,686
27,134
17,475
12,457
302,763
77.8
83.5
88.0
89.5
89.5
89.6
90.1
91.0
91.8
91.4
90.7
86.6
89.3
Adapted from Table 13.2 in Ries and Eisner (2003)
2Forages of exposure < 20, 5-y relative survival rate of 77.8% was assumed.
46
-------
Based on the method just outlined, the LAR for breast cancer mortality is
0.95x1 0"2 Gy"1. This is about 30% larger than in BEIR VII. Much of the
discrepancy between the two sets of results can be attributed to observed
increases in breast cancer incidence rates and declines in mortality rates. From
1980 to 2000, age-averaged breast cancer incidence rates (per 100,000 women)
increased by about 35% (102.2 to 136.0), whereas the mortality rates declined by
about 15% (31 .7 to 26.6) (Ries et a/. 2008).
To understand the effect these trends in incidence and mortality have on
the BEIR VII LAR projection for mortality, recall the BEIR VII formula:
M(D,e,d)= EAR(D,e,d)^^- . (3-41)
The underlying assumptions are that: a) the absolute risk of radiogenic cancer
death from an exposure at age e is equal to the absolute risk of a radiation-
induced cancer multiplied by a lethality ratio (that depends on attained age) and
b) lethality ratios can be approximated by current mortality to incidence rate
ratios. However, since the time between breast cancer diagnosis and death is
relatively long, lethality rates might be better approximated by comparing current
mortality rates to incidence rates observed for (much) earlier time periods. If, as
data indicate, current incidence rates are considerably higher than in the past,
the BEIR VII denominator is too large, and the estimated lethality ratio is too
small. This would result in a downward bias in the BEIR VII projection for
mortality.
Our projection has limitations which must be noted. First, its validity
depends on the extent to which estimates of relative survival functions can be
used to approximate mortality rates from breast cancer for people with breast
cancer. Long-term survival rates for breast cancer patients are desirable for
constructing valid estimates for this approach, but since these survival rates can
change rapidly, there is considerable uncertainty in extrapolating rates for
periods beyond 5-10 y. Finally, reduced expected survival among breast cancer
patients may be partly attributable to causes other than breast cancer. For
example, if some breast cancers are related to obesity, breast cancer patients as
a group may be at greater risk of dying from cardiovascular disease.
3.12 LAR by Age at Exposure
Sex-averaged LAR for incidence and mortality by age-at-exposure are
plotted in Figures 3-6 to 3-8 for selected cancer sites. More specifically, for both
males and females, LAR is calculated as described in previously according to:
110
LAR(D,e) = J M(EPA)(D,e,a)-S(a)/S(e)da , (3-42)
e+L
47
-------
where
M(£PA)^e^a} = w*\M(K)(D,e,a)] + (1 -w*)[M(A}(D,e,a)] , (3-43)
and sex-averaged LAR were calculated using Eq. 3-44:
LARAVG(D e) = 1 -04SSMALE(^L^RMALE(Ae) + SPEMALE(e}LARPEMALE(D,e) ^
(3_44)
where 1.048 is the ratio of the male to female births. Figures 3-6 to 3-8 show
that, for most cancer sites, the probability of premature cancer (or cancer death)
attributable to an acute exposure decreases with age-at-exposure. The notable
exception is leukemia mortality, for which the projected LAR increases slightly
from birth to about age 60.
For most cancers, the decrease in LAR with age-at-exposure is assumed
to be similar to the pattern shown for colon, lung, and bladder cancers: the LAR
decreases by a factor of about 2 or more from birth to age 30; it then levels off
until about age 50 and then gradually decreases towards 0. The same type of
relationship between LAR and age-at-exposure can be seen in Figure 3-9 for all
cancers combined. During the first 30 y, the decrease in LAR is almost entirely
attributable to the exponential decline in modeled age-specific ERR and EAR (in
the risk models, y< -0.3), whereas the decrease in LAR after age 50 is largely
attributable to competing risks - as people age, they have an ever-decreasing
chance of living long enough to contract a radiation-induced cancer. For breast
and thyroid cancers, the modeled age-specific ERR or EAR continue to decrease
after age 30, and the LARs do not level off after age 30. In general, the LAR
decreases more rapidly for breast, bone, thyroid, and residual cancers than for
other sites. For thyroid cancer, the modest discontinuities evident in LAR at ages
5, 10, and 15 are an artifact of the categorization used for age-at-exposure in the
thyroid risk model. Tables 3-12(a-c) and 3-13(a-c) provide sex- and age-at-
exposure-specific LAR values by cancer site.
Risks for childhood exposures are often of special interest. As shown in
Figures 3-6 through 3-8, for most cancer sites, the LAR per unit dose is sub-
stantially larger for exposures during childhood (here defined as the time period
ending at the 15th birthday) than later on in life. In addition, doses received from
ingestion or from inhalation are often larger for children than adults. Table 3-14
compares the average LAR per Gy for cancer incidence for exposures before
age 15 to the average LAR for all ages. For uniform, whole-body radiation, the
cancer risk coefficient (Gy"1) is 1.16x10"1 for people of all ages. This compares to
2.60x10"1 for exposures before age 15. The corresponding risk coefficients for
cancer mortality are 5.80x10"2 (all ages) and 1.15x10"1 (before age 15). Risks
from childhood exposures, like those for adults, are generally greater for females
48
-------
(3.29x10"1, incidence; 1.47x10"1, mortality) than for males (1.95x10~1, incidence;
v2
8.51x10 , mortality).
0.02
>, 0.015
o
g. 0.01
QL
-i 0.005
0
c
Stomach Cancer
\
-\
\
\
\
\
3 20 40 60 80
0.03
0 0.02
0)
Q.
< 0.01
0
Colon Cancer
\
\
\
\
\
^~~^\ '
3 20 40 60 80
Age at exposure Age at exposure
0.01
o
£. 0.005
5
n
Liver Cancer
\
- \
V ^-^^
0.06
O 0.04
S.
< 0.02
n
Lung Cancer
\
\
\
x
0 20 40 60 80
Age at exposure
0 20 40 60 80
Age at exposure
Figure 3-6(a): Sex-averaged LAR for incidence by age at exposure using a DDREF of 1.5
49
-------
O.OSr
O 0.02
5
Q.
< 0.01
0
0.2
I 0.1
s
0.05
0
Bladder Cancer
0 20 40 60 80
Age at exposure
Residual Cancers
0 20 40 60 80
Age at exposure
0.03r
Thyroid Cancer
O 0.02
5
Q.
< 0.01
o
0 20 40 60 80
Age at exposure
1.5
x 10
3 Bone Cancer
Q.
0.5
0
0 20 40 60 80
Age at exposure
Figure 3-6(b): Sex-averaged LAR for incidence by age at exposure: DDREF = 1.5 except
for bone cancer
Breast Cancer
Prostate Cancer
u. ^
>, 0.15
O
S. 0.1
£
-" 0.05
0
-
•\
\^^^
D 20 40 60 80
u.u^
>, 0.015
o
S. 0.01
o:
-" 0.005
0
(
\
\
\
- \
D 20 40 60 80
Age at exposure Age at exposure
8
>, 6
O
£4
{£.
<
-i 2
n
x 103Uterine Cancer
\
\
\
\
' X^\^J
0.01
o
S. 0.005
QL
_l
n
Ovarian Cancer
\
\
\
\
\
0 20 40 60 80
Age at exposure
0 20 40 60 80
Age at exposure
Figure 3-6(c): LAR for incidence by age at exposure using a DDREF = 1.5
50
-------
Stomach Cancer
Colon Cancer
0.01
8. 0.005
0.015r
O 0.01
<5
o.
< 0.005
x 10
20 40 60 80
Age at exposure
3 Liver Cancer
20 40 60 80
Age at exposure
Lung Cancer
o
-> 6
O
8.4
%
-" 2
n
\
\
~~~~~—
^-^___
u.uu
>,
O 0.04
8.
< 0.02
_i
n
\
X^
~^-^
^^-^_
20 40 60 80
Age at exposure
20 40 60 80
Age at exposure
Figure 3-7(a): Sex-averaged LAR for mortality by age at exposure using a DDREF of 1.5
6
>< „
O 4
8.
oc 2
n
x 10 3 Bladder Cancer
\
7"\
1.5
8.
< 0.5
n
x 103 Thyroid Cancer
A
/\
0 20 40 60 80
Age at exposure
Residual Cancers
0 20 40 60 80
Age at exposure
0.06
O 0.04
8.
< 0.02
x 10
4 Bone Cancer
•
>, 3
O
8.2
QL
0 20 40 60 80
Age at exposure
0
0 20 40 60 80
Age at exposure
Figure 3-7(b): Sex-averaged LAR for mortality by age at exposure: DDREF=1.5 except
for bone cancer
51
-------
Breast Cancer
0 20 40 60 80
Age at exposure
x 103 Uterine Cancer
> 1.5
o
0)
Q.
1
K
S0.5
0L
20 40 60 80
Age at exposure
x 10
3 Prostate Cancer
0)
Q.
5 1
0 20 40 60 80
Age at exposure
6r
O 4
i_
0)
Q.
x 10
3 Ovarian Cancer
0 20 40 60 80
Age at exposure
Figure 3-7(c): LAR for mortality by age at exposure using a DDREF of 1.5
52
-------
30 40 50 60
Age at exposure
Figure 3-8: LAR by age at exposure for leukemia for incidence (solid) and mortality (dashed)
using a DDREFof 1.5
40 50 60
Age at exposure
Figure 3-9: LAR for all cancers combined by age at exposure for exposures at low doses
and/or dose rates for incidence (solid) and mortality (dashed)
53
-------
Table 3-12a: LAR for cancer incidence1'2 by age at exposure for males
Age at exposure
Cancer site
Stomach
Colon
Liver
Lung
Prostate
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid3
Leukemia
Total3
0
168
342
103
320
198
219
123
1180
102
10.4
1720
2760
193
2950
5
139
292
86
268
172
188
107
653
55
8.0
917
1970
142
2110
10
114
248
71
222
148
159
58
498
44
6.1
484
1570
112
1680
15
94
210
59
185
127
135
32
394
37
4.6
256
1280
97
1370
20
77
179
49
154
110
116
23
313
31
3.5
136
1050
89
1140
1 Cases per 10,000 person-Gy.
2 DDREF of 1 .5 for sites other than leukemia, bone, and
3 Excludes nonfatal skin cancers
Table 3-12b:
Cancer site
Stomach
Colon
Liver
Lung
Breast
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid3
Leukemia
Total3
30
51
129
34
108
82
84
11
199
22
2.0
38
722
78
801
skin
40
48
126
33
107
83
84
5
174
20
1.1
10
682
79
761
50
43
117
29
104
80
81
2
142
16
0.6
3
616
83
699
: LAR for cancer incidence by age at exposure1'2
0
212
225
57
785
1260
66
91
221
386
1410
133
10.4
972
4850
173
5020
5
175
193
47
660
982
55
77
189
352
707
53
8.0
517
3500
117
3620
10
144
164
39
552
761
46
64
161
196
534
41
6.1
273
2710
88
2800
15
118
139
32
462
588
38
53
137
106
422
34
4.7
144
2130
75
2210
Age at
20
97
118
26
387
454
31
45
116
73
336
28
3.6
76
1720
69
1780
60
35
97
24
90
61
71
1
101
11
0.3
1
492
88
580
70
24
65
17
65
30
50
0
58
6
0.1
0
314
87
402
80
12
29
9
35
9
24
0
24
2
0.0
0
144
64
208
for females
exposure
30
64
84
18
272
265
21
31
84
30
213
20
2.1
21
1100
60
1160
40
61
82
18
269
146
19
28
83
12
184
17
1.2
6
920
61
981
50
55
76
16
255
72
16
24
78
4
151
14
0.6
2
764
63
827
60
46
65
14
217
32
12
17
67
1
112
10
0.3
0
594
65
659
70
33
46
10
150
12
8
11
48
0
69
5
0.1
0
393
63
456
80
18
23
6
79
4
4
5
24
0
31
2
0.0
0
195
47
242
1 Cases per 10,000 person-Gy.
2 DDREF of 1.5 for sites other than leukemia, bone, and skin
3 Excludes nonfatal skin cancers
54
-------
Table 3-12c: Sex-averaged LAR for cancer incidence1'2 by age at exposure
Age at exposure
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin3
Solid
Leukemia
Total3
0
190
285
81
547
614
101
32
44
220
252
1290
117
10.4
1360
3780
183
3970
5
157
244
67
459
480
88
27
38
188
227
680
54
8.0
722
2720
130
2850
10
129
207
55
383
372
75
22
31
160
126
515
43
6.1
381
2130
101
2230
15
106
175
46
320
288
65
18
26
136
68
408
36
4.7
201
1700
86
1780
20
87
149
38
268
222
56
15
22
116
47
324
30
3.5
106
1380
79
1460
30
58
107
26
188
130
42
10
15
84
21
206
21
2.0
30
910
69
979
40
55
104
25
187
72
42
9
14
83
8
179
19
1.1
8
799
70
870
50
49
97
23
179
36
40
8
12
80
3
146
15
0.6
2
690
73
763
60
41
81
19
154
16
30
6
9
69
1
106
10
0.3
1
543
77
620
70
29
55
13
110
6
14
4
6
49
0
64
6
0.1
0
356
75
430
80
15
26
7
60
2
4
2
3
24
0
28
2
0.0
0
173
54
227
1 Cases per 10,000 person-Gy.
2 DDREF of 1.5 for sites other than leukemia, bone, and skin
3 Excludes nonfatal skin cancers
55
-------
Table 3-13a: LAR for cancer mortality1'2 by age at exposure for males
Age at exposure
Cancer site
Stomach
Colon
Liver
Lung
Prostate
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid3
Leukemia
Total3
1 Deaths per 10
2DDREFof 1.5
Table 3-13b:
0
85
154
79
293
27
43
6.2
388
26
3.6
0.5
1110
65
1170
5
71
131
65
245
24
37
5.4
248
18
2.8
0.3
847
65
912
,000 person-Gy.
for sites other than
LAR for
10
58
112
54
203
20
31
2.9
195
15
2.1
0.1
693
65
758
15
48
95
45
169
17
27
1.6
160
12
1.6
0.1
576
63
638
20
39
81
37
141
15
23
1.1
134
10
1.2
0.0
482
61
542
leukemia, bone, and
cancer mortality1
30
26
58
26
99
11
17
0.6
93
7
0.7
0.0
338
58
396
skin
'2 by age at
Age at
Cancer site
Stomach
Colon
Liver
Lung
Breast
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid
Leukemia
Total
1 Deaths per 10
2DDREFof1.5
0
113
96
48
642
431
17
56
58
19
498
29
3.6
0.3
2010
53
2060
5
93
82
40
539
336
14
47
50
18
301
16
2.8
0.2
1540
51
1590
,000 person-Gy.
for sites other than
10
77
70
33
450
260
12
40
42
10
233
13
2.1
0.1
1240
50
1290
15
63
59
27
376
200
10
34
36
5
190
10
1.6
0.0
1010
49
1060
20
52
50
22
315
153
8
29
30
4
157
9
1.3
0.0
831
48
878
leukemia, bone, and
40
25
57
25
98
11
17
0.3
88
7
0.4
0.0
329
61
390
50
22
54
24
95
12
17
0.1
77
6
0.2
0.0
307
67
375
60
19
47
21
84
12
16
0.0
59
5
0.1
0.0
262
76
339
70
14
34
16
63
11
14
0.0
38
3
0.0
0.0
192
80
272
80
8
19
9
35
7
10
0.0
18
1
0.0
0.0
108
63
170
exposure for females
exposure
30
34
36
15
221
85
5
20
22
2
108
6
0.7
0.0
556
45
601
skin
40
33
35
15
219
42
5
20
22
1
100
6
0.4
0.0
499
48
547
50
30
33
14
210
17
5
18
22
0
88
5
0.2
0.0
444
52
496
60
26
30
13
183
6
4
15
21
0
70
4
0.1
0.0
372
57
429
70
20
23
10
135
2
3
10
18
0
48
3
0.0
0.0
273
58
331
80
13
15
6
77
0
2
5
13
0
24
1
0.0
0.0
156
47
203
56
-------
Table 3-13c: Sex-averaged LAR for cancer mortality1'2 by age at exposure
Age at
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid
Leukemia
Total
1 Deaths per
2 DDREF of 1
0
99
126
64
463
210
14
8
27
51
13
442
27
3.6
0.4
1550
59
1610
5
82
107
53
389
164
12
7
23
43
11
274
17
2.8
0.2
1190
58
1240
10,000 person-Gy.
.5 for sites other than
10
67
91
44
324
127
10
6
20
37
6
214
14
2.1
0.1
961
57
1020
15
55
77
36
270
98
9
5
17
31
3
175
11
1.6
0.1
789
56
845
leukemia, bone
20
45
66
30
226
75
8
4
14
26
2
145
10
1.2
0.0
652
54
707
, and
exposure
30
30
47
20
159
42
6
3
10
19
1
101
7
0.7
0.0
445
52
497
skin
40
29
47
20
158
21
6
3
10
19
0
94
6
0.4
0.0
413
55
468
50
26
44
19
153
9
6
2
9
19
0
83
6
0.2
0.0
375
60
435
60
23
38
17
134
3
6
2
7
18
0
65
4
0.1
0.0
318
67
384
70
17
29
13
100
1
5
2
5
16
0
43
3
0.0
0.0
234
69
303
80
11
17
7
59
0
3
1
3
11
0
22
1
0.0
0.0
135
54
189
57
-------
Table 3-14: LAR for cancer incidence1'2 for lifelong and childhood
exposures
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid3
Leukemia
Total3
Males
62
146
40
130
—
89
—
—
97
22
251
24
2.4
182
863
92
955
Lifelong exposure
Females
75
92
21
308
289
—
23
33
92
65
259
22
2.3
96
1280
69
1350
Exposures before
Sex-
averaged
68
119
30
220
146
44
12
17
95
44
255
23
2.39
138
1080
80
1160
Males
128
272
79
247
—
161
—
—
175
81
616
53
7.2
773
1820
132
1950
Females
161
179
43
611
885
—
51
71
176
265
675
53
7.2
436
3180
108
3290
age 15
Sex-
averaged
144
227
62
425
433
82
25
35
175
171
645
53
7.2
608
2480
120
2600
1 Cases per 10,000 person-Gy for a stationary population.
2 DDREFof 1.5 for sites other than leukemia, bone, and skin
3 Excludes nonfatal skin cancers
3.13 Summary of Main Results
New EPA LAR projections for incidence and mortality are given in Tables
3-15 and 3-16. The tables also provide 90% uncertainty intervals for the LAR. As
described in Section 4, a 90% uncertainty interval would be any interval which
contains the parameter of interest, e.g., the LAR, with a probability of 0.90 -
based on all that is known about the LAR from analyses of epidemiologic data
and additional sources of information on how radiogenic risk depends on dose
rate and other factors. The uncertainty intervals were calculated using Bayesian
methods, which involved a somewhat complex (Markov Chain) Monte Carlo
method for simulating site-specific LAR values. This approach allowed for the
quantification of uncertainties associated with sources such as: 1) sampling
variability, 2) transport of risk estimates from the Japanese A-bomb survivor
population, 3) uncertainty associated with the DDREF, and 4) dosimetry errors.
58
-------
Table 3-15: LAR projections for incidence1'2
Males
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
(Skin)
Solid4
Leukemia
Total4
LAR
62
146
40
130
—
89
—
—
97
22
251
24
2.4
182
863
92
955
90% Ul
(8, 220)
(40, 230)
(6, 110)
(58, 320)
—
(0,410)
—
—
(27, 230)
(5, 54)
(99, 610)3
—
—
(27,210)
(430, 1810)
Females
LAR
75
92
21
308
289
0
23
33
92
65
259
22
2.3
96
1280
69
1350
90% Ul
(9, 220)
(37,210)
(4, 88)
(95, 540)
(140,570)
—
(0, 130)
(1 1 , 82)
(14, 130)
(21 , 240)
(120, 700)3
—
—
(18, 160)
(650, 2520)
Sex-averaged
LAR
68
119
30
220
146
44
12
17
95
44
255
23
2.4
138
1080
80
1160
90% Ul
(9, 220)
(42, 220)
(6, 94)
(83, 420)
(70, 290)
(0, 200)
(0, 65)
(5, 42)
(24, 170)
(15, 140)
(120, 630)3
—
—
(29, 160)
(560,2130)
1 Cases per 10,000 person-Gy.
2 DDREFof 1.5 for sites other than leukemia, bone, and skin
3 Interval for residual, kidney and bone cancer cases combined
4 Excludes skin cancers
59
-------
Table 3-16: LAR projections for mortality1'2
Males
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid4
Leukemia
Total4
LAR
32
67
31
120
0
14
—
—
20
1.1
110
8.3
0.9
0.05
404
65
469
90% Ul
(4, 110)
(18, 110)
(5, 83)
(54, 290)
—
(0, 62)
—
—
(6, 48)
(0.3, 3)
(42, 260)3
—
—
(19, 150)
(230, 880)
Females
LAR
41
41
18
255
95
0.0
6.4
22
26
3.2
125
7.0
0.8
0.03
639
50
689
90% Ul
(5, 120)
(16,94)
(4, 76)
(78, 450)
(45, 190)
—
(0, 36)
(7, 56)
(4, 37)
(1,12)
(57, 330)3
—
—
(13, 110)
(320, 1230)
Sex-averaged
LAR
36
54
25
188
48
6.8
3.2
11
23
2.2
117
7.7
0.8
0.04
523
57
580
90% Ul
(5, 120)
(19,97)
(5, 77)
(72, 360)
(23, 95)
(0, 31)
(0,18)
(4, 28)
(6, 40)
(0.7, 7)
(55, 280)3
—
—
(20, 110)
(280, 1040)
1 Deaths per 10,000 person-Gy.
2 DDREFof 1.5 for sites other than leukemia, bone, and skin
3 Interval for residual, kidney and bone cancer deaths combined
4 Excludes skin cancers
For most cancer sites, BEIR VII derived parameter estimates for ERR and
EAR models based on a statistical analysis of LSS data that was cross-classified
by city, sex, dose, and intervals based on age-at-exposure, attained age, and
follow-up time. Sampling variability refers to the uncertainty in parameter esti-
mates associated with the variation in the observed numbers of cancer cases or
deaths within each of these subgroups. In contrast to BEIR VII, our uncertainty
analysis at least partially accounted for the sampling variability associated with
the site-specific risk model parameters for age-at-exposure and attained age.
Transport of risk estimate uncertainty refers to uncertainty associated with how to
apply the results from the analysis of the Japanese LSS cohort data to the U.S.
The ratio of LAR projections based on the EAR model divided by the projection
based on the ERR model is a crude indicator of the magnitude of this uncertainty.
It follows that "transport" uncertainty is greatest for sites such as stomach and
prostate cancer, for which Japanese and U.S. baseline rates are vastly different.
60
-------
A dominant source of uncertainty for all cancers combined is that
associated with the value of the DDREF. This includes some of the uncertainty
associated with the shape of the dose-response function at very low doses. As
discussed in Section 4, it does not incorporate uncertainty associated with the
validity of the assumption that the linear portion of the dose-response function
fitted to the LSS data can be equated to the response that would be observed at
lower doses or for chronic exposures. Additional sources of uncertainty, including
dosimetry errors, were also incorporated into the uncertainty analysis. Details are
provided in Section 4.
The new EPA risk projection is 955 cancer cases per 10,000 person-Gy
for males, and 1350 cancer cases for females. The 90% uncertainty intervals
suggest these projections are accurate to within a factor of about 2 or 3.
Uncertainties, as measured by the ratio of the upper to lower uncertainty bounds,
are greatest for stomach, prostate, uterine, bladder, liver, and thyroid cancers.
In the first four columns in Table 3-17, the new EPA risk projections for
incidence are compared to risk projections in the current (1999) version of FGR-
13. For all cancers other than esophagus, uterus, prostate, and residual site
cancers (which are defined differently for the two sets of projections), the EPA
risk projection for both males and females is about 35% higher than in FGR-13.
Cancer sites for which the relative change from the projected LAR in FGR-13 is
greatest include: female colon (|), female lung (|), female bladder (|), kidney (|),
and liver (|).
For the current version of FGR-13, the risk models were applied to 1989-
1991 mortality data to first derive projections for radiogenic cancer mortality. For
risk projections for cancer morbidity, the risk projections were then multiplied by
the inverse of cancer specific lethality ratios. For example, for ovarian cancers, it
was assumed that 70% of the radiogenic cancers would be fatal. The last two
columns of Table 3-17 show what the new EPA risk projections would be if the
new risk models were applied to baseline incidence rates derived from the same
1999-2001 mortality data used for FGR-13. These calculations indicate that the
overall increase in LAR for incidence is due to both changes in the risk models
(predominantly due to a reduction in the nominal DDREF for most cancer sites
from 2 to 1.5) and, for most cancers, increases in the baseline rates (and survival
probabilities) to which these models were applied. It is important to realize that
the data on baseline rates are not strictly comparable, in that the data were
derived from different sources (incidence data from SEER registries versus U.S.
mortality data and lethality ratios), and that it is not appropriate to conclude that
incidence rates actually increased for each of the cancers shown in Table 3-17.
61
-------
Table 3-17: Comparison of EPA and FGR-13 LAR projections for incidence1
New EPA
FGR-13 (1999)
New risk models
applied to 1989-1991
mortality & lethality
data
Cancer site
Stomach
Colon
Liver
Lung
Breast
Ovary
Bladder
Thyroid
Kidney
Bone
Leukemia
Sum of cancers
listed above2
Esophagus
Prostate
Uterus
Residual3
Total4
Males
62
146
40
130
Not
provided
—
97
22
24
2.4
92
675
Not
provided
89
—
251
955
Females
75
92
21
308
289
33
92
65
22
2.3
69
1070
Not
provided
—
23
259
7350
Males
36
152
19
81
Not
provided
—
66
21
10
1.3
65
457
7.7
Not
provided
—
191
657
Females
54
225
12
126
198
42
30
44
6
1.4
48
786
17
Not
provided
229
7030
Males
56
140
32
126
Not
provided
—
49
19
12
2
71
507
No direct
Females
72
88
20
267
287
32
59
29
11
2
56
923
comparison
for these sites
1 Cases per 10,000 person-Gy for low dose and/or chronic exposures
2 Excludes esophagus, prostate, uterine, and other "residual-site" cancers not specified here, and
skin cancer. FGR-13 did not provide an LAR projection for nonfatal skin cancer incidence.
3 Defined differently for new EPA projections and FGR-13.
4 Excludes nonfatal skin
Table 3-18 gives the LAR projections for mortality. From the first four
columns of results, the largest relative changes in LAR compared to the
projections in FGR-13 were for female colon (|), female lung (|) and skin (|)
cancers. A comparison of results in the last four columns - derived using the
same (1989-1991) mortality data - indicates that the effect of changes in the risk
models, mostly associated with the DDREF, was to increase risk projections by
about 20%. However, since the new EPA projections were based on mortality
rates that tended to be smaller by almost the same percentage, the LAR for all
62
-------
sites combined barely changed, i.e., from 462 to 469 per 10,000 person-Gy for
males and 683 to 689 for females.
Table 3-18 Comparison of EPA and FGR-13 LAR projections for mortality1
New risk models
New EPA
Cancer site
Stomach
Colon
Liver
Lung
Breast
Ovary
Bladder
Thyroid
Kidney
Bone
Leukemia
Sum of cancers
listed above2
Esophagus
Prostate
Uterus
Residual
Skin
Total
Males
32
67
31
120
Not
provided
—
20
1.1
8.3
0.9
65
346
Not
provided
14
—
109
0.05
469
Females
41
41
18
255
95
22
26
3.2
7.0
0.8
50
558
Not
provided
—
6.4
125
0.03
689
FGR-13
Males
33
84
18
77
Not
provided
—
33
2.1
6.4
0.9
65
319
7.3
Not
provided
—
135
1.0
462
Females
49
124
12
119
99
29
15
4.4
3.9
1.0
47
503
16
—
Not
provided
163
1.1
683
applied to 1989-1 991
mortality data
Males
50
77
31
120
Not
provided
—
24
1.9
8.0
0.9
69
382
Females
64
49
19
254
94
22
30
2.9
7.3
0.8
55
598
No direct comparison
for these sites
1 Deaths per 10,000 person-Gy for low dose and/or chronic exposures
2 Excludes esophagus, prostate, uterine, skin, and "residual-site" cancers not specified here.
3 Defined differently for new EPA projections and FGR-13.
Table 3-19 summarizes the sex-averaged LAR projections for cancer
incidence and mortality. Table 3-20 compares the new EPA LAR projections with
projections in BEIR VII. For some sites such as stomach, liver and prostate
cancers, which have very different baseline rates in U.S. compared to Japan, the
new EPA projections are substantially larger. This is due to EPA's adoption of the
weighted arithmetic mean for combining results derived from ERR and EAR
63
-------
models. For some other sites, in part because of our use of a stationary
population, EPA's projections tended to be slightly smaller.
Finally, Table 3-21 provides estimates of LAR for a (non-stationary) popu-
lation, in which the number of males and females at each age is based on the
2000 Census. Results given in this table are appropriate for assessing risks for
certain types of exposures to mixed populations with demographics similar to the
one targeted by the 2000 Census. Compared to the stationary population, the
census population contains a somewhat larger proportion of younger people.
Since projected radiogenic risks decrease with age-at-exposure, the LARs given
in Table 3-21 are slightly larger than LARs given in other tables of this report for
stationary populations. For example, the sex-averaged LAR for uniform whole-
body dose is 1.24xlO"2 Gy"1 for the census population as compared to the corre-
sponding LAR of 1.16x10"2 Gy"1 given in Table 3-19.
Table 3-19: Sex-averaged LAR projections for incidence and mortality1
Incidence
Mortality
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid
Leukemia
Total3
Projection
68
119
30
220
146
44
12
17
95
44
255
23
2.4
138
1080
80
1160
90% Ul
(9, 220)
(42, 220)
(6, 94)
(83, 420)
(70, 290)
(0, 200)
(0, 65)
(5, 42)
(24, 170)
(15, 140)
(120, 630)2
(29, 160)
(560,2130)
Projection
36
54
25
188
48
6.8
3.2
11
23
2.2
117
7.7
0.8
0.04
523
57
580
90% Ul
(5, 120)
(19,97)
(5, 77)
(72, 360)
(23, 95)
(0, 31)
(0, 18)
(4, 28)
(6, 40)
0.7, 7)
(55, 280)2
(20, 110)
(280, 1040)
1 Cases or deaths per 10,000 person-Gy
2 Interval for residual, kidney and bone cancer deaths combined
3 Excludes nonfatal skin cancers
64
-------
Table 3-20: Comparison of EPA and BEIR VII LAR calculations
Incidence1'2
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Solid cancers
Leukemia
Total
Sex
M
F
M
F
M
F
M
F
F
M
F
F
M
F
M
F
M
F
M
F
M
F
M
F
M
F
M
F
EPA
62
75
146
92
40
21
130
308
289
89
23
33
97
92
22
65
251
259
24
22
2.5
2.3
863
1280
92
69
955
1350
BEIR VII
34
43
160
96
27
12
140
300
310
44
20
40
98
94
21
100
290
290
None
None
None
None
800
1310
100
72
900
1382
Mortality1'2
EPA
32
41
67
41
31
18
120
255
95
14
6.4
22
20
26
1.1
3.3
109
124
8.3
7.0
0.9
0.8
404
639
65
50
469
689
BEIR VII
19
25
76
46
20
11
140
270
73
9
5
24
22
28
None
None
120
132
None
None
None
None
410
610
69
52
479
662
Cases or deaths per 10,000 person-Gy
' DDREF of 1.5 for sites other than leukemia
65
-------
Table 3-21: LAR incidence and mortality projections for a population based
on 2000 census data1'2
Males
Cancer site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual
Kidney
Bone
Skin
Solid3
Leukemia
Total3
Incidence
66
156
42
138
0
96
0
0
103
24
273
27
2.7
199
928
93
1020
Mortality
34
71
33
127
0
14
0
0
21
1.2
118
9.0
0.9
0
429
64
494
Females
Incidence
80
98
22
327
316
0
25
35
98
70
279
23
2.6
103
1380
71
1450
Mortality
43
43
19
269
104
0
6.7
24
27
3.5
133
7.5
0.9
0
680
50
730
Sex-averaged
Incidence
73
127
32
234
160
47
12
18
100
48
276
25
2.6
150
1150
82
1240
Mortality
39
57
26
199
53
7.0
3.4
12
24
2.4
126
8.2
0.9
0
556
57
613
1 Cases or deaths per 10,000 person-Gy.
2 DDREFof 1.5 for sites other than leukemia, bone, and skin
3 Excludes skin cancers
66
-------
3.14 Comparison with Risk Projections from ICRP and UNSCEAR
This section compares the new EPA risk models to the risk models used
in recent reports of the ICRP (2007) and UNSCEAR (2008). For most cancer
sites, UNSCEAR and ICRP ERR and EAR risk models were derived from
analyses of recent A-bomb survivor data with DS02 doses. As in BEIR VII, most
ICRP models were based on 1958-1998 incidence data, whereas the UNSCEAR
models were based on 1950-2000 mortality data. ICRP models were applied to a
mix of Euro-American and Asian populations; the UNSCEAR models were
applied to 5 populations (China, Japan, Puerto Rico, U.S., and United Kingdom).
3.14.1 ICRP risk models. For most cancer sites, the ICRP risk project-
ions are based on an approach very similar to that used by both EPA and BEIR
VII. For all three, an approximate LNT dose-response is assumed at very low
doses and dose rates. ICRP projections were based on a weighted average of
ERR and EAR risk model projections and a DDREF (of 2 instead of 1.5). For
most sites, ICRP used a weight (w) of 0.5 for the ERR model; exceptions include
breast, bone, and leukemia cancers (w=0), thyroid cancer (w=1), and lung cancer
(w=0.3). In the ICRP risk models, the dose-response for most solid cancer sites
is modified according to functions of age-at-exposure and attained age, which
are of similar or identical form to those used here and in BEIR VII. In ICRP, the
ERR and EAR for most solid cancer sites decrease with age-at-exposure by
about 17% (ERR) or 24% (EAR) per decade (even beyond age 30); in BEIR VII,
the per decade decrease for exposures before age 30 was somewhat steeper,
typically 26% (ERR) or 34% (EAR), but there is no decrease in risk with age-at-
exposure after age 30. A more detailed comparison of risk model parameter
values for solid cancers is given in Table 3-22. ICRP has separate risk models
for the most of the cancers with risk projections in this report. However, there is
an ICRP model for esophageal cancer and none for kidney, prostate, or uterine
cancers.
Table 3-23 compares LAR projections for the U.S. population - calculated
using ICRP and EPA risk models and 1998-2002 incidence data. The EPA pro-
jections tend to be somewhat larger, although much of the difference can be
attributed to the EPA's smaller nominal value for the DDREF (1.5 vs. 2). For the
vast majority of sites, the ICRP and EPA risk projections are well within a factor
of 2 of each other. Some of the largest differences are for lung, "other" solid (not
directly comparable), kidney, and leukemia, but even these differences are small
when compared to uncertainties associated with these risks. For leukemia, EPA's
risk projection is larger because EPA assigns a larger weight (0.7 vs. 0.5) to its
ERR model and baseline rates are higher in the U.S. than in Japan. ICRP risk
projection for skin cancer (see Section 3.3) is much larger than EPA's.
67
-------
Table 3-22: Comparison of ICRP (2007) and EPA risk model parameter
values for solid cancers
ERR Model
EAR Model
ICRP
EPA
ICRP
EPA
Linear Dose Response Parameter
Cancer Site
P^
Esophagus
Stomach
Colon
Liver
Lung
Breast1
Prostate
Uterus
Ovary
Bladder
Thyroid
Other solid
All but Esophagus,
Breast, Bladder,
Thyroid, Other solid
Esophagus
Lung
Breast
Bladder
Thyroid
Other Solid
All but Liver, Lung,
Breast, Bladder,
Thyroid, Other Solid
Liver
Lung
Breast1
Thyroid
Bladder
Other Solid
0.52 0.84 None
0.30 0.49 0.21 0.48
0.88 0.43 0.63 0.43
0.32 0.52 0.32 0.32
0.37 1.8 0.32 1.4
Not used
None 0.12
None 0.055
0.41 0.38
0.86 1.42 0.5 1.65
0.53 1.05 See text
0.28 0.22 0.27 0.45
0.33 0.46
4.6 6.4
4.0 1.7
2.9 0.9
3.4 4.7
— 10
None
None
1.0
0.75 1.0
Not
5.2 7.2
None
4.9 4.9
3.2 1.6
2.2 1
2.3 3.4
— 9.9
0.11
1.2
0.7
1.2 0.75
used
6.2 4.8
Age-at-exposure: per decade % change in ERR or EAR
100(l-exp(Y))
-26 for age<30;
0 otherwise
-17 None
-26 for age < 30;
0 otherwise
Not used
-26 for age <30;
0 otherwise
-56 See text
-26 for age < 30;
0 otherwise
-24
64
1
-39
-11
Not
-24
-34 for age < 30;
0 age > 30
None
-34 for age < 30;
0 age > 30
-40
-34 for age < 30;
0 age > 30
used
-26 for age < 30
0 age > 30
Power of attained age by which EAR varies ( // )
-1.65 -1.4
-1.65 -1.4
-1.65 -1.4
Not used
0 0
-1.65 -1.4
-1.65 -2.8
2.38
2.38
4.25
See text
Not
6.39
2.38
2.8
4.1
5.2
See text
used
6.0
2.8
ICRP and EPA use essentially the same model for female breast cancer (see Section 3.2).
68
-------
Table 3-23: Comparison of EPA and ICRP (2007) risk models: Projections
of incidence for chronic exposures to the U.S. population1'2
Cancer
Males
ICRP
EPA
Females
ICRP
EPA
Esophagus
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Other Solid
Kidney
Bone
Leukemia
Skin
153
48
100
32
87
—
No model
—
—
65
16
1573
13
2.0
483
10003
No model
62
146
40
130
—
89
—
—
97
22
251
24
2.4
93
182
163
74
46
13
207
230
—
No model
22
50
83
1313
10
2.0
363
10003
No model
75
92
21
308
289
—
23
33
92
65
259
22
2.3
71
96
1 Number of cases per 10,000 person-Gy
2 ICRP" projections for sites other than esophagus, leukemia and skin calculated using models
summarized in Table 3-18, a DDREF of 2, 1998-2002 SEER incidence data, and 1999-2001
U.S. life table data
3 ICRP projections for Euro-Asian population (ICRP 2007, Table A.4.14, p. 209)
3.14.2 UNCSCEAR risk models. Comparisons with the models used by
UNSCEAR (2008) are somewhat more complicated than for ICRP. The form of
the UNSCEAR ERR and EAR models depends on cancer site. For most cancer
sites, the models found to best fit the A-bomb survivor cancer incidence data
were LNT models for which radiogenic risk is modified only by attained age. For
many other cancer sites, the slope of the dose-response is also modified by sex
and/or TSE. In contrast to the BEIR VII models, age-at-exposure is seldom used
as a dose effect modifier - exceptions are the EAR and ERR models for thyroid
cancer and the ERR model for brain/CNS cancers. A summary of the UNSCEAR
risk models is given in Table 3-24. For the risk transport problem, UNSCEAR did
not recommend a method for combining site-specific ERR and EAR risk
projections. Although a value for the DDREF was not formally adopted, it was
noted that "values of DDREF of about 2, recommended by others [e.g., ICRP],
are consistent with...a large body of epidemiological and experimental data."
UNSCEAR provided separate risk models for cancers of the esophagus,
brain/CNS, bone, skin (nonmelanoma), and for all other BEIR VII cancer sites
except prostate, uterus and ovary.
69
-------
Table 3-24: Summary of UNSCEAR (2008) risk models for solid cancer
incidence and leukemia mortality
Cancer
Esophagus
Stomach
Colon
Liver
Lung
Female Breast
Bladder
Brain and CNS
Thyroid
Leukemia
Bone
Skin
Dose Response
Linear
Linear
Linear
Linear
Linear
Linear
Linear
Linear
Linear
ERR Linear-quadratic1
(Fit using Bayesian methods)
(Pure) quadratic
Quadratic-exponential2
Effect Modifiers
None
Attained age
ERR: Attained age
EAR: TSE
ERR: None
EAR: Attained age
ERR: Sex
EAR: Sex, Attained age
ERR: Attained age
EAR: TSE
ERR: None
EAR: Attained age
ERR: Age-at-exposure
EAR: None
ERR: Age-at-exposure, Attained
age
EAR: Sex,
Age-at-exposure
ERR: Attained age
EAR: Sex, TSE
ERR: Attained age
EAR: None
ERR: TSE, Attained age
EAR: TSE
1 One of several alternative models for leukemia fit using Bayesian methods
2 Product of quadratic and exponential functions of dose
Table 3-25 compares LAR projections for chronic exposures to the U.S.
population - calculated using UNSCEAR and EPA risk models. The UNSCEAR
ERR-model projection for all solid cancers combined is almost twice as large as
the corresponding EPA projection. However, much of this difference is due to the
much larger UNSCEAR projection for breast cancer, which in contrast to EPA's
projection, was based entirely on an analysis of A-bomb survivor data. Although
the models are often of quite different form, the UNSCEAR and EPA EAR risk
projections are often remarkably consistent, with almost identical projections for
all solid cancers combined: 1.17xl(T1 (UNSCEAR) vs. 1.04xl(T1 (EPA).
However, this last observation may be a bit misleading, since for the UNSCEAR
projections there was no explicit DDREF adjustment, and EPA applies a DDREF
of 1.5 for most cancer sites. Finally, we note that EPA's projections for skin
cancer risk are larger than UNSCEAR's (see Section 3.3).
70
-------
Table 3-25: EPA and UNSCEAR (2008) sex-averaged cancer incidence risk
projections from chronic exposures to the U.S. population1'2
Cancer site
ERR
EAR
UNSCEAR
EPA
UNSCEAR
EPA
Esophagus
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Other Solid
Kidney
Bone
Brain/CNS
Skin
Solid Total3
Leukemia
Total3
23
20
174
18
441
638
No model
No model
No model
184
118
408
No model
2
32
36
2095
55s
2150
No model
17
132
12
322
No model
62
6
17
106
44
283
25
2.4
No model
138
11804
97
12804
5
249
152
73
202
141
No model
No model
No model
81
86
165
No model
0
17
1
1171
Not used
No model
188
89
72
177
146
1.9
25
15
69
No model
190
18
2.4
No model
No model
1040s
42
1080s
1 Number of cases per 10,000 person-Gy
2 UNSCEAR (2008) solid cancer projections (Table 70, p. 254) for test doses of 0.01 Sv
3 Does not include skin cancer
4 Based on EAR projections for bone and breast cancer and ERR projections for all other sites
5 Based on ERR projection for thyroid cancer and EAR projections for all other sites
6 Mortality risk (deaths per 10,000 person-Gy) from Table 66 in UNSCEAR (2008) for a test dose
of 0.01 Gy, and based on a model fit using Bayesian methods. UNSCEAR did not provide risk
projections for leukemia incidence.
71
-------
4. Uncertainties in Projections of LAR for Low-LET Radiation
4.1 Introduction
This chapter describes uncertainties relating to the LAR projections given
in Section 3. After a brief description of sources of uncertainty (Section 4.2), a
simple analysis is presented to gain insight as to how large the uncertainties
might be for three of the most important sources: sampling errors in the epidem-
iological data underlying the risk models, the DDREF, and risk transport of the
radiogenic risks estimated from the cohort of Japanese A-bomb survivors to the
U.S. population. In this initial examination of uncertainties, the LAR is calculated
for ranges of "plausible" values for parameters in the ERR model, the DDREF,
and the weight assigned to the ERR model. Each parameter is varied in
sequence (one-at-a-time), while other parameters are set to nominal values, and
the corresponding range of LAR values is examined using graphical methods. (In
this Section, the term nominal value refers to the value for a parameter used in
Section 3 for calculating projections of radiogenic risk: e.g., for most cancer sites,
the nominal values are -0.3 for the age-at-exposure parameter and 1.5 for the
DDREF).
As discussed in Section 4.3, results indicate that for some cancers (e.g.,
bladder cancer) the (sampling) uncertainty associated with the linear dose-
response parameter dominates, whereas for others (e.g., stomach cancer, for
which baseline rates are much larger in Japan than in the U.S.) uncertainty
associated with risk transport is greatest. Colon cancer is an example for which
the DDREF is one of the most important sources of uncertainty, whereas for
prostate cancer, the uncertainty associated with both risk transport and sampling
errors are especially large. A problem with the simple approach is that it does not
adequately account for the combined effect of uncertainties associated with
several parameters.
In Section 4.4, a more sophisticated Monte Carlo approach is introduced,
which generates 90% uncertainty bounds associated with the sex- and cancer
site-specific LARs. The approach is similar to those used elsewhere, e.g., for the
Interactive RadioEpidemiological Program (IREP, see Kocher et al. 2008).
Probability distributions are assigned to parameter values associated with each
of several relevant sources of uncertainty, and Monte Carlo methods are used to
generate uncertainty bounds for quantities of interest. In our application of Monte
Carlo, the joint probability distribution of parameter values associated with the
ERR model and non-sampling sources of uncertainty are simulated through
repeated random sampling. Then, sex- and site-specific LAR values are calcu-
lated for each set of simulated parameter values, and 90% uncertainty bounds
are equal to the 5th and 95th percentile values of the simulated LAR values.
The fundamental difference between this approach and IREP's is that a
formal Bayesian analysis is used here to approximate probability distributions
72
-------
associated with sampling variability. First, initial (prior) subjective probability
distributions were assigned to each parameter in the risk models, i.e., the linear
dose-response parameter (/?), age-at-exposure parameter (/), and attained-age
parameter (77). Then, information on radiogenic risks from the LSS data was
applied to update these distributions. The Bayesian analysis of the LSS data is
described in Section 4.4 and in more detail in Appendix B. A Bayesian approach
for evaluating uncertainties in risk projections has also been used for the
UNSCEAR 2006 report (Little 2008).
For most cancer sites, the risk models used for this uncertainty analysis
are the same ERR models that BEIR VII fit to the LSS data. However, there are
two important differences between the two approaches. First, BEIR VII used
classical statistical methods to derive "best" estimates for the parameters which
describe how ERR depends on dose, age-at-exposure and attained age. In
contrast, we assigned (prior) probability distributions to these parameters and
then applied information gleaned from the LSS to update these distributions.
Second, for most sites, our Bayesian analysis placed fewer restrictions than
BEIR VII on the parameters for age-at-exposure and attained age.
Section 4.5 presents the main results of the quantitative uncertainty
analysis - uncertainty bounds summarizing the distributions for LAR, which
reflect both sampling and non-sampling sources of uncertainty. A comparison
with BEIR Vll's quantitative uncertainty analysis is given in Section 4.6. BEIR VII
used a non-Bayesian approach, which for most cancer sites produced results not
unlike ours. Although the BEIR VII uncertainty analysis has many desirable
features, it has several limitations which prompted us to consider an alternative
approach. Most notably, only uncertainties associated with sampling variation,
DDREF, and risk transport were quantified, and the non-Bayesian approach does
not ensure that results will be internally consistent; e.g., BEIR Vll's upper bound
for prostate cancer LAR is almost as large as the upper bound for all male
cancers combined.
Conclusions are given in Section 4.7. Foremost among them is that the
results of the analysis - the uncertainty distributions for the LAR summarized in
Section 4.5 - should not be over-interpreted. Results may be sensitive to
distributions which are subjectively assigned to sources of uncertainty (e.g., risk
transport), and not all sources of uncertainty can be quantified. Results of the
uncertainty analysis are meant primarily as guidance as to the extent to which
"true" site-specific risks for a hypothetical stationary U.S. population might differ
from the central estimates derived in Section 3.
4.2 Sources of Uncertainty Quantified in this Report
We quantified uncertainties associated with sampling variability, DDREF,
risk transport, errors in dosimetry, risk model misspecification, selection bias, and
errors in disease detection and diagnosis.
73
-------
Sampling variability. BEIR VII derived parameter estimates for most of
its ERR and EAR models from a statistical analysis of LSS solid cancer cases
and leukemia deaths, which were cross-classified by city, sex, dose, and inter-
vals based on age-at-exposure, attained age, and follow-up time. Here, sampling
variability refers to the uncertainty in parameter estimates associated with the
variation in the observed numbers of cancer cases or deaths in each of the
subcategories. For solid cancers, this includes uncertainties in parameters for the
linear dose-response (/?), and its modification by age-at-exposure (/) and
attained-age (77), but it does not include uncertainty relating to the shape
(curvature) of the dose-response.
DDREF. The dose and dose rate effectiveness factor was described in
Section 3.6. The uncertainty in the DDREF refers to problems associated with
extrapolating results on risks from studies of acute exposures and relatively large
doses to risks at low dose and dose rates. We adopted the BEIR VII nominal
value of 1.5, which was based on the curvature in the dose response observed in
data from the LSS and animal carcinogenesis studies. Uncertainty in the BEIR
VII DDREF estimate is due, in part, to the effect of sampling errors on estimates
of curvature.
Risk transport. This refers to the uncertainty in projecting risks to the
U.S. population using risk models derived from the Japanese A-bomb survivor
data. The uncertainty is due to lack of knowledge as to how radiogenic risks in
the Japanese cohort and the U.S. may differ.
The BEIR VII ERR models would be appropriate if radiogenic risks are
proportional to baseline rates. Likewise, the EAR model may be a reasonable
alternative if radiogenic risks are unrelated to baseline rates. For most sites, it is
plausible that projections based on some combination of the two models would
yield better approximations of risk. EPA's nominal risk projections are weighted
averages of ERR and EAR model-based projections. Here, risk transport
uncertainty is confined to the problem of assigning site-specific weights (among
competing plausible values) to the ERR risk model projections.
Incomplete follow-up. This uncertainty refers to the lack of any direct
information on risks for TSE outside the period of follow-up. For most solid
cancer sites, risk estimates were derived from data on cancers in the A-bomb
survivor cohort that occurred between 1958 and 1998. Thus, estimates of solid
cancer risks for TSE outside the interval (13y, 53y) must necessarily be based on
extrapolation. Models are fit to the data that best describe how ERR and EAR
depend on factors such as age-at-exposure and attained-age within the period of
follow-up. One then assumes that these age-related patterns hold for TSE
beyond the follow-up. Incomplete follow-up uncertainty is the uncertainty in risk
projections associated with these underlying assumptions.
74
-------
Errors in dosimetry. This refers to uncertainty in estimates of ERR and
EAR, and ultimately projections of risk, that result from errors in doses assigned
to the A-bomb survivors in the LSS cohort. The RERF report on DS02 (Kaul et a/.
2005) divides such dosimetry uncertainties into two broad categories: systematic
and random. "Systematic" refers to "the likelihood that doses to all individuals at
a given city will increase or decrease together [from imperfectly or unknown
effects]," whereas "random" refers to effects on individual survivor doses that act
more or less independently. Examples of systematic uncertainties are those
relating to the yields, neutron outputs, and burst heights, as well as the air
transport calculation method. Examples of random uncertainties are those
relating to survivor location and inputs needed to estimate shielding for individual
survivors. Both systematic and random uncertainties in dose estimates can lead
to bias in parameter estimates in the ERR and EAR models. Random errors will
tend to decrease the precision of estimates and can also have an effect on the
shape of the dose-response.
Errors in disease detection and diagnosis. Types of diagnostic
misclassification that can occur include classification of cancers as non-cancers
(detection error) and erroneous classification of non-cancer cases as cancer
(confirmation error). The former leads to an underestimate of the EAR but does
not affect the estimated ERR. Conversely, the latter leads to an underestimate of
the ERR but does not affect the EAR (EPA 1999a). Errors can also occur in the
misclassification of sites where cancers originate.
Selection bias in the LSS cohort. This refers to the possibility that risk
estimates derived from the LSS are biased downward because members of the
cohort, by being able to survive the bombings, demonstrated a relative insen-
sitivity to radiation.
4.3 "One-at-a-Time" Uncertainty Analysis
In this section, the LAR is calculated for ranges of "plausible" values (95%
Cl) for parameters in the ERR models, the DDREF, and the weight assigned to
the ERR model. (A more sophisticated uncertainty analysis, which accounts for
additional sources of uncertainty is presented in Section 4.4). Each parameter is
varied in sequence (one-at-a-time), while other parameters are set to nominal
values, and the corresponding range of LAR values is examined using graphical
methods. We start by examining how LAR for a solid cancer site (using stomach
cancer as an example) depends on sampling variability associated with para-
meters for linear dose-response (/?), age-at-exposure (/), and attained age (77)
in the ERR model:
ERR(D,e,d) =
(4-1)
75
-------
It is helpful to note that in Equation 4-1, /? is the ERR per Gy for age-at-exposure
30 and attained-age 70, and exp(y) is the increase in ERR per decade increase
in age-at-exposure (for e < 30). LAR projections for plausible values of each
parameter are then calculated using the methods described in Section 3, as is
shown next.
First, for any value of the linear dose-response parameter (/?), and
nominal values for the other parameters (r0 = -0.3 and 77 „ = -1.4), the estimate of
the LAR from an exposure (D) at age e is:
110
LAR(R} (D, e,j3) = J M(D, e, a, J3) • S(a) I S(e)da
e+L
110
= J /3Dexp(-0.3e*)(a/70)(-l4)-S(a)/S(e)da (4-2)
e+L
Using the nominal DDREF value of 1.5, the corresponding estimate of risk for a
lifelong chronic exposure is:
J S(e)-LAR(R\D,e,j3)-de
110-L
1.5 J S(e)de
LAR(R} (D, J3, stationary) = -* — (4-3)
The projection given in Equation 4-3 is based on the ERR model only. As
described in Section 3, EPA's (final) risk projections are weighted averages of
ERR and EAR risk projections. Analogously, we scale the ERR model-based
projection by a constant, which depends on EPA's nominal EAR and ERR model-
based projections and the weight assigned to the ERR model. For male stomach
cancer these are (per 10,000 person-Gy): 15 (ERR model) and 171 (EAR
model), so that for the nominal weight of 0.7, EPA's risk projection is 0.7(15) +
0.3(171) = 62. Thus, for stomach cancer, a reasonable scaling parameter would
be 4.1 (=62/15). (Note that this scaling factor is larger for stomach cancer than
for other cancers because, unlike most other cancers, baseline stomach cancer
rates are much greater in Japan than in the U.S.). Thus for any value of the ERR
weight parameter (w), the LAR is approximated as:
K(w, sex, site) x LAR(R} (D, f3, stationary) , with
,^ - ,. ...
K(w, male, stomach) = — ^^ — - - - - - (4-4)
The top-left panel in Figure 4-1 (a) shows how values for the LAR (based
on Equation 4-4) for both male and female stomach cancer depend on the linear
76
-------
dose-response parameter (/?). For males, plausible values of (3 are between 0.1
to 0.25, and the LAR ranges from about 15 to 60 (cancers per 10,000 person-
Gy). The corresponding range for females is somewhat narrower (about 25 to
60); there are more females than males in the LSS, and thus, for many sites, less
sampling variation.
The top-middle panel shows how values for LAR depend on parameters
for age-at-exposure (y}. First, note that for stomach cancer, y is likely between
about 0 and -0.5, i.e., ERR may be independent of age-at-exposure or,
alternatively, decrease by as much as 40% per decade (at y=-0.5) with age-at-
exposure < 30. (At y near -0.5, radiogenic risk would be almost 3 times as large
at age 0 as at age 30). In Figure 4-1 (a), it is seen that the LAR for male stomach
cancer risk can vary from about 80 (y about -0.5) to less than 50 (y ~ 0). For
females, the corresponding range is from about 90 to less than 60. These are
about half the width of the ranges for LAR associated with the linear dose-
response variable (/?), which suggests that there may be more uncertainty in
LAR associated with /? than there is with y.
The rightmost panel shows ranges of LAR values for the attained age
parameter. A comparison with results from the other panels (for /? and y}
indicates that uncertainty in LAR associated with attained age is relatively small.
The bottom panels in Figure 4-1 (a) provide results on how LAR for
childhood (ages < 15) exposures depend on the same three ERR parameters.
(The results were generated in exactly the same manner as above, except that in
Eq. 4-3 the integration is from age 0 to 15). Here, the uncertainty associated with
the age-at-exposure parameter (y} is much greater than seen before (for lifelong
exposures), and is comparable to the uncertainty associated with/?.
Figures 4-1(b)-(d) show the dependence of LAR on the ERR parameters
for three other sites: liver, lung, and bladder. The graphs for these sites share
many of the characteristics already noted. In general, for lifelong exposures,
uncertainty appears greatest for the linear dose-response parameter; for child-
hood exposures uncertainty in LAR associated with age-at-exposure is also
large. A comparison of figures for the four cancer sites indicates that the variation
in LAR is much larger for bladder cancer than for the other cancers. This is not
surprising since in the LSS dataset there were only 342 bladder cancers as
compared to 1146 liver, 1344 lung, and 3602 stomach cancers. For bladder
cancer, the LSS provides very little information on how ERR depends on age-at-
exposure or attained age, and the uncertainty in LAR associated with all three
ERR parameters is large.
77
-------
.
Linear Slope
Age-at-Exposure
Attained Age
100
I 80
Q_
X
LU
I 60
^0
'^
40
100
5 60
40
^
\
V
0.5
P
-0.5
0
y
100
g>
I 80
Q_
X
LU
I 60
-------
:
Linear Slope
Age-at-Exposure
70
60
1 5°
1 40
LU
I'0
i§ 20
10
(
j
1
- / /
/ /
- /
- /
3 0.5 1
70
60
1 5°
o
x 40
LU
!30
i§ 20
10
-
\
\
\
^K^^ .
-0.5 0 0
P y
Attained Age
&.
D
O
LU
O)
C
_0
:§
5
70
60
50
40
30
20
10
-J
120
0
3 100
o
x 80
LU
| 60
1 40
O
20
/
/ /
/ /
r" /
' /
. / &
1 /
/
120
tn
0
3 100
o
x 80
LU
| 60
1 40
O
20
\
\
\
\
\
\
\
\
\ v_
\ T
\ \
\\
\ \
\\
120
tn
0
3 100
(ft
o
x 80
LU
| 60
1 40
O
20
\
\
\
\k_
-^-
\
*• —
0 0.5 1 -1 -0.5 0 0.5 -5 0 5
P y ii
Figure 4-1 (b): Dependence of liver cancer LAR, for both lifelong and childhood exposures (age <
15) on ERR model parameters associated with the linear dose response (/?), age-at-exposure
(/), and attained-age (77).
79
-------
Linear Slope
Age-at-Exposure
Attained Age
400
| 300
Q.
X
LU
D)
i 200
100
c
800
n 600
o
Q.
X
LU
§ 40°
!E
0 200
C
/
/
/
./
) 1 ;
P
/
- /
- /
"/
400
| 300
Q_
X
LLJ
D)
i 200
-0
800
» 600
o
Q.
X
LU
o 400
S 200
\
Nir
\^
^~^~~~^
400
V)
§ 300
Q.
X
LU
D)
i 200
-------
Linear Slope
s
o
g-
LLJ
D)
C
0
o
x 150
LU
D)
I 10°
50
-1
\
\
^~~-C^^^—
250
g 200
O
x 150
LJJ
D)
I 10°
50
/ •
• -+^ '
•
012-5 0 5
400
e
| 300
Q.
X
LU
"§ 200
6 100
(.
• /
/ /
• T /
1 /
400
tn
&
| 300
Q.
X
LU
"§ 200
6 100
\
:
\\
400
£
| 300
Q.
X
LU
"§ 200
6 100
.
. ^-^.
-
) 1 2 3 -1 0 1 2 -5 0 5
Figure 4-1 (d): Dependence of bladder cancer LAR, for both lifelong and childhood exposures
(age < 15) on ERR model parameters associated with the linear dose response (/?), age-at-
exposure (y), and attained-age (77).
The graphs in Figures 4-2(a)-(c) compare "uncertainties" associated with
the linear dose response parameter to those associated with both risk transport
and the DDREF. Here, risk transport is evaluated by observing the dependence
of LAR on the ERR-model weight parameter (0
-------
Linear Slope
DDREF
Risk Transport
200 F
o 150 •
8 100
E
250
s 20°
ro
^ 150
^
5 100
50
/ /
r
D 0.5 1
P
/
r
//
/
D 0.5 1
P
200
150
100
50
(
250
200
150
100
50
(
\
\
\
^X '
312:
DDREF
\
\
\
\
\
\
\ X
\ \
"~~^-^_
312:
DDREF
200 '
150-
100-
50-
0'
c
180 r
160-
140-
120-
100-
80-
60 '•
0 0.5 1
ERR model weight
0 0.5 1
ERR model weight
Figure 4-2(a): Dependence of LAR (for lifelong exposures), for stomach and colon cancers on
the linear dose response parameter (f3), DDREF, and the ERR model weight parameter (w).
82
-------
Linear Slope
DDREF
Risk Transport
80
o 60
ra
O
o 40
ID
20
(
500
Cancers
D)
c
-1 200
100
(
- /
- / /
/ 4/
3 0.5 1
P
y
I /
2
f
I
80
60
40
20
(
500
400
300
200
100
\
\
\
\
\
\ X
\v_ ^-^^
312:
DDREF
\
\
\
\
\
\
\
, ^v
\
"^— -
3 1 2012:
80
60
40
20
0
3 (
500
400
300
200
100
3 (
•\ •
.^\N* .
^K^\
D 0.5 1
ERR model weight
y
/
.
D 0.5 1
p DDREF ERR model weight
Figure 4-2(b): Dependence of LAR (for lifelong exposures), for liver and lung cancers, on the
linear dose response parameter (/?), DDREF, and the ERR model weight parameter (w).
83
-------
Linear Slope
DDREF
Risk Transport
250
200
0
o
c
CJ 150
0
1 10°
50
7
/
/ / '
/ /
1 /
/ /
/ /
r
• /
7
250
200
150
100
50
\
\
\
\
\ \
\ \
X
V\ '
^~^~ .
01 2301 2 :
p DDREF
700
| 600
(0
0 500
0
±i
CO
ffl 400
D
1 300
200
/ /
/ /
/ /
/ /
/ /
- / /
/
700
600
500
400
300
->nn
: \ :
\
\
\\
\
\
\x
\\
Vv
x\
0 0.5 1012:
p DDREF
120
100
60
400
350
300
250
0
0 0.5 1
ERR model weight
0.5 1
ERR model weight
Figure 4-2(c): Dependence of LAR (for lifelong exposures), for bladder and residual site cancers,
on the linear dose response parameter (/?), DDREF, and the ERR model weight parameter (w).
84
-------
4.4 Monte Carlo Approach for Quantifying Uncertainties in LAR
For each cancer site: a multivariate probability distribution was assigned to
ERR model parameters, independent probability distributions were assigned to
parameters associated with non-sampling uncertainties, and Monte Carlo
methods were used to simulate the distribution of LAR for the U.S. population.
Bayesian methods used to generate the ERR model parameters are outlined in
the next section. Section 4.4.1 outlines the Monte Carlo method for simulating
the distribution for LAR. Section 4.4.2 describes how distributions were assigned
to non-sampling uncertainties, Section 4.4.3 describes the Bayesian method for
sampling variation, and Section 4.4.4 describes our Monte Carlo approach for
specific sites for which the Bayesian approach was used.
4.4.1 Monte Carlo method. The method is based on repeated random
sampling of all the parameters associated with uncertainty. In many important
aspects, the handling of sampling variation is identical to the approach described
in Section 4.3. For each iteration of the Monte Carlo process, a set of random
values for the ERR model parameters are generated, using Bayesian methods
described in the next Section, and a value for LAR is calculated based on
Equation 4-3. Then, to incorporate non-sampling uncertainties, the simulated
values of LAR are modified by randomly generated multiplicative "uncertainty
factors" (EPA 1999a), which are described next using the example of DDREF.
In Section 3, the LAR for low dose chronic exposures was calculated as
the ratio of the LAR for acute exposures divided by a nominal value of 1.5 for the
DDREF. For quantifying uncertainties, a subjective distribution is assigned to the
DDREF, which is lognormal with GM = 1.5 and GSD 1.35, corresponding to 2.5
and 97.5 percentile values of 0.8 and 2.7. Thus, if the only source of uncertainty
in our projections is uncertainty in the DDREF, the "true" LAR projection would -
with subjective probability of 95% - deviate from EPA's projection by a
multiplicative factor from (1.5/2.7) to (1.5/0.8). In general, an uncertainty factor
(UF) is the random factor by which a projection deviates from the "true" LAR due
to a specific source of uncertainty such as DDREF or risk transport. For
uncertainty associated with DDREF, the uncertainty factor is the ratio of 1.5
divided by the lognormal random variable with GM = 1.5 and GSD = 1.35. Then,
the Monte Carlo approach for simulating LAR is as follows:
1. Simulate N sets of ERR model parameters values using Bayesian
methods;
2. For each set of ERR model parameters, use Equation 4-4 to calculate an
initial value for LAR;
3. Assign a distribution for uncertainty factors associated with each non-
sampling source of uncertainty;
4. Generate N random values for each uncertainty factor;
85
-------
5. Multiply, element by element, the N initial values of the LAR generated in
Step 2 by the corresponding product of uncertainty factors generated in
Step 4.
4.4.2 Non-sampling sources of uncertainty. A summary of how each
source of uncertainty was treated is given in Table 4-1, with more detailed
discussions on each in the text below.
Table 4-1: Uncertainty factors for non-sampling sources of uncertainty
Source
Uncertainty Factor
Distribution1'2
Risk transport (quantified in BEIR VII)
DDREF (quantified in BEIR VII)
Incomplete follow-up3
See this Section
1.5/LN(1.0, 1.35)
LN(GSD= 1.2)
Errors in dosimetry
Random: linear dose response
Random: DDREF
Systematic
Nominal neutron RBE
LN(GSD= 1.16)
LN(GSD= 1.05)
LN(GSD= 1.1)
LN(GSD= 1.1)
LN(GSD= 1.05)
Errors in disease detection/diagnosis
Selection bias
Relative effectiveness of X-rays
Model misspecification for dose response
Total for sources not quantified in BEIR VII4
LN(GSD= 1.05)
LN(GSD= 1.1)
Not quantified
Not quantified
LN(GSD = 1.3): solid cancers
LN(GSD= 1.2); leukemia
1LN stands for lognormal. LN(a,/>) is the distribution with GM = a and GSD = b.
2Mean of distributions other than for DDREF is set to 1
3For solid cancers only
Includes incomplete follow-up, dosimetry, disease detection/diagnosis, selection bias
Risk transport. The uncertainty factor for risk transport is defined here as
the random factor by which the projection of LAR based on the ERR model,
derived from data on the Japanese A-bomb survivors, deviates from the "true"
LAR because radiogenic risk may not be proportional to baseline rates. For sites
other than thyroid, breast, bone, and lung, independent subjective probability
distributions were assigned to LAR(tme) as follows:
86
-------
P\LAR(ime) = LARW ] = 0.45 ; P[LAR(true} = LAR(A} ] = 0.05 ;
(LAR(tme) ~ Uniform between LAR(R} and LAR(A} ) with probability 0.5
This distribution assigns: with 50% probability, either the EAR or the ERR model-
based projection, or with 50% probability, a uniform distribution between the two
"extremes." For some sites, the LAR may not be bounded by the ERR and EAR
projections; however, in the absence of additional information, there is no way to
determine how far the probability distribution should be extended to account for
this.
For lung cancer, the only difference is that P[LAR(true} =LAR(R}] =0.05, and
P[LAR(tme} = LAR(A}] = 0.45. For bone, thyroid, and breast cancer, no risk trans-
port uncertainty was assumed. For the latter two cancer sites, note that the BEIR
VII projections were based on analyses of data from non-Japanese populations,
as well as from the LSS cohort.
The uncertainty factor for risk transport is the ratio: - ^- . It can also
LAR{R}
be defined with respect to a random ERR model weight parameter as
wLAR(R) +(\-w)LAR(A) . . ...... ... ...... __ ,,„.„,
- v 7 - , where w is U(0,1) with probability 0.5, equal to 1 with
/VlJV
probability 0.45, and equal to 0 with probability 0.05.
DDREF. A lognormal uncertainty factor with GM=1 and GSD=1.35 was
assigned to the DDREF for solid cancers (Figure 4-3). This is consistent with the
distributional assumptions made by BEIR VII, i.e., the variance associated with
the log-transformed DDREF is 0.09.
BEIR Vll's quantification of uncertainty in DDREF was primarily based on
a Bayesian analysis of the LSS data and animal carcinogenesis studies. The
main objective of their analysis was to estimate the curvature of the dose-
response, which, as described in Section 2.1.4, translates directly into an
estimate for DDREF. The analysis resulted in a posterior distribution for the
DDREF with GM=1.5 and GSD=1.28. The latter is equivalent to Var(\og(DDREF))
= 0.06. However, the BEIR VII Committee opined that: "the [Bayesian] DDREF
analysis is necessarily rough and the variance of the uncertainty distribution is
..., if anything, misleadingly small." Accordingly, they inflated the variance for the
log(DDREF) by 50% and set its variance equal to 0.09.
87
-------
Figure 4-3: Subjective probability density function for DDREF
Other non-sampling sources of uncertainty. Sources of uncertainty
considered here include uncertainties from: incomplete follow-up in the LSS,
dosimetry errors, and diagnostic misclassification. For cancers other than
leukemia, we assigned a single (encompassing) lognormal uncertainty
factor with GSD = 1.3. (For leukemia the GSD is 1.2, for reasons described in
the section on incomplete follow-up). The GMs for the component sources of
uncertainty considered in this subsection would range from about -0.9 to 1.1, with
about half of them greater than 1. The expected value (mean) of the overall
distribution was set to 1, since to use any other value would suggest a "precision"
in the stated uncertainties that is not warranted.
Incomplete follow-up. A-bomb survivors who were children at the time of
the bombings (ATB) still have substantial years of life remaining in which cancers
are to be expressed. Thus, uncertainty associated with incomplete follow-up is
greatest for childhood exposures, which accounts for about 40-45% of EPA's
projected cancer incidence radiation risk. For a crude indication of the relative
precision of the LAR for childhood exposures, we note that, for the BEIR VII
analysis of the LSS cohort, fewer than 2100 survivors with cancers were exposed
before age 15, compared to more than 3400 for age-at-exposure 15-30. Further-
more, approximately 90% of children < 10 ATB were still alive in the year 2000
(Little etal. 2008). More generally, about 45% of all survivors in the LSS were still
88
-------
alive in 2000, so that uncertainties in LAR projections from the incomplete follow-
up, especially for cancers that tend to develop relatively late in life, merit careful
consideration.
Combined with the potential for model misspecification of temporal and
age dependence, incomplete follow-up can lead to bias in projections of LAR. In
the UNSCEAR models described by Little et a/. (2008), ERR and EAR for solid
cancer mortality depend on TSE. The UNSCEAR risk projections differ from
those used here and in BEIR VII, in part, because different models are used for
extrapolating risks for cancers that might occur more than 53 y after exposure.
The uncertainty of projections based upon the parametric representations in
BEIR VII depend on the extent to which ERR and EAR for incidence and
mortality depend on TSE and other factors not accounted for in their risk models.
For EPA's previous assessment of radiogenic cancer risks, based
primarily on analysis of the LSS mortality data for follow-up until 1985, site-
specific uniform distributions were assigned to uncertainty factors to account for
sampling errors and possible model misspecification associated with temporal
dependence (1999a). For stomach, colon, lung, breast, thyroid and residual site
cancers, it was thought that these uncertainties might lead to an overestimate of
population risk. For these sites, a relative risk model was used that depended on
age-at-exposure but not attained age, and most of the projected risk was
associated with exposures before age 20. It was determined that "the contribu-
tion of childhood exposures was highly uncertain in view of statistical limitations
[i.e., sampling error] and possible decreases in relative risk with time after
exposure [i.e, modeling misspecification]." For most of these sites, the uniform
distribution, U(0.5, 1), was assigned to the uncertainty factor. In other words, the
ratio of the "true" population risks to the EPA projection was thought to range
between 0.5 and 1. For other solid cancer sites (except bone), the distribution for
the uncertainty factor was 0.8 to 1.5. Due to the extended follow-up period and
more flexible and appropriate modeling of age dependence in BEIR VII,
uncertainties associated with incomplete follow-up should be greatly reduced.
To update the uncertainty analysis to account for incomplete follow-up, the
new EPA risk models (see Section 3) were used to calculate the LAR for time-
since-exposure restricted to between 13 and 53 y, the period of follow-up for the
LSS incidence data. Slightly more than 50% of the projected LAR is associated
with this time period. Thus, unless the temporal dependence differs substantially
for time-since-exposure from what has been observed for the period of follow-up
in the LSS, it is unlikely to be a major source of uncertainty, with the possible
exception of childhood exposures. A common lognormal uncertainty factor
with GSD = 1.2 was used for solid cancers.
Leukemia deserves special mention. To paraphrase Little et al. (2008),
uncertainties in risk projections for leukemia would have more to do with risks for
times soon after exposure than for times extending beyond the current LSS
89
-------
follow-up. This is because the mortality follow-up in the LSS began in October,
1950, about 5 y after the bombings in Hiroshima and Nagasaki, and there is
evidence of risk for TSE < 5 y from other studies. In particular, a substantial
number of leukemia cases reportedly occurred in the LSS before 1951, with an
apparent subsequent decline; a significant increase in leukemia within 5 y of
radiotherapy was observed in the International Cervical Cancer study; and in an
analysis of the Mayak worker study (Shilnikova et al. 2003), the ERR/Gy for
leukemia mortality was significantly higher for external doses received 3-5 y prior
to death than for doses received more than 5 y earlier. We did not quantify
uncertainty associated with time-since-exposures < 5 y because, although it
might be larger than for most solid cancers, it is judged to be small compared to
sampling uncertainties for leukemia (see Section 4.4.4). For leukemia an UF was
not assigned for incomplete follow-up.
Errors in dosimetry. In 2003, RERF implemented a revised dosimetry
system called DS02, which is the culmination of efforts stemming from concerns
about the previous (DS86) system for assigning doses to the A-bomb survivors.
Chief among these concerns were discrepancies between DS86 calculations and
measured thermal neutron activation values (Roesch 1987). These measure-
ments indicated that DS86 might have seriously underestimated neutron doses
for Hiroshima survivors, and, as a result, y-ray risk estimates for solid cancers
could possibly be underestimated by more than 20% (Preston et al. 1993, EPA
1999a). However, the magnitude of this bias would depend on factors such as
the RBE for neutrons. Other factors motivating development of the new system
included improved computer models for radiation transport and biodosimetric and
cancer data indicating overestimation of doses for Nagasaki factory workers
(Preston et al. 2004).
A comprehensive review adequately resolved issues relating to the
discrepancies with neutron activation measurements (Preston et al. 2004). As
summarized in Preston et al. 2004 and detailed elsewhere (Cullings et al. 2006,
Young and Hasai 2005), major changes in DS02 include: 1) changes in the burst
height and yield for the Hiroshima bomb; 2) changes in the gamma radiation
released by the Nagasaki bomb; 3) use of new data on neutron scattering, etc.,
to improve calculations for radiation transport; 4) more detailed information and
better methods to account for in-home and terrain shielding; 5) more detailed
information for computing free-in-air fluences; and 6) new weighting factors for
fluence-to-kerma and fluence-to-dose calculations.
The RERF report on DS02 (Kaul et al. 2005) divides uncertainties
associated with the dosimetry system into two broad categories: systematic and
random. Systematic uncertainties include those relating to the yields, neutron
outputs and burst heights, and the air transport calculation method. Random
uncertainties include those relating to survivor location and inputs needed to
estimate shielding for individual survivors. In Kaul et al. (pp. 989, 991), a
90
-------
coefficient of variation (CV) of 12-13% (corresponding to a GSD of about 1.12)
was associated with systematic uncertainties.
For assessing the effects of random dose errors on risk projections, we
refer to the recent contribution by Pierce et al. (2008). As they note, "RERF has
for more than 15 years made adjustments to individual (DS86 and DS02) dose
estimates to reduce the effects of imprecision" on estimates of risk. Without
adjustment, it is well-established that random dose errors would cause a
downward bias in risk estimates if a linear dose-response is assumed. They may
also introduce a bias in the estimate of curvature, which is used for evaluating
the DDREF. RERF adjustments are currently based on the assumption that the
random errors are independent and lognormal with GSD = 1.42.
Pierce et al. argue for adjustments based on more sophisticated treatment
of the random errors that account for effects of "the use of smoothing formulae in
the DS02 treatment of location and shielding." Results in Pierce et al. (Table 1,
p. 123) indicate that the more realistic and sophisticated modeling of random
dose errors would result in a change of about 2% in the estimated linear dose-
response estimate of ERR and about a 15-20% change in the estimate of
curvature, compared to estimates based on current methods and assumptions.
The effect on the estimate of DDREF would be somewhat less than this, in part
because it depends also on data from animal carcinogenesis studies. Perhaps,
somewhat conservatively, we assign lognormal uncertainty factors with a
GSD equal to 1.05 (effects of random errors on the linear dose parameter
estimate) and 1.1 (effects on the estimate for the DDREF).
Finally, we quantify uncertainties relating to the use of a nominal neutron
RBE of 10. The use of this nominal weight assigned to the neutron component of
dose has already been discussed in Section 3.1. Calculations in Preston et al.
(2004) indicate that the use of an RBE of 20 would result in a relative decrease in
ERR estimates for solid cancers by about 5%. Radiobiological data (Sasaki et
al.) indicate an RBE of 20 or greater cannot be ruled out. A lognormal uncertainty
factor with GSD of 1.05 is assigned to this source.
Errors in disease detection and diagnosis. The BEIR VII Committee
concluded that "this is unlikely a serious source of bias in risk estimates." As
stated earlier, both detection and confirmation error can occur. Detection error
leads to an underestimate of the EAR, but does not affect the estimated ERR,
whereas confirmation error leads to an underestimate of the ERR but does not
affect the EAR (EPA 1999a).
Analyses of LSS mortality data formed the basis for EPA's previous risk
assessment. For that assessment, results from studies of Sposto et al. (1992)
and Pierce et al. (1996) were used to estimate the bias in risk estimates due to
diagnostic misclassification in the LSS mortality data. Conclusions from these
studies were that the ERR estimate for solid cancers in the LSS should be
91
-------
adjusted upward by about 12% and that the EAR estimate should be adjusted
upward by about 16%. Based on these results and results from the uncertainty
analysis by the NCRP 126 Committee (NCRP 1997), EPA assigned an
uncertainty factor of N(1.15, SD=0.06) for diagnostic misclassification. Here
N(a, SD=b) refers to a normal distribution with mean a and standard deviation b.
Misdiagnosis is likely to lead to a somewhat smaller bias in the BEIR VII
projections than in EPA's 1994 projections because the former were based on
the LSS incidence data. As noted in the BEIR VII report, "cancer incidence data
are probably much less subject to bias from under-ascertainment or from
misclassification, and this was an important reason for the committee's decision
to base models for site-specific cancers on incidence data. However, incidence
data are not available for survivors who migrated from Hiroshima to Nagasaki.
Adjustments are likely to account for this (Sposto et a/. 1992), but there is likely
to be some uncertainty in the adequacy of these adjustments." We have
assigned a lognormal uncertainty factor with GSD = 1.05 to diagnostic
misclassification. Admittedly, this understates the uncertainty for some cancers
since the uncertainty factor does not account for misclassification among
different cancer types.
Relative effectiveness of medical X-rays. For breast and thyroid
cancers, the BEIR VII risk models were based on pooled analyses of data from
the LSS and several medical studies. These medical studies were generally
based on data from patients who had received therapeutic or diagnostic X-rays,
which are of lower energy than the bulk of the photons irradiating the A-bomb
survivors. If the risk per unit dose for lower energy photons is >1 (see Section
5.2), there may be an upward bias in risk estimates from the pooled studies.
However, in many of the medical studies the doses were fractionated, so the
DDREF of 1.5 would not be applicable. Thus, any upward bias due to the higher
effectiveness of lower energy photons may be somewhat offset by the difference
in DDREF.
We did not incorporate any uncertainty associated with a higher effective-
ness in inducing cancer for medical X-rays compared with the photons from the
atomic bombs. It should be relatively small compared to the uncertainties
associated with sampling variability - especially for thyroid cancer.
Selection bias in the LSS cohort. The question as to whether there is a
serious selection bias has been a subject of considerable controversy. For
example, Little (2002) cited several papers by Stewart and Kneale from 1973 to
2000 which argued that the selection bias may be substantial. In a recent
analysis, Pierce et al. (2007) argue that the magnitude of the bias on the ERR
estimate for solid cancer is unlikely to be greater than 15-20%. (The bias might
be greater for non-cancer effects). We assign a lognormal distribution with
GSD 1.1 to the uncertainty factor for selection bias.
92
-------
Shape of the dose response. As described in Section 3.5, BEIR VII
models explicitly (leukemia) or implicitly (solid cancers) assume a linear-
quadratic (LQ) dose response for cancer induction by radiation. Although
epidemiological data are generally consistent with linearity at low doses (Section
2.2), recent mechanistic studies have revealed complex phenomena (Section
2.1) that could conceivably modulate risks at very low doses and dose rates,
either up or down, from what would be projected based on a LQ model. The
BEIR VII Committee did not attempt a quantification of this source of uncertainty.
Attempting to assign a probability distribution to the dose-response model would
necessarily be highly speculative and subjective; consequently, EPA has not
included this source of uncertainty in its quantitative uncertainty analysis.
However, it is acknowledged that a breakdown in the model at low doses, leading
to substantial errors in our risk projections, cannot be ruled out.
4.4.3 Bayesian analysis for sampling variability. This section describes
a Bayesian analysis of LSS incidence data, which we used to derive uncertainty
distributions for LAR for sampling variability. Distributions were derived for all
solid cancer sites except breast and thyroid. (Our treatment of sampling
variability for the latter two sites and leukemia is given in Section 4.4.4).
Uncertainties for bone and kidney cancers, which for this analysis were added
into the residual category, were not explicitly calculated.
The Bayesian analysis is in many respects similar to BEIR Vll's analysis of
LSS data (see also Preston et al. 2007). In BEIR VII, confidence intervals were
derived for parameter values in ERR risk models by fitting these models to the
LSS data. The fitting of the models was based on their likelihood (defined next),
and the LSS data includes observed rates for solid cancer incidence and
leukemia deaths for subgroups defined by city, sex, dose, and intervals based on
age-at-exposure, attained age, and follow-up time. The likelihood refers to the
probability of a set of observations given values for a set of parameters (Everitt
1995). In essence, the confidence intervals derived in BEIR VII contain values for
parameters (/?,/, and 77) for which the probabilities of observed cancer rates are
largest.
The fundamental difference between EPA's Bayesian analysis and the
analyses in BEIR VII is that the Bayesian analysis formally accounts for
subjective information about parameter values using prior distributions. Prior
distributions are probability distributions that summarize information about a
parameter that is known or assumed, prior to obtaining further information from
empirical data (Everitt 1995). For EPA's analysis of LSS data, the most important
prior distributions were the ones assigned to parameters in the ERR model. An
example is the prior distribution assigned to the age-at-exposure parameter for
most cancer sites. Under the assumption that for most sites, ERR decreases with
age-at-exposure, but that for most cancer sites the per decade decrease in ERR
(before age 30) would not be much greater than 3 (for which y < -1.1), a prior of
93
-------
N(-0.3, SD=0.5) was assigned to/. For this prior distribution, P(y < 0) « 0.7 and
P(y <-1.1)«0.05.
In any Bayesian analysis, distributions of the parameter values are
updated using the likelihood (which incorporates all the information from the LSS
about the parameters), yielding what is referred to as the posterior distribution.
The posterior distribution incorporates all that is known about parameter values,
and it can be used directly to calculate uncertainty intervals (often referred to as
credible intervals). A 90% uncertainty interval would be any interval for which the
quantity of interest, e.g., the linear dose response parameter or the LAR, is
included with posterior probability 0.90.
The relationship between the likelihood (the basis for the BEIR VII
analysis) and the posterior distribution (the basis for the EPA uncertainty
analysis) is that the posterior distribution is proportional to the product of the prior
distribution and the likelihood. If, for example, a constant prior is used, as is often
the case when there is very little prior information about a parameter value, the
likelihood and posterior distributions will be (outside of a multiplicative constant)
identical to each other, and the two types of approaches will yield similar, if not
identical, results. However, if there is information from another epidemiological
study about one or more parameters, then the prior can have a substantial
influence on the posterior distribution. For example, suppose it is "known" that
the linear dose-response parameter (/?) for a particular cancer site cannot be
greater than 0.5. The prior probability would be 0 for values above 0.5, and the
corresponding posterior probability would also be 0 - regardless of the likelihood.
Here, the confidence interval might contain values above 0.5, but the Bayesian
uncertainty interval would not.
From the previous example, it should be clear that Bayesian posterior
distributions and uncertainty intervals depend on the prior distributions assigned
to parameter values. In general, posterior distributions are more sensitive to the
choice of prior distributions when there is only limited data for updating them. At
the end of this Section, we describe the underlying prior distributions and their
rationale for our Bayesian analysis of the LSS data.
Although it is true that a different set of priors would have led to different
results than presented in Section 4.5, non-Bayesian analyses also depend on
assumptions made about parameter values. Unfortunately, it is often not obvious
what these assumptions are. For example, many may be unaware of the implicit
assumption in BEIR VII that the per decade decrease in ERR with age-at-
exposure may be the same or similar for most solid cancer sites. In contrast,
Bayesian analyses, through the use of prior distributions, provide a relatively
straightforward and flexible approach for incorporating what might be assumed
about parameter values.
94
-------
The main task in Bayesian analyses is to calculate the posterior distri-
bution - upon which inferences are based - from the data and prior distributions
for the parameters. It is usually very difficult, or impossible, to calculate it using
analytical means, so, instead, one typically simulates the posterior distribution
using complex sampling techniques such as Markov Chain Monte Carlo (MCMC)
- see for example, Gelman et a/. (2003) for a description of Bayesian methods
and computational methods such as MCMC. To simulate the posterior distri-
bution for ERR model parameters, we applied MCMC to the LSS incidence data
and the prior distributions for those parameters, as described next. This was
accomplished using the software program WinBUGS (Lunn et al. 2000). Further
details are given in Appendix B.
Prior distributions for ERR model parameters. An important feature of
our uncertainty analysis is that the age-at-exposure and attained-age parameter
values are allowed to depend on site. Separate sets of these two parameters
were used for almost all cancer sites; exceptions are cancers of the prostate,
ovary, and uterus. For these 3 sites, BEIR VII nominal values were used for age-
at-exposure (y = -0.3) and attained age (77=-1.4) because of insufficient data
on these cancers to provide stable estimates for these parameters or their
uncertainties. It should be noted that the uncertainty intervals for these sites are
not meant to account for uncertainties relating to age and temporal dependence
in risk.
Age-at-exposure parameter: Under the assumption that, for most cancer
sites, the ERR decreases with age-at-exposure, but the per decade decrease in
the ERR (before age 30) would not be > 3 (implying that y < -1.1), a prior
distribution of N(-0.3, 0.25) was assigned to y. This allows the ERR to be up to
« 20 times larger at birth than at age 30. For this prior distribution, P(/< 0) « 0.7,
and there is a 95% probability for the interval (-1.3, 0.7). As seems appropriate,
this interval for any site-specific parameter is considerably wider than BEIR Vll's
95% Cl for the age-at-exposure parameter for all solid cancers: (-0.51, -0.10).
Attained age parameter. The attained age parameter represents the
power to which the ERR increases (r/> 0) or decreases (77 < 0) with attained
age. For cancers other than prostate, uterine, or ovarian, independent N(-1.4, 2)
distributions were used. The distribution was chosen to be centered at the BEIR
VII nominal value for solid cancers and to have a lower limit of around -4. At this
lower limit, excess absolute risks for many cancers would not depend on attained
age because baseline rates typically increase by a power of about 4 with age.
The prior distribution assigns about a 95% probability to the interval (-4.2, 1.4).
Linear dose dose-response parameter: A lognormal prior distribution was
used for each of the linear dose-response parameters. Log-transformed
parameters for each cancer site were assumed to have prior distributions with a
common (unknown) mean and variance (r2). Lognormal priors were chosen, in
part, to ensure that ERR values cannot be negative. Details are given in
95
-------
Appendix B. In essence, the method represents a flexible approach of sharing
information on radiogenic risks among cancer sites. Such sharing of information
is desirable - especially for cancer sites for which ERR estimates are less
precise than for other sites. The variance (r2) determines how much information
is shared among sites. For example, if the variance is set to zero, the linear
dose-response parameter would be forced to be equal to the same value for
each site. In contrast, if the variance is (essentially) infinite, posterior distributions
for the site-specific dose-response parameters would be independent. In general,
the site-specific posterior distributions are "shrunk together" by an amount
dependent onr2. However, instead of specifying a value for r2 in advance, we
assigned it a prior distribution, so that the data also has a role in deciding how
similar values for the linear-dose-response parameter might be. The rationale for
this type of approach is further discussed in Pawel et al. (2008).
There are two main reasons for choosing more complicated prior distri-
butions for the linear dose-response parameters than for the age-related
parameters. First, for most cancers, the LAR is more sensitive to the linear dose-
response parameter than the other parameters, which warranted consideration of
a more sophisticated approach. Second, "extreme" values for the distributions of
the age-response parameters could be more easily determined and justified; e.g.,
it is reasonable to assume for the purposes of this analysis that attained-age
parameters would not be much less than -4 (for which EAR is constant with
attained age) and not much greater than 0.
4.4.4 Approach for other cancers. Cancer sites included here are
leukemia, breast, thyroid, and bone. EPA risk models for the latter three are not
based exclusively on analyses of the LSS data. We also discuss the approach for
uniform whole-body radiation.
Leukemia. The uncertainty from sampling variability was assumed to be
lognormal with GM equal to the nominal sex-specific estimates presented in
Section 3. The GSD was derived from the 95% Cl in Table 12-7 of BEIR VII for
the LAR associated with an exposure of 1 mGy per year throughout life. For
example, for males, the Cl is (19, 230) per 104 PY-Sv, which corresponds to a
GSD « 1.9. The BEIR VII confidence intervals account for uncertainties relating
to the linear and quadratic components in the dose response. Values for other
parameters were set to nominal values.
Breast and thyroid cancers. The EPA nominal estimates for these
cancer sites were based on risk models derived from a pooled analysis of data
from medical cohorts as well as the LSS. It would thus be inappropriate to
calculate sampling variability uncertainties from an analysis of only the LSS data
(as we did for almost all other cancer sites).
For breast cancer, the uncertainty from sampling variability was assumed
to be lognormal with GM equal to nominal EPA estimates presented in Section 3.
96
-------
The GSD was derived from the 95% Cl in Table 12-2 of BEIR VII for the EAR
linear dose-response parameter, i.e. (6.7, 13.3) per 104 PY-Sv (GSD = 1.2).
Results from Preston et a/. (2002) indicate that the data from the LSS was
extremely influential in the derivation of the BEIR VII risk model. The BEIR VII
risk model is quite similar to an EAR model that would have been derived from
LSS data alone. Among the cohorts used for the final BEIR VII model, the LSS
cohort had by far the greatest number of breast cancers and the largest number
of excess cases among those exposed to 0.02 Gy or more (see Preston et a/.,
Tables 6, 10, and 12). It is thus reasonable to assume that uncertainties relating
to DS02 dosimetry errors, selection bias, and other sources specific to the LSS
would have a similar impact on the BEIR VII breast cancer as for other cancer
sites.
For thyroid cancer, there is considerable uncertainty as to how risks may
depend on TSE. The pooled analysis of epidemiological studies by Lubin and
Ron (1998) indicates that radiogenic thyroid cancer risk decreases with time for
TSE greater than about 30 y. Recent results on data on children treated for an
enlarged thymus (Adams et a/., 2010) and tinea capitis in Israel (Sadetzki et a/.
2006) are consistent with these findings. However, it is unclear whether
radiogenic risk might reach a peak at TSE about 15-20 y, and at what TSE the
decline in risks might be most precipitous. To account for uncertainty in risks
associated with TSE, uncertainty intervals were derived using the two ERR
models recommended by NCRP (Models 3 and 4, NCRP 2008, pp. 291-292) and
the BEIR VII model (see Section 3-2, Eq. 3-7). For both of the NCRP models, the
ERR is a categorical function of age-at-exposure and TSE. In the BEIR VII
model, ERR does not depend on TSE. In one of these NCRP models (Model 3),
ERR declines precipitously at TSE around 30 y and then remains flat. In the
other (Model 4), the ERR is the same as in Model 3 for TSE up to 50 y, and then
halved for TSE > 50 y.
A 25% probability was subjectively assigned to each of the NCRP models
and a 50% probability to the BEIR VII model. For the NCRP models, a lognormal
distribution was assigned to the linear dose-response parameter with a GM equal
to the NCRP nominal value for this parameter (11.7) and a GSD = 1.6. The GSD
was derived using the 95% Cl (5.4, 24.9) given in the NCRP report (NCRP 2008,
Table 5.11), but adjusted upward to account for possible differences in the ERR
for males and females. For the BEIR VII model, lognormal distributions were
assigned to the linear dose-response parameters with GSD chosen to coincide
with 95% Cl of (0.14, 2.0) for males and (0.28, 3.9) for females (see NAS 2006,
Table 12-2). Although the thyroid risk models depended on data from medical
cohorts, it is unlikely that uncertainties associated with sources such as dosi-
metry error (for both the LSS and the medical cohorts) would be smaller than for
most other cancer sites. For thyroid cancer, an UF with GSD = 1.3 - the same
as for other solid cancers - was assigned to sources of uncertainty not quantified
in BEIR VII.
97
-------
Bone cancer. The nominal EPA risk model was derived from data on
radium dial painters exposed to 226Ra and 228Ra and patients injected with the
shorter-lived isotope 2 Ra. The risk of bone cancer is a relatively small
component of the risk for all cancers from uniform whole-body radiation.
Uncertainties for bone cancer are not quantified here, but EPA plans to address
this issue when it revises FGR-13.
Uniform whole-body radiation. To quantify uncertainties for the LAR for
all cancers from uniform whole-body radiation the simulated site-specific LAR
values were summed (over all cancer sites) at each iteration. For the Monte
Carlo simulation, uncertainties in transfer weightings for different cancer sites
were assumed to be independent. Transfer weightings for males and females for
the same cancer site were assumed to be fully correlated. To avoid under-
estimating the uncertainty for the LAR for all cancers combined, the DDREF was
assumed to be fully correlated for different cancer types: i.e., the DDREF was
assumed to be identical for all cancers other than bone and leukemia. Similarly,
the UFs associated with sources not quantified in BEIR VII were assumed to be
identical for each solid cancer type. For leukemia (as mentioned earlier) the UF
associated with incomplete follow-up was set to zero.
4.5 Results
The mean, median, and 90% uncertainty intervals for male and female
cancer incidence LARs are given in Tables 4-2a and b. Sex-averaged uncertainty
intervals are given in Table 4-2c. The lower bounds of 0 for prostate and uterine
cancers coincide with analyses of LSS incidence data, which provides insufficient
evidence to indicate a positive dose-response (Preston et al. 2007, MAS 2006).
In general, it is important not to over-interpret the lower bounds for other sites,
because they can be sensitive to prior distributional assumptions, e.g., whether a
lognormal or normal distribution is used. Except for thyroid and breast cancers,
the uncertainty bounds do not account for information about radiogenic risks
gained from studies other than the LSS. These include the Techa River study,
which, for example, showed a statistically significant effect of chronic radiation on
leukemia incidence (Krestinina et al. 2010).
The tables also include the EPA nominal projections described in Section
3. For almost all cancer sites, differences between the mean of the uncertainty
distribution and the EPA nominal projections are extremely small when compared
to the range of plausible values for LAR indicated by the uncertainty bounds.
When one also accounts for the different assumptions used for the uncertainty
analysis - compared to those used for deriving the nominal estimates - results
are remarkably consistent. For almost all individual cancers, and for all cancers
combined, the mean of the uncertainty distributions and the nominal estimates
are within 25% of each other. An exception is female bladder cancer, for which
the LSS data provides relatively little information on radiogenic risk. EPA's
projections are based the BEIR VII risk models, which were derived from LSS
98
-------
data specific to those sites, whereas the uncertainty distribution is based, in part,
on information on ERR "borrowed" from other sites. LSS data, although sparse,
indicate that relative risk for female bladder cancer may be somewhat larger than
for other sites. The mean of the uncertainty distribution for female bladder cancer
is greater than the EPA estimate because it "averages" observed risks for
prostate cancer with the larger observed risks for other cancer sites. In this way,
the uncertainty analysis takes into account that some of the difference in
estimates of site-specific ERRs may be due to sampling variation.
Table 4-2a: EPA projection and uncertainty distribution for the LAR for
male cancer incidence1'2
Uncertainty Distribution
Cancer Site
Stomach
Colon
Liver
Lung
Prostate
Bladder
Thyroid
Residual3
Leukemia
Total4
EPA
Projection
62
146
40
130
89
97
22
278
92
955
Mean
67
110
36
160
892
100
22
290
93
970
Lower (5%)
Limit (L)
8
39
6
58
O2
28
5
99
27
430
Median
32
99
24
140
O2
86
17
250
77
880
Upper (95%)
Limit (U)
220
230
110
320
410
230
54
610
210
1800
1 Cases per 10,000 person-Gy for exposures at low dose and/or dose rates.
2 Dose response for prostate cancer is not significant at 0.05 level. Posterior mean equated to
EPA projection. See Appendix B for further details.
3 Includes kidney and other cancers not here specified.
4 Excludes skin cancer
99
-------
Table 4-2b: EPA projection and uncertainty distribution for the LAR
for female cancer incidence
1
Uncertainty Distribution
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Uterus
Ovary
Bladder
Thyroid
Residual3
Leukemia
Total4
EPA
Projection
75
92
21
308
289
23
33
92
65
283
69
1350
Mean
70
100
28
260
310
23
37
57
91
340
69
1380
Lower (5%)
Limit (L)
9
37
4
95
140
O2
11
14
21
120
18
650
Median
36
91
16
220
280
O2
32
47
67
290
57
1270
Upper (95%)
Limit (U)
220
210
88
540
570
130
82
130
240
700
160
2520
1 Cases per 10,000 person-Gy for exposures at low dose and/or dose rates.
2 Dose response for uterine cancer is not significant at 0.05 level. Posterior mean equated to
EPA projection. See Appendix B for further details.
3 Includes kidney and other cancers not specified here.
4 Excludes skin cancer
100
-------
Table 4-2c: EPA projections and uncertainty distributions for the sex-
averaged LAR for cancer incidence1
Uncertainty Distribution
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Prostate
Uterus
Ovary
Bladder
Thyroid
Residual3
Leukemia
Total4
EPA
Projection
68
119
30
220
146
44
12
17
95
44
281
80
1160
Mean
69
110
32
210
160
44
12
19
79
57
310
81
1180
Lower (5%)
Limit (L)
9
42
6
83
70
O2
O2
5
24
15
120
29
560
Median
34
97
20
180
140
O2
O2
16
68
44
270
72
1090
Upper (95%)
Limit (U)
220
220
94
420
290
200
65
42
170
140
630
160
2130
1 Cases per 10,000 person-Gy for exposures at low dose and/or dose rates
2 Dose response for these cancers is not significant at 0.05 level. Posterior mean equated to
EPA projection. See Appendix B for further details.
3 Includes kidney and other cancers not specified here.
4 Excludes skin cancer
A comparison of EPA's nominal estimates to the 90% uncertainty bounds
indicates that, for some cancer sites, the nominal site-specific estimate may differ
from the LAR by a factor as large as 5 or more (stomach, prostate, liver, uterus).
For other sites (e.g., ovary or bladder) results suggest the EPA projection
underestimates the LAR by a factor of only about 2 and may overestimate risk by
a factor of about 4. Estimates may be accurate to within a factor of 3 or less for
lung, breast, colon, and residual site cancers, and to within a factor of about 2 for
all cancers combined. The sex-averaged 90% uncertainty interval for uniform
whole-body radiation, excluding skin cancer, is 5.6x10"2 to 2.1x10"1 Gy"1.
The contribution to uncertainties associated with sampling variability, risk
transport, DDREF, and other non-sampling sources of uncertainty are compared
in Tables 4-3a and b. Sampling variability is the dominant source of uncertainty
for bladder, thyroid, ovarian, leukemia, and residual site cancers. Risk transport
uncertainty is dominant for stomach, liver, and uterine cancers. For prostate
cancer, both sampling and risk transport uncertainties are large. DDREF is an
important contributor of uncertainty (but accountable for < 50% of the total
uncertainty) for many cancer sites. It is also an important source of uncertainty
for risk relating to uniform whole-body radiation.
101
-------
Table 4-3a: Percentage of uncertainty in LAR for male cancer incidence
attributable to sampling, risk transport, and DDREF1
Source of Uncertainty
Cancer Site
Stomach
Colon
Liver
Lung
Prostate
Bladder
Thyroid
Residual
Leukemia
Sampling
8
39
18
34
82
54
72
50
84
Risk Transport
76
5
60
3
11
3
0
3
12
DDREF
9
32
12
35
4
24
16
26
0
Other
7
24
10
27
3
18
12
20
4
1 Based on relative variance associated with each source of uncertainty
Table 4-3b: Percentage of uncertainty in LAR for female cancer incidence
attributable to sampling, risk transport, and DDREF1
Source of Uncertainty
Cancer Site
Stomach
Colon
Liver
Lung
Breast
Uterus
Ovary
Bladder
Thyroid
Residual
Leukemia
Sampling
7
36
18
19
16
83
55
56
78
53
77
Transport
77
7
64
27
0
11
1
6
0
5
18
DDREF
9
32
10
31
47
3
25
21
12
23
0
Other
7
25
8
24
37
2
19
17
10
18
5
1 Based on relative variance associated with each source of uncertainty
Results in Tables 4-2(a)-(c) were used to calculate uncertainty intervals for
radiation-induced cancer mortality. This was accomplished by applying crude
estimates of radiogenic cancer fatality rates, equal to the ratio of the nominal
EPA projection for mortality divided by the corresponding projection for incidence
to the lower and upper bounds for cancer incidence. For uniform whole-body
radiation, 90% Uls for cancer mortality (Gy~1) are 2.4x10~2 to 1.0x10~1 for males,
3.4x10"2 to 1.3x10~1 for females, and 2.9x10~2 to 1.1x10~1 when sex-averaged.
These intervals do not account for uncertainties associated with the fractions of
cancers that are fatal.
102
-------
Tables 4-4a,b provide uncertainty intervals for the LAR for incidence
associated with childhood exposures for selected sites. Results for most cancers
are reasonably consistent with the estimates of radiogenic risk from childhood
exposures, which were given in Section 3. However, for female bladder and lung
cancers, the means of the posterior distributions are noticeably different than the
central estimates derived using the BEIR VII models. As described above, the
difference can be partially attributed to the sharing of information among cancer
sites for the uncertainty analysis.
Another reason relates to BEIR VII assumptions relating to trends in the
dose-response with age-at-exposure. In BEIR VII, the age-at-exposure para-
meters for these two sites were set to the same value as for most other cancer
sites. This is because, when models without this restriction were fitted to the LSS
data, differences among the age-at-exposure parameters for cancer sites other
than the thyroid were not statistically significant. However, this only means that
the LSS provides insufficient information to show that trends with age-at-
exposure are different for these two sites. In fact, for lung cancer, data from the
LSS suggests that radiogenic risks might not be as dependent on age-at-
exposure as for other cancer sites. In contrast, the Bayesian uncertainty analysis
allowed for different values for site-specific age-at-exposure parameters.
For all cancers combined, the 90% Ul for LAR (Gy~1) associated with
childhood exposures is 7.7x10~2 to 3.6x10~1 for males and 1.2x10~1 to 5.5x10~1 for
females. These uncertainties for childhood exposures may be somewhat
understated because it is difficult to fully account for uncertainties relating to
incomplete follow-up in the LSS. Also, for some cancer sites - not listed but
included in the total, such as prostate and ovarian cancers - the analysis does
not account for age and time-related uncertainties.
Table 4-4a: EPA projection and uncertainty distributions for male cancer
incidence for childhood exposures for selected sites1'2
Uncertainty Distribution
Cancer Site
Stomach
Colon
Liver
Lung
Bladder
Residual3
All4
EPA
Projection
128
272
79
247
175
676
1950
Mean
110
200
68
200
120
790
1860
Lower (5%)
Limit (L)
11
63
11
50
21
240
770
Median
52
170
44
160
88
630
1640
Upper (95%)
Limit (U)
370
440
200
480
330
1780
3620
1 Cases per 10,000 person-Gy for exposures at low dose and/or dose rates.
2 Risks for exposures before the 15th birthday.
3 Includes kidney and other cancers not specified here.
4 Excludes skin cancer
103
-------
Table 4-4b: EPA projection and uncertainty distributions for female cancer
incidence for childhood exposures for selected sites1'2
Uncertainty Distribution
Cancer Site
Stomach
Colon
Liver
Lung
Bladder
Residual3
All4
EPA
Projection
161
179
44
611
176
736
3290
Mean
120
200
57
350
71
1010
2870
Lower (5%)
Limit (L)
13
59
7
86
12
300
1230
Median
59
170
31
280
51
780
2550
Upper (95%)
Limit (U)
400
450
190
880
200
2340
5490
Cases per 10,000 person-Gy for exposures at low dose and/or dose rates.
2 Risks for exposures before the 15th birthday.
3 Includes kidney and other cancers not specified here.
4 Excludes skin cancer
4.6 Comparison with BEIR VII
4.6.1 Quantitative uncertainty analysis in BEIR VII. The BEIR VII
Report includes a quantitative uncertainty analysis with 95% subjective CIs for
each site-specific risk estimate of LAR for low-LET radiation. The analysis
focused on three sources of uncertainty thought to be most important: sampling
variability in the LSS data, the uncertainty in transporting risk from the LSS to the
U.S. population, and the uncertainty associated with values for the DDREF for
projecting risk at low doses and dose rates from the LSS data. The BEIR VII
Committee did not assign specific distributions (e.g., normal or lognormal) to
sources of uncertainty. Instead, the quantification was based on variances for
log-transformed random variables (uncertainty factors) for each source of
uncertainty. Their treatment of specific sources of uncertainty is outlined next.
Sampling variability. For most cancer sites, BEIR VII derived parameter
estimates for ERR and EAR models based on a statistical analysis of LSS cancer
cases and deaths, cross-classified by city, sex, dose, and intervals based on
age-at-exposure, attained age, and follow-up time. For all solid cancer sites
except breast and thyroid, the BEIR VII uncertainty analysis accounted for only
the sampling variability associated with the linear dose parameter (ft). The
uncertainty analysis made use of an approximation for the variance of the
log(LAR) associated with sampling variability:
Var,
SAMPLING
[\og(LAR(D, ej)]
(4-6)
Risk transport. To quantify uncertainties from risk transport, BEIR VII
essentially assumed that either the EAR or ERR model is "correct" for risk
104
-------
transport, and that a weight parameter (w) equals the probability the ERR model
is correct. BEIR VII approximated Var[log(LARJ] as follows:
VarTRANSPORT [\og(LAR)] « \og[LARm (0(R) ) / LAR(A) (0(A) )]2 W(\ - w) . (4-7)
Here, 6(R) denotes the vector of estimated and nominal parameter values for ft,
y, r\, and DDREF for the ERR model, and LAR(R\9(R)} represents the
corresponding nominal LAR estimate. Likewise, 9(A} and LAR(A\6(A)) represent
the estimated parameter values and nominal LAR values for the EAR model.
EPA's use of a subjective probability distribution for risk transport repre-
sents a significant departure from BEIR Vll's approach. The BEIR VII method
tends to yield larger estimates of uncertainty for risk transport, particularly for
cancer sites such as the prostate, for which U.S. and the A-bomb survivor cohort
have very different baseline rates.
DDREF. As detailed in Section 4.4.2, BEIR VII assumed that the variance
of the log-transformed DDREF equals 0.09. EPA assigned a normal distribution
to log(DDREF), also with variance = 0.09.
Combining sources of uncertainties. To calculate the var(log(LAR)),
the BEIR VII Committee simply summed the variances for \og(LAR) associated
with sampling error, risk transport, and DDREF. To calculate 95% subjective
confidence intervals, they further assumed that the combined uncertainty for LAR
follows a lognormal distribution.
Unquantified sources of uncertainty. BEIR VII noted several other
sources of uncertainty but did not quantify them, arguing instead that uncer-
tainties for many of these other sources are relatively small. These other sources
of uncertainty include: 1) uncertainty in the age and temporal pattern of risk,
especially for individual sites, which was usually taken to be the same as that
derived for all solid tumors; 2) errors in dosimetry; 3) errors in disease detection
and diagnosis; and 4) unmeasured factors in epidemiological experiments.
4.6.2 Comparison of results. Results from EPA's quantitative uncer-
tainty analysis are compared with BEIR VII uncertainty intervals for LAR cancer
incidence (Table 4-5). For purposes of comparison, 95% uncertainty intervals
were used. For most sites, results are reasonably consistent. Exceptions include
prostate cancer (BEIR VII upper bound appears to be too large), ovarian cancer
(BEIR VII upper bound much larger than EPA's), and female thyroid cancer (for
which we considered different risk models than in BEIR VII).
105
-------
Table 4-5: EPA and BEIR VII 95% uncertainty intervals for LAR of solid
cancer Incidence1
Males
Females
Cancer Site
Stomach
Colon
Liver
Lung
Prostate
Breast
Uterus
Ovary
Bladder
Remainder
Thyroid
Solid cancers
EPA
(6, 270)
(32, 280)
(5, 130)
(49, 360)
(0, 520)
None
—
—
(22, 270)
(82, 740)
(4, 69)
(320, 1970)
BEIR VII
(3, 350)
(66, 360)
(4, 180)
(50, 380)
(<0, 1860)
None
—
—
(29, 330)
(120,680)
(5, 90)
(490, 1920)
EPA
(8, 270)
(30, 250)
(3, 110)
(80, 650)
—
(120,650)
(0, 180)
(8, 99)
(11, 160)
(100,860)
(17,310)
(520, 2800)
BEIR VII
(5, 390)
(34, 270)
(1, 130)
(120,780)
—
(160,610)
(<0, 131)
(9, 170)
(30, 290)
(120,680)
(25, 440)
(740, 2690)
1 Cases per 10,000 person-Gy for exposures at low dose and/or dose rates
4.7 Conclusions
The main results given in Section 4.6 suggest that the EPA risk
projections for uniform whole-body radiation (total for all cancer sites) are likely to
be within a factor of 2 or 3 of the "true" risk for the U.S. population. For many
individual cancer sites, the projections and actual risks might differ by a factor of
roughly 3 to 5, and even more for cancers of the stomach, prostate, liver, and
uterus. For childhood exposures, the uncertainties are somewhat larger. An
important caveat is that the analysis did not fully account for important
uncertainties associated with the shape of the dose response at low doses and
dose rates.
The quantitative uncertainty analysis did allow for sources of uncertainty,
such as dosimetry errors and some cancer misdiagnosis, which were not
quantified in BEIR VII. For sources of uncertainty quantified in BEIR VII, results
from this analysis and BEIR VII are consistent for most sites.
Results from the EPA uncertainty analysis should not be over-interpreted.
The results presented in Section 4.6 are meant solely as rough guidance on the
(relative) extent to which "true" site-specific risks for a hypothetical stationary
U.S. population might differ from the central estimates derived in Section 3.
Distributions for uncertainty factors rely on subjective judgment, and it is not
always possible to satisfactorily evaluate "biases" associated with sources of
uncertainty such as risk transport. Modeling uncertainties, e.g., the uncertainty
associated with BEIR VII model assumptions on how ERR depends on attained
106
-------
age, age-at-exposure, and TSE, are often especially difficult to quantify. Uncer-
tainties in mortality risk projections associated with changes in cancer fatality
rates were not evaluated.
107
-------
5. Risks from Higher LET Radiation
5.1 Alpha Particles
Assessing the risks from ingested or inhaled a-emitting radionuclides is
problematic from two standpoints. First, it is often difficult to accurately estimate
the dose to target cells, given the short range of a-particles in aqueous media
(typically < 100 urn) and what is often a highly non-uniform distribution of a
deposited radionuclide within an organ or tissue. Second, for most cancer sites,
there are very limited human data on risk from a-particles. For most tissues, the
risk from a given dose of alpha radiation must be calculated based on the
estimated risk from an equal absorbed dose of y-rays multiplied by an "RBE"
factor that accounts for different carcinogenic potencies of the two types of
radiation, derived from what are thought to be relevant comparisons in
experimental systems.
The high density of ionizations associated with tracks of a-particles
produces DMA damage which is less likely to be faithfully repaired than damage
produced by low-LET tracks. Consequently, for a given absorbed dose, the
probability of inducing a mutation is higher for alphas, but so is the probability of
cell killing. The effectiveness of a-particle radiation relative to some reference
low-LET radiation (e.g., 250 kVp X- rays or 60Co y-rays) in producing a particular
biological end-point is referred to as the a-particle relative biological effectiveness
(RBE). The RBE may depend not only on the observed end-point (induction of
chromosome aberrations, cancer, etc.), but on the species and type of tissue or
cell being irradiated, as well as on the dose and dose rate.
In most experimental systems, the RBE increases with decreasing dose
and dose rate, apparently approaching a limiting value. This mainly reflects
reduced effectiveness of low-LET radiation as dose and dose rate are decreased
— presumably because of more effective repair. In contrast, the effectiveness of
high-LET radiation in producing residual DMA damage, transformations, cancer,
etc. may actually decrease at high doses and dose rates, at least in part due to
the competing effects of cell killing. For both low- and high-LET radiations, it is
posited that, at low enough doses, the probability of a stochastic effect is
proportional to dose and independent of dose rate. Under these conditions, the
RBE is maximal and equal to a constant RBEM. In order to estimate site-specific
cancer risks for low dose alpha radiation, we need a low-dose, low-LET risk
estimate for that site and an estimate of the RBEM.
5.1.1 Laboratory Studies. The experimental data on the RBE for a-
particles and other types of high-LET radiation have been reviewed by the NCRP
(NCRP 1990) and the ICRP (ICRP 2003). From laboratory studies, the NCRP
concluded that: "The effectiveness of a-emitters is high, relative to (3-emitters,
being in the range of 15 to 50 times as effective for the induction of bone
108
-------
sarcomas, liver chromosome aberrations, and lung cancers." The NCRP made
no specific recommendations on a radiation weighting factor for alpha radiation.
The ICRP has reiterated its general recommendation of a radiation
weighting factor of 20 for a-particles (ICRP 2003, 2005). However, ICRP
Publication 92 further states (ICRP 2003):
Internal emitters must be treated as a separate case because their RBE depends
not merely on radiation quality, but also, and particularly for a-rays with their
short ranges, on their distribution within the tissues or organs. It is, accordingly,
unlikely that a single WR should adequately represent the RBEM for different a
emitters and for different organs. The current WR of 20 for a-rays can thus serve
as a guideline, while for specific situations, such as the exposure to radon and its
progeny, or the incorporation of 224Ra, 226Ra, thorium, and uranium, more
meaningful weighting factors need to be derived.
Another set of recommendations for a-particle RBE is contained in the
NIOSH-lnteractive RadioEpidemiological Program (NIOSH-IREP) Technical
Documentation intended for use in adjudicating claims for compensation of
radiogenic cancers (NIOSH 2002, Kocher etal. 2005). For a-particle caused solid
cancers (other than radon-progeny-induced lung cancer), IREP posits a
lognormal uncertainty distribution for its radiation effectiveness factor (REF,
equivalent to RBEM) with a median of 18 and a 95% Cl [3.4, 101]. For leukemia,
IREP employs a hybrid distribution: REF = 1.0 (25%); LN with Ul[1,15] (50%); LN
with UI [2,60] (25%).
Studies comparing groups of animals inhaling insoluble particles with
attached a- or (3-emitters have been performed to assess RBE for lung cancer. In
a large long-term study of beagle dogs, Hahn et al. (1999) reported that the RBE
was at least 20. An RBE of about 20 was also found in F344 rats for inhaled a-
emitting 239Pu02 particles, relative to (3-particles from inhaled 144Ce02 or
fractionated X-irradiation (Hahn et al. 2010). An analogous study of lung cancer
induction in CBA/Ca mice found that, in the limit of low doses, 242Cm a-particles
were 9.4 times (90% Cl: 5,23) as effective in producing adenocarcinomas as
45Ca (3-particles; however, the apparent RBEM was only 1.5 (90% Cl: 0.7,9) for
adenomas (Priest et al. 2006).
5.1.2 Human Data. Results from epidemiological studies of groups with
intakes of a-emitting radionuclides can be used directly to develop site-specific
cancer risk coefficients for alpha radiation; they can also be used in conjunction
with low-LET studies to estimate RBE; finally, these results can be used in
combination with estimates of RBE to derive low-LET risk estimates where none
can be obtained from low-LET studies.
There are 4 cancer sites for which there are direct epidemiological data on
the risks from alpha irradiation: bone, bone marrow, liver, and lung. Not coinci-
dentally, these are sites for which we are particularly interested in obtaining high-
LET risk estimates because they are ones which tend to receive higher than
109
-------
average doses of alpha radiation from certain classes of internally deposited
radionuclides. For each of these sites except bone, we also have risk estimates
for low-LET radiation derived from the LSS.
Bone cancer. Although new data are being obtained from research on
Mayak plutonium workers (Koshurnikova et a/. 2000, Sokolnikov et a/. 2008), the
most extensive sources of information on radiogenic bone cancer in humans
continue to be from: (1) radium dial painters ingesting 226Ra and 228Ra and (2)
patients injected with the shorter-lived isotope Ra.
Given their long radioactive half-lives, the radionuclides ingested by the
dial painters had time to redistribute throughout the mineral bone before
decaying. It is estimated that the average a-dose to target endosteal cells is
about 50% of the average skeletal dose (Marshall et a/. 1978). The shorter-lived
224Ra, however, is largely confined to the bone surface so the endosteal dose is
higher than the average skeletal dose. Speiss and Mays (1970) estimated that
the endosteal dose was higher by a factor of 9, but a subsequent determination
of the surface-to-volume ratio in bone reduced the estimated factor to 7.5 (Lloyd
and Hodges 1971, MAS 1980).
EPA has taken its estimates of risk of a-particle induced bone sarcoma
from the BEIR IV analysis of the 224Ra data, which is consistent with a linear, no-
threshold dose response (MAS 1988, EPA 1994). The corresponding low-LET
risk estimate (per Gy) was assumed to be a factor of 20 lower than that based on
the assumed a-particle RBEMof 20.
Subsequent to BEIR IV, improvements have been made in the dosimetry
for the 224Ra patients, especially those treated as children. Some additional
epidemiological data have also become available. The updated data set has
been analyzed by Nekolla et a/. (2000) and found to be well-described by an
absolute risk model, which for small acute doses reduces to the form:
where Ar is the increment in bone cancer incidence from an endosteal dose, D,
of a-particle radiation; g(e) reflects the variation in risk with age at exposure (e)\
and h(f) represents the variation with time after exposure (f). A good statistical fit
was found for g(e) as an exponentially decreasing function of age at exposure,
and for h(f) as a lognormal function of time after exposure.
Normalizing the time integral of h(f) to unity, a maximum likelihood
calculation yielded:
a = 1.782 x KT'Gy1,
110
-------
g(e) = exp[-0.0532 (e - 30)],
i\2'
1
where ?0is 12.72 y and a is 0.612. Thus, the temporal response, h(t), has a GM =
12.72yandaGSD = e'7= 1.844.
For estimates of bone cancer risk from alpha radiation, we adopt the
model and calculational methods of Nekolla et a/., with two modifications. First,
those authors assumed, for simplicity, a fixed life-span of 75 y; our lifetime
estimates are derived using their derived mathematical models, but, as with our
other risk estimates, applied in conjunction with gender-specific survival functions
determined from U.S. vital statistics. Second, Nekolla et a/, adopted the ratio of 9
for endosteal to skeletal dose published by Speiss and Mays; we employ the
updated estimate of 7.5. The effect of this change is to increase the coefficient a
in the model above by a factor of 1.2 (= 9/7.5). With these modifications, the
calculated average lifetime risk of bone cancer incidence is 2.5x10~3 Gy"1 for
males and 2.3x10~3 Gy"1 for females. The population average of 2.4x10~3 Gy"1 is
close to the FGR-13 estimate of 2.72x10"3 Gy"1 (EPA 1999b). About 35% of all
bone cancers are fatal (SEER Fast Stats), and it is assumed here that the same
lethality holds for radiogenic cases - half that previously assigned (EPA 1994).
Thus, the mortality risk projections for a-particle induced bone cancer are:
8.6x10"4 Gy"1 (males), 8.2x10"4 Gy"1 (females), and 8.4x10"4 Gy"1 (sex-averaged).
There has been a great deal of discussion in the scientific literature
concerning a possible threshold for induction of bone sarcoma (MAS 1988).
Often cited is a plot of bone cancer risk versus dose in radium dial painter data,
which appears to show a rather abrupt threshold at about 10 Gy. However, it has
been pointed out that such an apparent threshold may be an artifact of pre-
senting the data on a semi-log plot (incidence vs. log dose); Mays and Lloyd
found that a conventional plot of incidence vs. dose is consistent with linearity
(Mays and Lloyd 1972, MAS 1988). In laboratory studies, Raabe et a/. (1983)
found that the mean time to tumor increases with decreasing dose rate,
suggestive of a "practical threshold" in dose rate below which the latency period
would exceed the lifespan of the animal. However, interpretation of this finding
remains controversial (MAS 1988), and Rowland has noted that - contrary to the
practical threshold hypothesis - bone sarcomas sometimes appeared in dial
painters at short times after low intakes of radium (Rowland 1994). It has also
been postulated that a sub-linear dose response relationship, resulting in a
practical threshold below which the risk is negligible, might be produced by a
requirement for two radiation-induced initiation steps (Marshall and Groer 1977,
MAS 1988) or by the need for radiation-induced stimulation of cell division
(Brenner et a/. 2003).
111
-------
A more recent statistical analysis of bone sarcomas in the dial painters
concluded that the data could be fit with a linear model having a threshold at
about 9 Gy but was inconsistent with a linear no-threshold model, even when a
cell killing term was included (Games et al. 1997, Hoel and Games 2005). In
contrast, the incidence of bone cancer among the 224Ra patients was consistent
with a linear no-threshold dose-response (BEIR IV, Nekolla et al. 2000), and
there was evidence of an excess of bone cancers among a group of ankylosing
spondylitis patients who received an estimated average endosteal dose of about
6 Gy - somewhat below the "threshold" dose estimated from the dial painter
data.
One possible explanation for the discrepancy is that the differences in
temporal and spatial pattern of the doses for two sets of nuclides give rise to a
threshold in the case of 226Ra and 228Ra, but not 224Ra. However, no plausible
mechanism has been put forth for a linear, threshold model, and it is very hard to
reconcile it with the standard paradigm for radiation-induced cancer in which
cancer risk is enhanced by radiation-induced mutations in target cells, here
presumably those contained in the endosteal cell layer.
Hoel and Games fit the radium dial painter data to a number of different
dose response functions. Of these, the linear-threshold model provided the best
fit. However, that model is only one example of a simple 2-component spline
function (one in which there is zero slope up to an estimated threshold at 9 Gy
and a positive slope at higher doses). Alternatively, the data could be fit to other
2-component spline functions having a positive (non-zero) slope up to some
break point, above which the slope is increased. A range of such models would
also be consistent with the dial painter data and are arguably as biologically
plausible as the linear-threshold model. In particular, as shown below, the lack of
observed bone cancers in the dial painters between 0 and 10 Gy is not
inconsistent with the slope inferred from the patients injected with Ra. In
addition, an important limitation to the analysis of Hoel and Games is that it does
not factor in the uncertainties in dosimetry, which could distort the shape of the
dose-response in a way that produces an apparent threshold.
For many of the dial painters, there were possible complicating effects of
tissue damage (fibrosis) associated with very high doses (10-200 Gy) of alpha
radiation in the bone (Lloyd and Henning 1983). The usefulness of the dial
painter data for low dose risk estimation also suffers from several other
problems: the intake of radium was estimated many years after the event and
may be inaccurate; the distribution of radium in the bone is nonuniform and "hot
spots" capable of extensive cell killing may have occurred; the continuous receipt
of dose makes it difficult to separate out the fraction of dose associated with
cancer induction; the contributions from a-emitters and other radiations accomp-
anying radium decay cannot be separated; and the fraction of the total dose to
the endosteal cells cannot be specified precisely (Boice 2006).
112
-------
Although no bone sarcomas were observed in dial painters who received
an estimated dose of less than 10 Gy, this is not inconsistent with the linear
projection based on the 224Ra patients. Overall, 449 dial painters were classified
as receiving an average skeletal dose > 0 but < 10 Gy. The estimated collective
skeletal dose for this group was 738 person-Gy (Hoel and Games 2005). As
noted above, the endosteal dose has been estimated to be « 50% of the average
skeletal dose, so the collective endosteal dose among these dial painters is
calculated to be about 370 person-Gy. The risk coefficient derived from the 224Ra
patients is « 2.4x10"3 Gy"1 (see above); hence, < 1 radiogenic bone cancer would
be projected among the dial painters whose doses were below the posited
threshold. On this basis, the dial painter data does not have the power to reject
the low-dose risk estimate derived from the 224Ra patients.
On the other hand, only 4 bone cancers were observed among the low-
dose 224Ra-treated spondylitis patients, whereas 7.8 radiogenic cases would be
projected from the linear model, and 1.3 spontaneous cases would be expected.
Moreover, none of the 4 cases were osteosarcomas, even though the majority of
cases at higher doses were of that type. According to Nekolla et al. (2000), these
findings suggest that the model projection based on the 224Ra patient data may
be conservative. These authors further note that a zero initial slope could not be
rejected based on a linear-quadratic fit to the 224Ra patient data.
Bijwaard et al. (2004) have carried out a biologically based modeling study
of radiation-induced bone cancer incidence in beagles and in radium dial painters
in which it was assumed that: (1) mutation of a stem cell produces an altered
"intermediate" cell; (2) clonal expansion creates a pool of the intermediate cells;
(3) mutation in an intermediate cell produces a malignant cell; and (4) the single
malignant cell repeatedly divides to form a tumor. In earlier work, where the 2-
mutation model was applied to data on radon-induced lung cancer in rats, it was
found that the process was dominated by a linear increase in the first mutation
rate with dose, leading to a linear increase in cancer risk with dose (Bijwaard et
al. 2001). However, in the case of bone cancer induction in beagles, it appeared
that the radiation had affected both mutational steps, leading to a linear-quadratic
dose-response at lower doses, where cell-killing effects could be neglected. The
data on radium dial painters showed a similar dependence (Leenhouts and
Brugmans 2000, Bijwaard et al. 2004).
Sokolnikov et al. (2008) found an excess of bone cancer in plutonium
exposed workers at the Mayak nuclear plant in Russia. The evidence for a bone
cancer dose-response rests on only 3 deaths, all occurring in individuals with an
estimated bone surface dose exceeding 10 Gy. Nevertheless, the data were not
inconsistent with a linear dose-response relationship.
Studies of patients receiving radiotherapy for childhood cancers indicate
that low-LET radiation exposure also increases the risk of bone cancer (Tucker et
al. 1987, Hawkins et al. 1996, Vu et al. 1998). Also noteworthy, especially in view
113
-------
of the presumed lower biological effectiveness of low-LET radiation, is that Vu et
al. (1998) found an excess risk among patients receiving a localized bone dose
of 1-10 Gy, with a mean dose of approximately 3 Gy, substantially lower than the
threshold absorbed dose for a-particles suggested by the spline fit to radium dial
painter data discussed above (Hoel and Carnes 2005). Hawkins et al. (1996), on
the other hand, reported no observed risk below 10 Gy; however, in their study
the median dose among irradiated patients receiving < 10 Gy was only about 0.1
Gy. The data are sparse, and the authors did not derive quantitative risk
coefficients. Nevertheless a rough estimate of the ERR/Sv based on the data in
Hawkins et al. is in reasonable agreement with that derived from the 224Ra data,
but with wide uncertainty bounds (unpublished calculations).
An RBE for bone cancer induction can be derived from a comparative
analysis of data on beagles injected with the a-emitter 226Ra or the (3-emitter 90Sr,
both of which are distributed fairly uniformly throughout the volume of calcified
bone (Mays and Finkel 1980, Bijwaard et al. 2004) . Employing a two-mutation
model for bone cancer induction, Bijwaard et al. found that the dose-response
relationships for both these radionuclides were approximately linear-quadratic at
low doses, and that the linear coefficient was approximately 9.4 times higher for
radium than for strontium. Based on this finding, EPA is adopting a revised RBE
value of 10 for bone cancer; i.e., the risk per Gy for low-LET radiation is assumed
to be 1/10 that estimated for a-particle radiation.
Uncertainty. Based on a consideration of sampling error alone, Nekolla et
al. derived a standard error of only ±33% on the slope of the linear dose-
response relationship derived from the 224Ra patient data, but a zero initial slope
could not be excluded. A linear-quadratic fit to that data yielded about a 20%
reduction in the best estimate of the linear coefficient. As discussed above, the
226Ra data in both animals and humans are suggestive of a sublinear dose-
response relationship for bone cancer, but the case for a threshold is
unconvincing.
Recognizing that the estimate may be conservative, EPA has adopted the
model for bone cancer risk due to alpha radiation derived by Nekolla et al. from
the 224Ra patient data. The uncertainty distribution is taken to be triangular with
the vertex at the nominal estimate and the lower and upper bounds at zero and
twice the nominal estimate, respectively. For low-LET radiation, the nominal
estimate of risk per Gy is 10 times lower but the upper bound is taken to be 4
times the low-LET nominal estimate, reflecting additional uncertainty associated
with the difference in biological effectiveness between low-LET radiation and a-
particles.
Leukemia. Excess leukemia cases have not been observed in studies of
radium dial painters or patients injected with high levels of 224Ra, although in
some cases there was evidence of blood disorders that may have been
undiagnosed leukemias (NAS 1988). It appears from these studies that bone
114
-------
sarcoma is a more common result of internally deposited radium, and that the
radium leukemia risk is much lower than that calculated using ICRP dosimetry
models together with a leukemia risk coefficient derived from the LSS weighted
by an RBE of 20 (Mays et al. 1985, MAS 1988, Harrison and Muirhead 2003,
Cerrie 2004). More recently, however, an excess of myeloid leukemia has been
found in ankylosing spondylitis patients receiving lower doses of 224Ra (Wick et
al. 1999, 2008). Supported also by data on 224Ra injected mice (Humphreys et al.
1993), it was hypothesized that at high doses the bone cancer risk is predom-
inant, but at low doses the bone cancer risk is diminished and replaced by a
leukemia risk (Wick etal. 2008).
In part, the anomalously low risk of leukemia from a-particles might be
attributed to microdosimetry: i.e., target cells may be non-uniformly distributed in
the bone marrow in such a way that the dose to these cells is considerably lower
than the average marrow dose. Evidence suggests, however, that microdos-
imetric considerations do not fully account for the lower risk, and that high-LET
radiation is only weakly leukemogenic. Thorotrast patients, who are expected to
have a more even distribution of a-particle energy, do show an excess of
leukemia, but only about twice the risk per Gy as seen in the LSS (ICRP 2003).
Moreover, an RBE of only about 2.5 has been found for neutron-induced
leukemia in mice (Ullrich and Preston 1987), a situation in which the high-LET
radiation dose would have been nearly uniform throughout the marrow.
The BEIR VII low-LET risk estimate for leukemia incidence is roughly 50%
higher than that of UNSCEAR (2000b) or EPA (1994). Using a Bayesian
approach, Grogan et al. (2001) estimated the a-particle leukemia risk to be
2.3x10"2 per Gy. If one adopts the BEIR VII low-LET leukemia (incidence) risk
estimate, this would correspond to an RBE of approximately 2.9. Through a
comparison of Thorotrast and A-bomb survivor data, Harrison and Muirhead
(2003) also estimated the RBE to be 2-3. However, the authors noted that the
Thorotrast doses were likely to be underestimated by a factor of 2-3 (Ishikawa et
al. 1999), and that the RBE was perhaps very close to 1.
Ankylosing spondylitis patients (mostly young adult males) injected with
relatively low amounts of 224Ra had a higher rate of leukemia than that projected
from the general population or that observed in a group of unirradiated control
patients (Wick et al. 1999, 2008). After 26 y of average follow up, the exposed
group of 1471 patients had 19 leukemias compared to 6.8 expected based on
age- and gender-specific population rates; after 25 y of average follow up, the
1324 control patients had 12 leukemias (7.5 expected). The average dose to
bone surface was estimated at 5 Gy in these patients. According to ICRP
dosimetry models, the average marrow dose is about 10% of the bone surface
dose for internally deposited 224Ra (ICRP 1993). Thus, the estimated average
marrow dose is « 0.5 Gy, and the excess risk, calculated using the population
projected rate is « 1.7x10"2Gy"1. This is about twice the leukemia risk projection
for 30-y old males derived in BEIR VII from the LSS data (MAS 2006, p. 281).
Thus, these radium-injection data are also roughly consistent with an RBE of
115
-------
about 2. Alternatively, if the unirradiated control patients are used as the com-
parison group, the estimated risk per Gy and RBE are roughly halved. Hence,
these data also support an RBE for leukemia induction of about 1-2. It should be
noted, however, that the temporal variation of excess leukemias appeared
different in this study from that observed in the LSS (Wick et al. 1999).
EPA has been employing an RBE of 1 for a-particle induced leukemia
(EPA 1994). Based on the information discussed above, the RBE is being
adjusted upward to a value of 2, with a confidence interval of 1 -3.
Liver cancer. The LSS shows a statistically significant excess of liver
cancer. The uncertainty bounds derived by BEIR VII are wide, both because of
the large sampling error and the uncertainty in the population transport (liver
cancer rates are about an order of magnitude lower here than in the LSS cohort).
The BEIR VII central estimate for gamma radiation is « 2x10"3 Gy"1; the EPA
central estimate based on the weighted AM rather than the weighted GM of the
two methods for transferring risk from the LSS to the U.S. population is « 3 x10~3
Gy"1. For comparison, updated analyses of data on Thorotrast patients from
Denmark (Andersson et al. 1994) and Germany (van Kaick et al. 1999) yielded
estimates of 7x10"2 and 8x10"2 excess liver cancers per Gy, respectively.
Assuming an RBE of 20 for the a-particle RBE, these values are about 20%
higher than what would be projected from the EPA liver cancer model - quite
reasonable agreement given the large uncertainties and difference in age and
temporal distribution. However, Leenhouts et al. (2002) has reanalyzed the
Danish Thorotrast data, employing a biologically based, two-mutation model of
carcinogenesis, and derived a lifetime liver cancer risk estimate of 2x10"2 Sv"1 (4
x10"1 Gy"1), an order of magnitude higher than the BEIR VII central estimate, but
consistent with the BEIR VII upper bound. One reason given by Leenhouts et al.
for the higher risk estimate is that the model projects risk over a whole lifetime,
whereas the original analysis by Andersson et al. addressed only the risk over
the period of epidemiological follow-up. The increase may also partly stem from a
correction for downward curvature in the dose-response function at high doses.
An excess of liver cancer has been found among workers at the Mayak
nuclear facility in the Russian Federation, especially among workers with
Plutonium body burdens and among female workers (Gilbert et al. 2000).
Averaged over attained age, the ERR per Gy for plutonium exposures was 2.6
for males and 29 for females. (Sokolnikov et al. 2008). For comparison, the BEIR
VII risk model for y-ray induced liver cancer derived from the LSS yields an ERR
per Gy of 0.32 for males and females, calculated for exposure age 30 and
attained age 60. Thus, the RBEs that would be derived from the LSS and Mayak
worker study would be roughly 8 for males and 90 for females.
In conclusion, the Danish and German Thorotrast results are in good
agreement with one another, and the risk projections derived from them are in
quite reasonable agreement with what would be projected from the LSS,
116
-------
assuming a plausible RBE of about 20. There is considerable uncertainty in the
estimates, relating to uncertainty in the dose estimates, the fraction of the dose
"wasted" because it was delivered after the cancer was initiated, and the
extrapolation from high doses (several Gy) to low environmental doses. In
addition, as seen from the Leenhouts et al. modeling exercise, there is consid-
erable uncertainty in projecting risk over a whole lifetime, especially the
contribution from childhood exposures. The results from the Mayak worker study
appear to be in only fair agreement with those from the Thorotrast studies. Based
on its review of the available information, EPA adopts a model for calculating a-
particle induced liver cancer, which is a scaled version of the BEIR VII model,
equivalent to multiplying the corresponding BEIR VII low-LET risk estimates, on
an age- and gender-specific basis, by an RBE of 20. The population average risk
isthen6x10"^Gy"1.
Lung cancer. Excess lung cancers have been associated with the
inhalation of a-emitting radionuclides in numerous epidemiological studies.
Cohort studies of underground miners have shown a strong association between
lung cancer and exposure to airborne radon progeny. This association has also
now been found in residential case-control studies. In addition, a cohort study of
workers at the Mayak nuclear plant has also shown an association with inhaled
Plutonium (Gilbert et al. 2004). The miner studies serve as the primary basis for
BEIR VI and EPA estimates of risk from radon exposure (MAS 1999, EPA 2003),
and results from the residential studies are in reasonable agreement with those
risk estimates (Darby et al. 2005, Krewski et al. 2005). The Agency has no plans
at this time to reassess its estimates of risk from exposure to radon progeny, but
it is the intent here to reassess estimates of risk from inhaled plutonium and other
a-emitters, along with the lung cancer risk associated with low-LET exposures.
Table 5-1 compares summary measures of risk per unit dose for the U.S.
population derived from the LSS in BEIR VII and from the pooled underground
miner studies in BEIR VI. For radon, the estimation of lung dose requires a
conversion from radon progeny exposure, measured in working level months
(WLM). Estimating this conversion factor involves a model calculation of the
deposition of radon progeny in the airways, the distribution of a-particle energy
on a microdosimetric scale, and the relative weights attached to different tissues
in the lung (MAS 1999, EPA 2003, James et al. 2004). Results are presented for
the dose conversion factor of« 12 mGy per WLM derived by James et al. (2004)
and for the estimate of 6 mGy per WLM recommended in UNSCEAR 2000a.
When compared to results from animal studies, the inferred a-particle
RBEs in Table 5-1 may appear to be unreasonably low - especially for females.
It should be recognized, however, that the risk model used to derive risk
estimates for radon are in certain ways incompatible with the models for low-LET
lung cancer risk in BEIR VII. They differ not only with respect to their functional
dependence on age, gender, and temporal factors, but also with respect to the
interaction with smoking. In contrast to the BEIR VII models, the radon risk
117
-------
models do not incorporate a higher risk coefficient for females or for children. The
miner cohorts from which the radon models were derived consisted essentially
entirely of adult males, and it is possible that radon risks are underestimated for
children and females. The radon risk appears to be almost multiplicative with
smoking risk (or the baseline lung cancer rate), whereas the LSS data suggests
an additive interaction. It is unclear whether these apparent differences with
respect to gender and smoking reflect a real mechanistic difference in
carcinogenesis by the two types of radiation exposure (chronic alpha vs. acute
gamma) or unexplained errors inherent in the various studies.
Table 5-1: Lung cancer mortality and RBE
Data
Source
A-bomb
mortality
EPA radon
risk model
Gender
Male
Female
Combined
Male
Female
Combined
Risk per 106
Person-WLM
—
—
640
440
540
Risk per 104
Person-Gy
140
270
210
8001 16002
5501 11002
6751 13502
RBE
1.0
1.0
1.0
5.71 11.42
2.01 4.12
3.21 6.42
1 Risk per Gy to the whole lung or RBE calculated assuming: (1) 12 mGy/WLM, on average, to
sensitive cells in the bronchial epithelium (James et al. 2004) and (2) lung risk partitioned 1/3
(bronchi): 1/3 (bronchioles): 1/3 (alveoli).
2 Calculated assuming 6 mGy/WLM, on average, to sensitive cells in the bronchial epithelium
(UNSCEAR 2000a).
Lung cancer results from the LSS cohort can also be compared with those
on Mayak workers, whose lungs were irradiated by a-particles emitted by inhaled
Plutonium (Gilbert et al. 2004), but the results of such an analysis must be
viewed critically. The dose from inhaled Pu is highly uncertain, as is the relative
sensitivity of different target cells to radiation. Information on smoking in both
cohorts is limited. The populations are quite different with respect to gender and
age profile. Males account for about 75% of the PY and over 90% of the lung
cancers among the internally exposed Mayak workers, but for only about 30%
and 55% of the PY and lung cancers, respectively, among the LSS cohort.
Another issue is that the dependence of the risk on attained age appears to be
quite different in the two studies - a monotonically increasing EAR for the LSS,
but a sharp decrease in the EAR above age 75 for the Mayak workers. There
are, however, very few data on these older Mayak workers. Focusing just on lung
118
-------
cancers appearing between ages 55 and 75, one finds that the central estimates
of risk per Sv in the two studies are comparable, consistent with an RBE for a-
particles of 10 or more.
A more recent analysis of the Mayak plutonium worker data, based on
improved dosimetry, has been published (Sokolnikov et al. 2008). From a statis-
tical modeling of the lung cancer data, it was estimated that the ERRs per Gy at
age 60 were 7.1 for males and 15 for females. For comparison, the LSS study
yielded an ERR per Gy of 0.32 and 1.4, respectively, for males and females, for
exposure age 30 and attained age 60. Thus, the two sets of data together would
suggest an RBE of roughly 20 for males and 10 for females.
The risk per unit dose estimate from the plutonium exposed Mayak
workers appears to be considerably higher than that from the radon studies
despite the fact that the lung dose from radon progeny is projected to be almost
entirely to the epithelial lining of the airways, whereas the inhaled plutonium dose
is expected to be concentrated in the alveoli, which is generally thought to be a
much less sensitive region for cancer induction.
There seems to be no fully satisfactory way to reconcile all the results
from the LSS, miner, and Mayak worker studies with what one would expect from
the dosimetry and experimental determinations of a-particle RBE, even taking
into account the sampling errors in the various epidemiological studies. The
Mayak study is ongoing, with possible improvements in the dosimetry still to be
made; the LSS risk estimates are also somewhat suspect, especially their
dependence on gender and age at exposure (see Section 3.2). In particular, it is
odd that the risk among the A-bomb survivors is higher in females than males,
despite the much lower lung cancer incidence among Japanese women than
men. Also, the BEIR VII lung cancer model reflects the negative trend with age at
exposure obtained from the analysis of all solid tumors, but there is very little
evidence to directly support a higher lung cancer risk for childhood exposures.
5.1.3 Nominal risk estimates for alpha radiation. Information on a-particle
RBEM (relative to y-rays) for induction of cancer is sketchy, especially in humans.
Laboratory studies are mostly indicative of a value of about 20, but with likely
variability depending on cancer type and animal species or strain. There is also
evidence in both animals and humans that the RBEM is much lower for induction
of leukemia. Comparisons of data on lung cancer induction by inhaled radon
progeny or plutonium with data on the A-bomb survivors yields somewhat
conflicting results, suggesting possible errors in the data or in the underlying
assumptions regarding the form of the models, internal dosimetry, or the
sensitivity of different parts of the lung. At this point, comparisons between the
radon data and the LSS data suggest an RBE « 20 for lung cancer induction,
but the Mayak results so far fail to substantiate this. Further follow-up of the LSS
cohort and additional information on the Mayak workers may help to resolve this
issue.
119
-------
EPA's site-specific a-particle risk estimates will be obtained by applying an
RBE of 20 to our y-ray risk estimates, with two exceptions: 1) an RBE = 2 for
leukemia and 2) continued use of models derived from BEIR VI to estimate lung
cancer risk from inhaled radon progeny (MAS 1999, EPA 2003). The low-dose, y-
ray risk estimate for bone cancer is obtained by dividing the risk per Gy for a-
particles - estimated from patients injected with Ra - by an RBE of 10.
Aside from those revisions pertaining to leukemia, liver cancer, and bone
cancer described above, this approach is consistent with previous EPA practice
except in the case of breast cancer, where previously an RBE of 10 was
employed rather than 20 (EPA 1994). The justification for the lower RBE was that
the estimated (y-ray) DDREF was 1 for breast cancer but 2 for other solid
tumors. The evidence for such a difference in DDREF appears weaker now, and,
for simplicity, we are now applying the same nominal DDREF (1.5) and RBE (20)
for most solid tumors, including breast.
5.1.4 Uncertainties in risk estimates for alpha radiation. For most
cancer sites, the uncertainty in a-particle risk can be calculated from the
combined uncertainties in y-ray risk, as presented in Section 4, and in a-particle
RBE. For solid cancers, EPA previously assigned a lognormal uncertainty
distribution to the a-particle RBE, with a 90% Cl from 5 to 40. The median value
is thus ~ 14.1 and the GSD « 1.88 (EPA 1999a). This distribution still appears
reasonable for solid tumors other than bone cancers. The uncertainty distribution
for leukemia induced by a-emitters deposited in the bone was previously taken to
be uniform over the interval [0,1] (EPA 1999a). Based on the more current
information discussed above, a lognormal distribution is now assumed, with GM
= 2 and GSD = 1.4.
In the case of a-particle induced liver cancer, EPA is basing its 95% upper
confidence limit on the risk estimate derived from the modeling approach of
Leenhouts et al. (4x10"1 Gy"1). This upper bound value is consistent with a log-
normal distribution with a GM equal to EPA's nominal central estimate of 8x10"2
Gy"1 and a GSD of 2.66. The lower 95% confidence limit on the distribution is
then 1.6x10"2 per Gy, which corresponds to what would be inferred from the LSS
liver cancer risk estimate in conjunction with an assumed a-particle RBE of 8.
Risk projections for bone cancer are only important when considering
internally deposited "bone-seekers." Given the difficulties in estimating the dose
to target cells in bone, EPA is deferring the quantification of uncertainty in bone
cancer risks until the Agency reevaluates the risks from specific internal emitters.
5.2 Lower Energy Beta Particles and Photons
As energetic electrons lose energy in passing through matter, they
become more densely ionizing: i.e., the average distance between ionization
120
-------
events shrinks, and more energy is deposited in ionization clusters. As discussed
earlier, such clusters produce DSBs and complex DMA damage that are more
difficult for the cell to repair. Indeed it has been suggested that a large fraction of
the residual damage caused by low-LET radiation may stem from such clusters
produced at the ends of electron tracks (Nikjoo and Goodhead 1991). For this
reason, it might be expected that lower energy (3-particles would be more
biologically damaging than higher energy betas. Furthermore, since the energy
distribution of secondary Compton electrons is shifted downward as incident
photon energy is reduced, the biological effectiveness of photons might also be
expected to rise with decreasing energy, implying that lower energy photons,
including medical X-rays, which typically have energies below 150 keV, might be
more damaging than the y-rays to which the LSS cohort was exposed.
Results from many studies tend to confirm these predictions for low-LET
radiations, including measurements of chromosome aberrations, mutations, cell
transformation and cancer induction. The most extensive source of data on the
subject consists of comparative studies of X- and y-ray induction of dicentrics in
human lymphocytes. In these studies, 220-250 kVp X-rays generally produced 2-
3 times as many dicentrics as 60Co y-rays (NCRP 1990, MAS 2006). The
relevance of such findings for cancer induction is unclear; in particular, a dicentric
will render a cell incapable of cell division. Other laboratory studies directed at
ascertaining the RBE for various types of radiation, relative to X- or y-rays,
provide additional indirect information, suggesting again that the orthovoltage X-
rays often used in radiobiology may be a factor of 2-3 times more hazardous than
Y-rays with energies above about 250 keV (Kocher et al. 2005, NCRP 1990, MAS
2006). Kocher et al. further conclude that X-rays with energies < 30 keV, such as
those used in mammography, may have a slightly higher RBE than those in the
range 30-250 keV.
Kocher et al. also consider what RBEs should be applied to (3-particles.
Noting that the average energy of a Compton electron produced by an incident
250 keV photon is 60 keV, they conclude that (3-particles above « 60 keV should
have about the same RBE as > 250 keV photons - i.e., = 1.0. One important
radionuclide that emits a substantial fraction of its decay energy in the form of a
lower energy beta is 3H, for which the mean (3-energy is 5.7 keV and the
maximum is 18.6 keV. For 3H and other betas with average energy below 15
keV, the authors recommend a lognormal uncertainty distribution with a GM = 2.4
and a GSD = 1.4, corresponding to a 95% Cl of (1.2, 5.0). The reference
radiation is again chronic y-rays. In addition, they assign the same probability
distribution to the RBE for internal conversion or Auger electrons with energy <
15 keV as for 3H. This uncertainty range is comparable to what was
recommended for < 30 keV photons and is generally consistent with experiments
in which investigators compared 3H with y-rays in the induction of various end-
points.
121
-------
Kocher et a/, also state that electrons of energy 15-60 keV would be
expected to have about the same RBE as 30-250 keV photons but that direct
biological data are lacking.
A review of tritium risks has recently been conducted by an independent
advisory group for the Health Protection Agency of the UK (HPA 2007). The
authors found that, in a wide variety of cellular and genetic studies, the RBE
values for tritiated water (HTO) were generally in the range of 1-2 when
compared with low dose-rate orthovoltage X-rays and 2-3 when compared with
chronic y-rays. The HPA Report also surveyed several laboratory studies
comparing animal carcinogenesis by HTO and by chronic X-rays or y-rays.
Derived RBEs from those studies were generally consistent with those obtained
in vitro, but it was pointed out that the carcinogenesis studies all suffered from
methodological problems. Overall, the HPA Report concluded that "an RBE of
two compared with high energy gamma radiation would be a sensible value to
assume." Although much of the experimental evidence suggested a value
between two and three, fractional values were "not considered appropriate."
The conclusions of the HPA report were supported by experimental and
theoretical evidence (Nikjoo and Goodhead 1991, Goodhead 2006) that the
biological effects of low-dose, low-LET radiation predominantly reflect complex
DMA damage generated by ionization and excitation events produced by low
energy electrons near the ends of their tracks with energies > 100 eV but no
more than about 5 keV. Figure 5-1 shows a plot, for various incident radiations,
of F, the cumulative fraction of the total dose deposited in an aqueous medium
by electrons of energy T(> 100 eV). These fractions were estimated by Nikjoo &
Goodhead (1991) using track-structure simulation codes, and results were found
to be similar to those of a numerical approximation method developed by Burch
(1957). Assuming that the amount of critical damage is proportional to F(5 keV),
the estimated RBE is = 2.3 for 3H (3-particles and = 1.4 for 220 kVp X-rays, both
relative to 60Co y-rays or 1 MeV electrons. Alternatively, if the critical damage is
taken to be proportional to F(1 keV), the estimated RBEs would be = 1.6 for 3H
and = 1.2 for the X-rays.
Through a more accurate Monte Carlo procedure, Nikjoo and Goodhead
calculated, for each of several initial electron energies, the cumulative fraction of
the total dose deposited by electrons with energies between 100 eV and a
specified energy. Those results are shown in Figure 5-2. From the figure, it is
estimated that the contribution of low-energy (0.1-5 keV) electrons to the total
dose from an electron with initial energy 10 keV would be = 63%, compared to
=51% for an incident 100 keV electron. The authors did not calculate the
distribution for higher energy incident electrons, but assuming that the fractional
increase in F obtained in applying the Monte Carlo method in place of the Burch
approximation is about the same as for 100 keV electrons (=10%), the result
would be =37% for the higher energy electrons or 60Co y-rays. Using this
approach, it should be possible to estimate average RBEs for a whole range of
122
-------
low-energy (3-emitters. Furthermore, from spectral information on the secondary
electrons produced by a photon source of a given energy, RBEs could also be
estimated for y-ray emitters.
1.0
0.8
Cumulative
fraction Q.6
of
total dose
F
0.4
0.2
2MeV e"
1MeV
10'
10'
10'
Electron Energy. T(eV)
Figure 5-1: Cumulative fraction of the total dose, F, plotted against secondary electron
kinetic energies, T, for a variety of low-LET radiations calculated by Nikjoo & Goodhead
(1991) using the method of Burch (1957)
A comprehensive review on the subject of low energy electron and photon
RBEs has recently been published (Nikjoo and Lindborg 2010). The authors
tabulated results from experimental data on cell inactivation, chromosome
aberrations, cell transformation, micronucleus formation, and DSBs and found a
wide range of values, dependent on electron and photon energies, but apparently
also on irradiation conditions, cell type, and experimental conditions. They also
summarized results from biophysical modeling of DSB formation. Again there
was a considerable spread in the estimated RBEs, presumably due to dif-
ferences in the underlying assumptions and details of the calculations.
No firm conclusions can be drawn from human epidemiological data on
the RBE for lower energy photons and electrons. Risk coefficients derived from
studies of cohorts medically irradiated with X-rays are in some cases lower than
what has been observed for the A-bomb survivors. Nevertheless, given the
various uncertainties, such as those relating to dosimetry, sampling error,
population differences, and possible confounders, it is still possible that medical
X-rays are significantly more carcinogenic, per unit dose, than y-rays (ICRP
2003, NAS 2006).
123
-------
1.0
0.8
Cumulative
fraction Q.6
of
total dose
F 0.4
0.2
101
10-
10-
Electron Energy, T(eV)
Figure 5-2: Cumulative fraction of total dose, F, plotted against secondary electron kinetic
energies, T, for a variety of slow and fast initial electron energies calculated by the Monte Carlo
track structure method (Nikjoo and Goodhead 1991).
In conclusion, there is strong experimental and theoretical support for the
contention that low energy photons and electrons are more biologically effective
than the y-rays from a Co source or those accounting for most of the dose
received by the atomic bomb survivors in Hiroshima and Nagasaki (MAS 2006).
However, this issue can only be fully resolved through experiment and a better
understanding of the dependence of DMA damage and carcinogenesis on micro-
dosimetric parameters. EPA is sponsoring a project aimed at deriving RBE
values for low-LET emissions by specific radionuclides based on calculations of
energy deposition and DMA damage events produced by low-energy electrons.
The NCRP has also convened a committee to address the issue of RBEs for low
energy, low-LET radiation. It is anticipated that these efforts can advance to the
point where adjustments to the risk estimates for tritium, and possibly for other
radionuclides, can be incorporated into the next version of FGR-13.
124
-------
6. Risks from Prenatal Exposures
First carried out by Stewart and coworkers (Stewart et al. 1958, Bithell and
Stewart 1975), case-control studies of childhood cancer have shown about a
40% increase in risk associated with exposure to diagnostic X-rays in utero.
Typically, the X-rays employed in Stewart's "Oxford series" were 80 kVp and the
mean dose was 6-10 mGy; this corresponds to only about 1 photon per cell
nucleus. Hence this finding argues against the likelihood of a threshold for
radiation carcinogenesis.
The estimate of risk for childhood cancer derived from the Oxford survey
is about 0.06 per Gy (95% Cl 0.01-0.126) for all cancers and about 0.025 per Gy
for leukemia (Mole 1990, Doll and Wakeford 1997). Although numerous other
case-control studies have shown a similar radiation-related risk as the Oxford
survey (Doll and Wakeford 1997), the evidence from cohort studies is equivocal
(Boice and Miller 1999). Children exposed in utero to radiation from the atomic
bomb explosions have not experienced any detectable increase in cancer, and
the derived upper bound is lower than the estimate derived from the case-control
studies (Doll and Wakeford 1997). Results from a large cohort study did show an
increase in leukemia of about the same magnitude as the Oxford series, but the
observed increase in childhood solid tumors was much lower and not statistically
significant (Monson and MacMahon 1984). Another question regarding the risk of
solid tumors has been that the excess relative risk seen in the case-control
studies is about the same, regardless of the type of tumor. This may suggest that
the increase is due to some unaccounted for source of confounding (Boice and
Miller 1999).
On balance, the evidence from the epidemiological studies indicates that
the fetus is at risk of childhood cancer from ionizing radiation (Doll and Wakeford
1997). Following the recommendations of Doll and Wakeford (1997) and the
ICRP (2000), EPA adopts the estimate of 0.06 Gy"1 for prenatal exposures to
diagnostic X-rays. Since the individual radiation doses in the Oxford study were
generally quite low, no DDREF adjustment is required to project risks at low
doses or dose rates. However, as discussed in Section 5.2, an RBE > 1 should
perhaps be assigned to X-rays commonly used in medicine. It would then be
appropriate to divide the above estimate by the X-ray RBE to obtain the estimate
of risk for higher energy y-rays and electrons.
It can be inferred from recent SEER data (Altekruse et al. 2010: Tables
28.10 and 29.6) that long-term survival rates for childhood leukemias and solid
cancers are approximately 70-80% (although this may not adequately account for
delayed mortality due to second cancers resulting from the treatment). Based on
those survival rates, the estimated childhood cancer mortality risk coefficient for
prenatal exposures would be 20-30% of the incidence estimate.
125
-------
The studies of medically irradiated fetuses only address the induction of
childhood cancers. Epidemiological follow-up of the A-bomb survivors has
indicated that individuals irradiated in utero may have a lower risk of adult
cancers than those irradiated as young children, but the difference is not
statistically significant (Preston et al. 2008). Based on this finding, we adopt the
same set of models employed for calculating risk for exposure to young children
to assess the risk of adult cancers due to an in utero exposure. More specifically,
we directly applied the risk models of Section 3 with age-at-exposure set to 0.
The sex-averaged projected risk for adult cancers (attained age > 15) is 0.29
Gy"1 for incidence and 0.12 Gy"1 for mortality. This risk is 2 or 3 times higher than
that for the general U.S. population. It is also about 5 times the estimated risk of
a radiogenic childhood cancer from prenatal exposures. Nevertheless it
constitutes only a small fraction (< 3%) of the risk from a uniform whole-body
exposure to the U.S. population.
126
-------
7. Radionuclide Risk Coefficients
Subsequent to publication of this report, EPA plans to use its revised
radiation risk models and ICRP's latest dosimetric models to update the
radionuclide risk coefficients in Federal Guidance Report 13 (EPA 1999b).
Radionuclide risk coefficients are EPA's best estimates of the lifetime excess
mortality or morbidity risk per unit intake of a given radionuclide by ingestion or
inhalation, or per unit exposure for external irradiation. The current version of
FGR-13 contains risk coefficients for environmental exposure to over 800 radio-
nuclides.
Based on the values in Tables 3-17 and 3-18, EPA expects that updated
mortality risk coefficients for those radionuclides that irradiate the body uniformly
will be similar to currently published values, whereas corresponding morbidity
risk coefficients will likely increase by about 35%. For radionuclides irradiating
the body nonuniformly, both increases and decreases are anticipated, depending
on the target organ. For example, updated risk coefficients for inhaled
radionuclides retained in the lung may be larger than present estimates because
the population-averaged lung cancer risk has increased substantially over time.
Conversely, updated risk coefficients for radionuclides that are poorly absorbed
from the intestines into the bloodstream and that emit short-range radiation,
especially a-particles, should be smaller than current values because of a
reduced colon cancer risk coefficient and the adoption of new ICRP alimentary
tract models (ICRP 2006) that place the location of target cells in the intestinal
wall out of range of a-particles emitted from the contents of the colon.
127
-------
8. Noncancer Effects at Low Doses
Hereditary effects. Ionizing radiation can produce mutations in the DMA
of reproductive cells, which may be expressed as harmful hereditary effects in
subsequent generations. Radiation-induced hereditary effects have been demon-
strated in a number of species and have been extensively studied in laboratory
mice. However, a statistically significant excess of these effects has not been
detected in irradiated human populations, including the Japanese atomic bomb
survivors. Epidemiological data can, therefore, only provide an upper bound on
the magnitude of the genetic risk of radiation to humans.
Based on a careful consideration of the data on mice and known differ-
ences in the genetic make-up of mice and humans, the BEIR VII Committee
arrived at a quantitative estimate of the genetic risk to humans. The total risk for
all classes of genetic diseases was estimated to be about 3,000-4,700 cases per
million first-generation progeny per Gy of low dose rate low-LET radiation (MAS
2006). This numerical estimate is defined relative to the "genetically significant
dose," i.e., the combined dose received by both parents prior to conception. The
average parental age at the time of conception is roughly 30. So, for example, in
a population receiving 1 mGy annually, the average genetically significant dose
for each newborn will be approximately 30x2 mGy, or 60 mGy, and the estimated
risk of an adverse genetic effect in the progeny will be (180-280)x10~6. For
comparison, the estimated average lifetime risk of an incident cancer from a
1-mGy/y exposure is: (1.1x10'1 Gy"1) (75x10'3 Gy/lifetime) « 8,000x10'6 Thus, the
estimated number of hereditary effects is low compared to the number of
projected cancers.
Cardiovascular Disease. It is well established that high radiation doses
(> 5 Gy), such as those sometimes administered therapeutically, can produce
cardiovascular disease through direct damage to the structures of the heart and
the coronary arteries (MAS 2000, UNSCEAR 2006, Little et a/. 2008b). In
addition, there is evidence of an increase in cardiovascular disease associated
with much lower doses in the LSS cohort (Preston et a/. 2003, Little et a/. 2008b,
Shimizu et a/. 2010) and a few other groups (Little et a/. 2008b) but the asso-
ciation in the LSS is not statistically significant at doses under 0.5 Sv (Shimizu et
a/. 2010), and the other studies, which focus on occupational cohorts, may suffer
from bias.
Various biological mechanisms have been proposed that might increase
the risk of cardiovascular disease at low doses: e.g., mutational events in dividing
epithelial cells of blood vessels, generating abnormal clones, which can, in turn,
serve as sites for plaque formation. Although such mechanisms cannot be ruled
out, the evidence for a low dose (< 0.5 Gy) risk of cardiovascular disease is not
persuasive, and further research is required to understand the nature of the
association between cardiovascular risk and radiation dose observed at
moderate doses in epidemiological studies (Little etal. 2008b).
128
-------
Cataracts. It is well established that exposure to ionizing radiation leads
to the formation of cataracts (Ainsbury et al. 2009). The suggested mechanism
involves radiation damage to dividing cells in the lens and their subsequent
differentiation and migration, leading to the occurrence of opacities. Cataracts
have been classified as a deterministic effect with a threshold of approximately 2
Gy, but recent data suggest a threshold of no more than about 0.5 Gy. There is,
moreover, evidence of opacity formation in people exposed to chronic low-dose
rate gamma radiation (Chen et al. 2001). Based on current data, it is possible
that cataract induction is a linear, non-threshold phenomenon with a doubling
dose of the order of 2 Gy (Ainsbury et al. 2009).
129
-------
APPENDIX A: Baseline Rates for Cancer and All-Cause Mortality
and Computational Details for Approximating LAR
Baseline rates. Age, gender and cancer site-specific cancer rates were
obtained from NCI using the software packages SEER-Stat (for single ages from
0 to 84) and DevCan (for age categories 85-89, 90-94, and > 95). DevCan, which
is available from the NCI's Surveillance Research Program's website
(http://surveillance.cancer.gov), is software for calculating probabilities for
developing and dying from cancer. Cancer rates in DevCan are conditioned on
being alive and cancer free (at the beginning of each age interval), and were
adjusted upward - usually by no more than about 1% - to account for individuals
with two or more cancers.
For ages 0-84, SEER 13 data were used for 1998-1999 cancer rates, and
SEER 17 data for 2000-2002 cancer rates. Generalized additive models were
applied to the combined cancer rate data (from both SEER and DevCan), using
the software package mcgv (version 1.6-1) in R (Wood 2006). Essentially, cancer
rates were modeled as the sum of smooth (spline) functions of age with terms
that allowed dependence on sex, and dataset (SEER 13 or SEER 17). The R
program used to fit the cancer rate data is available on request.
Cancer rates for both incidence and mortality are graphed in Figure A-1.
Computational details. In Section 3.5, the integrals in Eq. 3-17 and 3-23
for calculating LAR were approximated using monotonic spline functions (Fritsch
and Carlson 1980). However, before applying the spline functions, discontinuities
in inte-grands were removed using a simple smoothly varying function. For
almost all solid cancer sites, these discontinuities occur at the time of minimum
latency (5 y), at which point the BEIR VII models specify that the ERR and EAR
suddenly jump from 0 to some positive value.
The LAR for an exposure at age e is:
110
LAR(D,e)= \M(D,e,d)-S(d)IS(e)da.
0
Here, M(D,e,d) is the excess risk at attained age a that, for most sites, would be
calculated using a BEIR VII ERR or EAR model. For all solid cancer sites other
than bone, M(D,e,d) would be discontinuous at a-e = TSE = 5. In part, because
such discontinuities are not biologically plausible, we replaced values of M with
M*, where
130
-------
M*(D,e,a) = 0, TSE<4
,,
(TSE-4)2 +(TSE-6)
M*(D,e,d)=M(D,e,d), TSE>6.
-M(D,e,d), 4
-------
Liver (males)
Liver (females)
>JU
40
1 30
O
°- 20
o
1
10
/ / '
i I
/ j •
f /
/ /
//
>JU
40
1 30
O
°- 20
o
1
10
-
,— x"
/y~\-
/
^^^
0 20 40 60 80 0 20 40 60 80
Age Age
Lung (males) Lung (females)
500
1 400
8
I 300
o
Q.
£ 200
8.
100
r\
// \ •
I \\
/ \\
// \-
I/ •
1
/
/
\J\J\J
500
1 400
8
I 300
o
Q.
£ 200
8.
100
n
,
/ ~\
/ /^"^ \
/ / v -
//
/
0 20 40 60 80 0 20 40 60 80
Age Age
Bladder (males) Bladder (females)
o 300
I
I 200
o
Q.
100
/ \
/
/
/
1 /
/ /
/ /
tuu
o 300
I
I 200
o
Q.
£
8. 100
^ — ;
0 20 40 60 80 0 20 40 60 80
Age Age
Thyroid (males) Thyroid (females)
8 15
O
8
I 10
o
Q.
^ 5
, — x
,/ X
,/ \-
^_ _^-^
£.\J
8 15
O
8
I 10
o
Q.
1 5
n
0 20 40 60 80 0 20 40 60 80
Age Age
Figure A-1 (continued)
132
-------
Kidney (males)
20 40 60 80
Age
Residual Site Cancers (males)
100
80
1 6°
O
°- 40
I
20
0
Kidney (females)
,
0 20 40 60
Age
80
Residual Site Cancer (females)
IUUU
800
3
O
o 600
o
°- 400
1
200
n
/ '
/
/ / '
/ /
/ /
/ /
jS /^
IUUU
800
8
o
o 600
o
°- 400
I
200
n
^-— J
y/
./ X'-
,/ y^
20 40 60 80
Age
0 20
40 60
Age
80
Leukemia (males)
Leukemia (females)
80
§60
|
S 40
Q.
O
& 20
^~-~^'~
f"
1
n
/
n
/
;t^==^:::/ •
0 20 40 60 80
Age
80
§60
o
S 40
Q.
O
* 20
-
,-
/ '
y/
"^v__ _^^
0 20 40 60 80
Age
Skin (males) Skin (females)
5000
1 4000
3
I 3000
o
Q.
£ 2000
8.
1000
•
•
I
I
/
/
^^^
uuuu
5000
1 4000
3
I 3000
o
Q.
£ 2000
8.
1000
n
/
/••
/
0 20 40 60 80 0 20 40 60 80
Age Age
Figure A-1 (continued)
133
-------
Prostate
Female Breast Cancer
§
o
0~
o
*~
b
Q.
&
8.
I^UU
1000
800
600
400
200
n
L L L L
/ \ /
/ \ '
1
j I
! I -
j i
/ / '
/ /
_^y ^^~/
20 40 60
Age
Uterus
80
0
§
8
b
Q.
i
3.
IZU
100
80
60
40
20
n
/ " — -^
/ \
/ \ '
/ \
/ \
/ \
1
I //"~-
/ ,/
/ ^-^
~/-^- — ; — ~~^^
20 40 60
Age
80
500
400
|
| 300
-------
APPENDIX B: Details of Bayesian Analysis
Data. The dataset is a subset of the incidence data for the follow-up
period 1958-1998, which was analyzed by Preston et al. (2007). The data can be
downloaded from RERF at http://www.rerf.or.jp/library/dle/lssinc07.html (file
Issinc07.csv). The dataset incorporates the latest (DS02) dosimetry and is
otherwise essentially the same as the one used for the BEIR VII analysis, in that
it excludes the "not-in-city" group (see Preston et al. 2007 for details). The data is
in the form of an event-time table, which includes the number of cancer cases
and person-years for subgroups defined by city, sex, and intervals based on
dose, age-at-exposure, attained age, and follow-up time.
Risk models. For most solid cancer sites, BEIR VII ERR models were
used. That is, for a specific cancer site, the ERR for an atomic bomb survivor is:
ERR(D,s,e,d) = j3sDexp(ye*)(a/7Qy ,
where a and e denote attained age and age-at-exposure, and e* is the age-at-
exposure function, which is set to 0 for ages > 30. The corresponding cancer rate
is:
Here, A0 (s, a, b,c) denotes the baseline rate, which depends on sex (s ), attained
age (a ), year of birth (b ), and city (c).
Baseline cancer rate models. For each cancer site, the same sex-
specific parametric models as in Preston et al. are used for the baseline rates
A0(-)'. "In the most general models, for each sex, the log rate was described
using city and exposure status effects together with piecewise quadratic
functions of log age joining smoothly at ages 40 and 70 and piecewise quadratic
functions of birth year joining smoothly at 1915 (age at exposure 30) and 1895
(age at exposure 50). A smooth piecewise quadratic function of x with join points
at xi and x2 can be written as/?0+/^jc+/?2jc2+/?3max(jc-jc1,0)2+/?4max(jc-jc2,0)2. This
parameterization provides flexible but relatively parsimonious descriptions of the
rates."
Prior distributions for baseline cancer rates. For baseline cancer rate
para-meters, the priors were normal distributions with mean 0 and extremely
large variances. This is an example of what are sometimes referred to as non-
informative priors. Use of non-informative priors will often yield results similar to
what would be obtained from more traditional statistical methods, e.g., maximum
likelihood.
135
-------
Prior distributions for ERR model parameters. Lognormal prior distri-
butions were assigned to linear dose-response parameters for age-at-exposure
20 and attained age 70. The younger age-at-exposure is chosen for technical
reasons to increase the speed of the Monte Carlo algorithm. Normal distributions
were assigned to the age-at-exposure and attained-age parameters. Prior
distributions are detailed in Table B-1.
Table B-1: Prior distributions for ERR model parameters1
Parameter
Log(ERR) at age-at-exposure 20 Age-at-
and attained age 70 exposure Attained age
Males Females
Cancer Site (j) 'j l^^M,jJ /,• >»&^F,J) \r d Wj >
Stomach
Colon
Liver
Lung
Bladder
Prostate
Uterus
Ovary
Other solid
#(/Vs7,r2) N(juFj,T2) #(-0.3,0.25)
#(//M,T2) — -0.3
— , -0.3
— 'J -0.3
#(-0.5,10) N(UFJ,T2) #(-0.3,0.25)
#(-1.4
-1.4
-1.4
-1.4
#(-1.4
,2)
,2)
// ~ #(-0.5, 10), pFJ =v + $j, /Vy =f-Sj
All sites
1/r2 ~Gamma(3.5,V), 1/r2. ~Gamma(5.5,l)
1 Linear dose response parameter (/?) in Preston et a/. (2007) represents the ERR for age-at-
exposure 30 and attained-age 70. Here a prior distribution is assigned to the ERR for age-at-
exposure 20 and attained age 70. The younger age-at-exposure (20 instead of 30) is chosen to
reduce correlations that can increase run times for the MCMC algorithm.
Likelihood. The likelihood is based on the assumption that the hazard
function for each cancer can be approximated as a piecewise-linear function of
time. It can then be shown that the likelihood is identical to that for a Poisson
model in which, for each cell within the event time table, the number of expected
cases is equal to the product of the hazard rate and the total person-years. For a
set of several cancers, the likelihood is the product of Poisson likelihoods
associated with each cancer type (Larson 1984).
136
-------
Simulation of posterior distributions using MCMC. The software
package, WinBUGS, was used to simulate three independent "chains" of 25,000
sets of ERR and baseline rate parameter values. In MCMC, burn-in time refers to
the time during which the chains of simulated values have not yet converged to
the target distributions, and it is common practice not to use values simulated
during burn-in. For this analysis, the first 12,500 sets of parameters of each chain
were discarded. To save computer time, the sequences were then "thinned" by
using every third value. The final analysis was based on 12,500 sets of
parameter values.
Considerable care was taken to make sure that the results generated from
MCMC would converge to the target (posterior) distribution. For example, an
initial analysis - based on the assumption that maximum likelihood estimates of
parameter values follow a multivariate normal distribution - was used to generate
starting values, and modified Gelman-Rubin statistics (Brooks and Gelman 1998)
were used to determine whether convergence had been achieved.
The WinBUGS program for simulating the parameter values - together
with starting values used - is available upon request.
Prostate and uterine cancers. A lognormal prior distribution for a linear
dose response parameter (/3) assures that simulated values from the posterior
distribution for that parameter will be positive. However, for both prostate and
uterine cancers, the evidence for a positive dose-response is not statistically
significant. For these two cancers, a set of the simulated values for the linear
dose response parameter (/3), generated using WinBUGS, were randomly
chosen and set to zero. For each site, the percentage of values set to zero was
determined so that the mean of the posterior distribution for LAR would equal the
nominal value for the LAR given in Section 3.
Posterior distributions for ERR parameters. Table B-2 compares
posterior distributions for the linear dose-response parameters (ERR Gy"1 for
age-at-exposure 30 and attained age 70) to the corresponding estimates in BEIR
VII. Except for bladder and colon cancer, the mean and uncertainty interval
bounds for the posterior distributions are remarkably similar to the corresponding
confidence intervals in BEIR VII. The 95% uncertainty interval calculated in this
report for stomach cancer is (0.09, 0.33), whereas the 95% confidence interval
reported in BEIR VII is (0.09, 0.32). In contrast, for female bladder cancer, the
upper 95% uncertainty bound calculated here is only 2.2, versus the upper 95%
Cl bound of 3.2 in BEIR VII. Histograms of posterior distributions for ERR
parameters are given in Figures B-1, B-2 and B-3. We note that, for specific
cancer sites, parameter values are correlated; e.g., the age-at-exposure
parameter and the linear dose response parameter for bladder cancer are
positively correlated.
137
-------
Table B-2: Comparison of posterior distributions for ERR linear dose
response parameter1 with estimates in BEIR VII
Males
Cancer
Stomach
Colon
Liver
Lung
Bladder
Remainder
Prostate
Uterus
Ovary
Posterior
Distribution2
0.19
(0.09,0.33)
0.38
(0.12,0.79)
0.23
(0.07, 0.48)
0.37
(0.17,0.62)
0.50
(0.13,1.1)
0.16
(0.05,0.32)
0.09
(0, 0.42)
BEIR VII3
0.17
(0.09,0.32)
0.51
(0.30, 0.89)
0.26
(0.13,0.52)
0.26
(0.12,0.56)
0.40
(0.15, 1.13)
0.18
(0.10,0.32)
0.10
(0,0.56)
Females
Posterior
Distribution2
0.38
(0.20, 0.60)
0.38
(0.14, 0.73)
0.32
(0.10, 0.65)
1.12
(0.63, 1.7)
0.97
(0.24, 2.2)
0.29
(0.12, 0.52)
0.04
(0, 0.21)
0.34
(0.11, 0.69)
BEIR VII3
0.39
(0.25, 0.59)
0.35
(0.15, 0.77)
0.26
(0.08, 0.81)
1.13
(0.76, 1.7)
1.33
(0.56,3.2)
0.29
(0.18, 0.49)
0.04
(0, 0.18)
0.31
(0.08, 1.13)
1ERR Gy"1 for age-at-exposure 30 and attained age 70
2Mean and 95% uncertainty bounds
3Maximum likelihood estimate and 95% confidence interval
138
-------
0 0.2 0.4 0.6 0.8
Figure B-1: Posterior distributions for ERR at age-at-exposure 30 and attained-age 70
for selected cancer sites. For prostate and uterine cancer, only positive values for ERR
are shown.
139
-------
Figure B-2: Posterior distributions for the age-at-exposure parameter for selected sites
140
-------
Figure B-3: Posterior distributions for the attained age parameter for selected sites
141
-------
REFERENCES
Adams MJ, RE Shore, A Dozier, SE Lipshultz, RG Schwartz, LS Constine et al.
fhyroid cancer risk 40+ years after irradiation for an enlarged thymus: an update
of the Hempelman cohort. 2010. Radiat Res 174: 753-762.
Ainsbury EA, SD Bouffler, W Dorr, J Graw, CR Muirhead, AA Edwards, J
Cooper. 2009. Radiation cataractogenesis: A review of recent studies. Radiat
Res 172: 1-9.
Altekruse SF, CL Kosary, M Krapcho, N Neyman, R Aminou, W Waldron et al.
(eds). 2010. SEER Cancer Statistics Review, 1975-2007, National Cancer
Institute. Bethesda, MD, http://seer.cancer.goV/csr/1975 2007/, based on Nov.
2009 SEER data submission, posted to the SEER website, 2010.
Andersson M, M Vyberg, J Visfeldt, B Carstensen, HH Storm. 1994. Primary liver
tumours among Danish patients exposed to Thorotrast. Radiat Res 137: 262-
273.
Arias E, LR Curtin, RWWei, RN Anderson. 2008. U.S. Decennial Life Tables for
1999-2001, United States Life Tables. National Vital Statistics Reports, Vol 57 No
1. Hyattsville, Maryland: National Center for Health Statistics. Accessed 10/31/08
at CDC website: http://www.cdc.gov/nchs/data/nvsr/nvsr57/nvsr57 01 .pdf
Azzam El, SM de Toledo, GP Raaphorst, REJ Mitchel. 1996. Low-dose ionizing
radiation decreases the frequency of neoplastic transformation to a level below
the spontaneous rate in C3H 10T1/2 cells. Radiat Res 146: 369-373.
Azzam El, SM de Toledo, T Gooding, JB Little. 1998. Intercellular communication
is involved in the bystander regulation of gene expression in human cells
exposed to very low fluences of alpha particles. Radiat Res 150: 497-504.
Bauer S, Bl Gusev, LM Pivina, KN Apsalikov, B Grosche. 2005. Radiation
exposure to local fallout from Soviet atmospheric nuclear weapons testing in
Kazakhstan: solid cancer mortality in the Semipalatinsk historical cohort, 1960-
1999. Radiat Res 164: 409-419.
Berrington A, SC Darby, HA Weiss, R Doll. 2001. 100 years of observation on
British radiologists: mortality from cancer and other causes 1897-1997. Br J
Radiol74:507-519.
Bijwaard H, MJP Brugmans, HP Leenhouts. 2004. Two-mutation model for bone
cancer due to radium, strontium and plutonium. Radiat Res 162: 171-184.
142
-------
Bland Kl, HR Menck, CEH Scott-Conner, M Morrow, DJ Winchester, DP
Winchester. 1998. The national cancer data base 10-year survey of breast
carcinoma treatment at hospitals in the United States. Cancer 83: 1262-1273.
Boice JD, Jr. 2006. Ionizing Radiation, pp 259-293, in: Cancer Epidemiology and
Prevention. Third Edition. D Schottenfeld and JF Fraumeni (eds). Oxford: Oxford
University Press.
Boice JD, Jr., D Preston, FG Davis, RR Monson. 1988. Frequent chest X-ray
fluoroscopy and breast cancer incidence among tuberculosis patients in
Massachusetts. Radiat Res 125: 214-222.
Boice JD, Jr., RW Miller. 1999. Childhood and adult cancer after intrauterine
exposure to ionizing radiation. Teratol 59: 227-233.
Boice JD, Jr., JH Hendry, N Nakamura, 0 Niwa, S Nakamura, K Yoshida. 2010.
Low-dose-rate epidemiology of high background radiation areas. Radiat Res
171:849-854.
Brenner DJ, R Doll, DT Goodhead, EJ Hall, CE Land, JB Little et al. 2003.
Cancer risks attributable to low doses of ionizing radiation: Assessing what we
really know. PNAS 100: 13761-13766.
Brenner DJ, RK Sachs. 2003. Domestic radon risks may be dominated by
bystander effects-but the risks are unlikely to be greater than we thought. Health
Phys85: 103-108.
Brenner AV, D Mykola, MD Tronko, M Hatch, Tl Bogdanova, VA Oliynik et al.
1-131 dose-response for incident thyroid cancers in Ukraine related to the
Chornobyl accident. Environ Health Perspect 2011 Mar 14. [Epub ahead of print].
Brenner DJ, RK Sachs. 2006. Estimating radiation-induced cancer risks at very
low doses: rationale for using the linear no-threshold approach. Radiat Environ
Biophys 44: 253-256.
Breslow NE, NE Day. 1987. Statistical Methods in Cancer Research. Volume II:
The Design and Analysis of Cohort Studies. International Agency for Research
on Cancer. Lyon.
Brooks SP, A Gelman. 1998. General methods for monitoring convergence of
iterative simulations. J Comput Graph Stat 7: 434-455.
Bucci A, E Shore-Freedman, T Gierlowski, D Mihailescu, E Ron, AB Schneider.
2001. Behavior of small thyroid cancers found by screening radiation-exposed
individuals. J Clin Endocrinol Metab. 86: 3711-3716.
143
-------
Burch PRJ. 1957. Some physical aspects of relative biological efficiency. Br J
Radiol 30: 524-529.
Cardis E, M Vrijheid, M Blettner, E Gilbert, M Hakama, C Hill etal. 2005a. Risk of
cancer after low doses of ionising radiation: retrospective cohort study in 15
countries. BMJ 331: 77 doi:10.1136/bmj.38499.599861.EO (published 29 June 2005).
Cardis E, A Kesminiene, V Ivanov, I Malakhova, Y Shibata, V Khrouch et al.
2005b. Risk of thyroid cancer after exposure to 131I in childhood. JNCI 97: 724-
732.
Cardis E, M Vrijheid, M Blettner, E Gilbert, M Hakama, C Hill etal. 2007. The 15-
country collaborative study of cancer risk among radiation workers in the nuclear
industry: Estimates of radiation-related cancer risks. Radiat Res 167: 396-416.
CDC (Centers for Disease Control) and NCI (National Cancer Institute). U.S.
Cancer Statistics Working Group. 2005. United States Cancer Statistics: 2002
Incidence and Mortality. Atlanta: U.S. Department of Health and Human
Services, Centers for Disease Control and Prevention and National Cancer
Institute. Accessed at: http://www.cdc.gov/cancer/npcr/uscs/pdf/2002 uscs.pdf
(7/24/06).
Chen W-L, J-S Hwang, T-H Hu, M-S Chen, WF Chang. 2001. Lenticular
opacities in populations exposed to chronic low-dose-rate gamma radiation from
radiocontaminated buildings in Taiwan. Radiat Res 156: 71-77.
Coates PJ, SA Lorimore, EG Wright. 2004. Damaging and protective cell
signalling in the untargeted effects of ionizing radiation. Mutat Res 568: 5-20.
Cornforth MN, SM Bailey, EH Goodwin. 2002. Dose responses for chromosome
aberrations produced in noncycling primary human fibroblasts by alpha particles,
and gamma rays delivered at sublimiting low dose rates. Radiat Res 158: 43-53.
Cronin K, E Feuer, M Wesley, A Mariotto, S Scoppa, D Green. 2003. Current
Estimates for 5 and 10 Year Relative Survival. Statistical Research and
Applications Branch, National Cancer Institute. Technical Report #2003-04.
Accessed 10/17/08 at: http://srab.cancer.gov/reports/tech2003.04.pdf
Darby S, D Hill, A Auvinen, JM Barros-Dios, H Baysson, F Bochicchio, H Deo et
al. 2005. Radon in homes and risk of lung cancer: collaborative analysis of
individual data from 13 European case-control studies. BMJ 330: 223.
Delongchamp RR, K Mabuchi, Y Yoshimoto, DL Preston. 1997. Cancer mortality
among atomic bomb survivors exposed to in utero or as young children, October
1950-May 1992. Radiat Res 147: 385-395.
144
-------
Doll R, R Wakeford. 1997. Risk of childhood cancer from fetal irradiation. Brit J
Radiol70: 130-139.
Doody MM, JE Lonstein, M Stovall, DG Hacker, N Luckyanov, CE Land. 2000.
Breast cancer mortality after diagnostic radiography: Findings from the U.S.
Scoliosis Cohort Study. Spine 25: 2052-2063.
Edwards AA. 1999. Neutron RBE values and their relationship to judgements in
radiological protection. J Radiol Prot 19: 93-105.
El Ghissassi F, R Baan, K Straif, Y Grosse, B Secretan, V Bouvard et a/., on
behalf of the WHO International Agency for Research on Cancer Monograph
Working Group. 2009. A review of human carcinogens - Part D: Radiation.
Lancet Oncol. 10: 751-752.
Elkind MM. 1994. Radon-induced cancer: a cell-based model of tumorigenesis
due to protracted exposures. Int J Radiat Biol 66: 649-653.
Elmore E, X-Y Lao, R Kapadia, E Gedzinski, C Limoli, JL Redpath. 2008. Low
doses of very low-dose-rate low-LET radiation suppress radiation-induced
neoplastic trans-formation in vitro and induce an adaptive response. Radiat Res
169:311-318.
EPA (Environmental Protection Agency). 1994. Estimating Radiogenic Cancer
Risks. EPA Report 402-R-93-076, Washington, DC: U.S. EPA.
EPA (Environmental Protection Agency). 1999a. Estimating Radiogenic Cancer
Risks. Addendum: Uncertainty Analysis. EPA Report 402-R-99-003, Washington,
DC: U.S. EPA.
EPA (Environmental Protection Agency). 1999b. Cancer Risk Coefficients for
Environmental Exposure to Radionuclides. EPA Report 402-R-99-001,
Washington, DC: U.S. EPA.
EPA (Environmental Protection Agency). 2003. EPA Assessment of Risks from
Radon in Homes. EPA Report 402-R-03-003, Washington, DC: U.S. EPA.
Everitt BS. 1995. The Cambridge Dictionary of Statistics in the Medical Sciences.
Cambridge University Press: NY.
Fritsch FN, RE Carlson. 1980. Monotone piecewise cubic interpolation. SIAM J
Numer Anal 17: 238-246.
Furukawa K, DL Preston, S Lonn, S Funamoto, S Yonehara, T Matsuo et al.
2010. Radiation and smoking effects on lung cancer incidence among atomic
bomb survivors. Radiat Res 174: 72-82.
145
-------
Gelman A, JB Carlin, HS Stern, DB Rubin. 2003. Bayesian Data Analysis.
Chapman and Hall: NY.
Gilbert ES. 1991. Chapter 3: Late somatic effects. In: S Abrahamson, MA
Bender, BB Boecker et al. Health Effects Models for Nuclear Power Plant
Accident Consequence Analysis. Modifications of Models Resulting from Recent
Reports on Health Effects of Ionizing Radiation, Low LET Radiation, Part II:
Scientific Bases for Health Effects Models. NUREG/CR-4214, Rev. 1, Part II,
Addendum 1, LMF-132. U.S. Nuclear Regulatory Commission: Washington, DC.
Gilbert ES, NA Koshurnikova, ME Sokolnikov, VF Khokhryakov, S Miller, DL
Preston et al. 2000. Liver cancers in Mayak workers. Radiat Res 154: 246-252.
Gilbert ES, NA Koshurnikova, ME Sokolnikov, NS Shilnikova, DL Preston, E Ron
et al. 2004. Lung cancer in Mayak workers. Radiat Res 162: 505-516.
Goodhead DT. 2006. Energy deposition stochastics and track structure: What
about the target? Radiat Prot Dosim 122: 3-15.
Grogan HA, WK Sinclair, PG Voilleque. 2001. Risks of fatal cancer from
inhalation of 239'240plutonium by humans: a combined four-method approach with
uncertainty evaluation. Health Phys 80: 447-461.
Hahn FF, BA Muggenburg, RA Guilmette, BB Boecker. 1999. Comparative
stochastic effects of inhaled alpha- and beta-particle-emitting radionuclides in
beagle dogs. Radiat Res 152: S19-S22.
Hahn FF, B Scott, DL Lundgren. 2010. Comparative stochastic effects of alpha,
beta or X-irradiation of the lung of rats. Health Phys 99: 363-366.
Harrison JD, CR Muirhead. 2003. Quantitative comparisons of induction in
humans by internally deposited radionuclides and external radiation. Int J Radiat
Biol 79:1-13.
Hatch M, E Ron, A Bouville, L Zablotska, G Howe. 2005. The Chernobyl disaster:
Cancer following the accident at the Chernobyl Nuclear Power Plant. Epidemiol
Rev 27: 56-66.
Hawkins MM, LMK Wilson, HS Burton, MHN Potok, DL Winter, HB Marsden, MA
Stovall. 1996. Radiotherapy, alkylating agents, and risk of bone cancer after
childhood cancer. JNCI 88: 270-278.
Hildreth NG, RE Shore, PM Dvoretsky. 1989. The risk of breast cancer after
irradiation of the thymus in infancy. N Engl J Med 321: 1281-1284.
146
-------
Howe GR, J Mclaughlin. 1996. Breast cancer mortality between 1950 and 1987
after exposure to fractionated moderate-dose-rate ionizing radiation in the
Canadian fluoroscopy cohort study and a comparison with breast cancer
mortality in the atomic bomb survivors study. Radiat Res 145: 694-707.
HPA (Health Protection Agency) (UK). 2007. Review of Risks from Tritium.
Report of the Independent Advisory Group on Ionising Radiation. Available at:
www.hpa.org.uk/publications/2007/tritium advice/RCE Advice ontritium.pdf
Humphreys ER, KR Isaacs, TA Raine, J Saunders, VA Stones, DL Wood. 1993.
Myeloid leuke
64:231-235.
Myeloid leukemia and osteosarcoma in CBA/H mice given 224Ra. Int J Radiat Biol
Hwang S-L, J-S Hwang, Y-T Yang, WA Hsieh, T-C Chang, H-R Guo et a/. 2008.
Estimates of relative risks for cancers in a population after prolonged low-dose-
rate radiation exposure: A follow-up assessment from 1983 to 2005. Radiat Res
170: 143-148.
ICRP (International Commission on Radiological Protection). 1991. The
Biological Basis for Dose Limitation in the Skin. ICRP Publication 59. Ann ICRP
22(2).
ICRP (International Commission on Radiological Protection). 1993. Age-
dependent Doses to Members of the Public from Intake of Radionuclides: Part 2.
Ingestion Dose Coefficients. ICRP Publication 67. Ann ICRP 23(3/4).
ICRP (International Commission on Radiological Protection). 2000. Pregnancy
and Medical Radiation. ICRP Publication 84. Ann ICRP 30(1).
ICRP (International Commission on Radiological Protection). 2003. Relative
Biological Effectiveness (RBE), Quality Factor (Q), and Radiation Weighting
Factor (WR). ICRP Publication 92. Ann ICRP 33(4).
ICRP (International Commission on Radiological Protection). 2006. Human
Alimentary Tract model for Radiological Protection. ICRP Publication 100. Ann
ICRP 36 (1-2).
ICRP (International Commission on Radiological Protection). 2007. The 2007
Recommendations of the International Commission of Radiological Protection.
ICRP Publication 103. Ann ICRP 37(2-4).
Ishikawa Y, JAH Humphreys, CG Collier, ND Priest, Y Kato, T Mori, R
Machinami. 1999. Revised organ partition of thorium-232 in thorotrast patients.
Radiat Res 152:8102-8106.
147
-------
Jacob P, W Ruhm, L Walsh, M Blettner, G Hammer, H Zeeb. 2009. Is cancer risk
of radiation workers larger than expected? Occup Environ Med 66: 789-796.
James AC, A Birchall, G Akabani. 2004. Comparative dosimetry of BEIR VI
revisited. Radiat Prot Dosimetry 108: 3-26.
Kadhim MA, DA MacDonald, DT Goodhead, SA Lorimore, SJ Marsden, EG
Wright. 1992. Transmission of chromosomal instability after plutonium alpha-
particle irradiation. Nature 355: 738-740.
Kadhim MA, SR Moore, EH Goodwin. 2004. Interrelationships amongst radiation-
induced genomic instability, bystander effects, and the adaptive response. Mutat
Res 568: 21-32.
Karagas MR, ER Greenberg, SK Spencer, TA Stukel, LA Mott. 1999. The New
Hampshire Skin Cancer Study Group. Increase in incidence rates of basal cell
and squamous cell skin cancer in New Hampshire, USA. Int J Cancer 81: 555-
559.
Kaul DC, SD Egbert, WA Woolson. 2005. Chapter 13: DS02 uncertainty
analysis, pp. 921-926 in: Reassessment of the Atomic Bomb Radiation
Dosimetry for Hiroshima and Nagasaki: Dosimetry System 2002. Volume
2. RW Young and GW Kerr, eds. Hiroshima: Radiation Effects Research
Foundation.
Kesminiene A, A-S Evrard, VK Ivanov, IV Malakhova, J Kurtinaitis, A Stengrevics
et al. 2008. Risk of hematological malignancies among Chernobyl liquidators.
Radiat Res 170: 721-735.
Kocher DC, Al Apostoaei, FO Hoffman. 2005. Radiation effectiveness factors for
use in calculating probability of causation of radiogenic cancers. Health Phys
89:3-32.
Kocher DC, Al Apostoaei, RW Henshaw, FO Hoffman, MK Schubauer-Berigan,
DO Stancescu et al. 2008. Interactive Radioepidemiological Program (IREP): A
Web-based tool for estimating probability of causation/assigned share of
radiogenic cancers. Health Phys 95: 119-147.
Koshurnikova NA, ES Gilbert, M Sokolnikov, VF Khokhryakov, S Miller, DL
Preston et al. 2000. Bone cancers in Mayak workers. Radiat Res 154: 237-245.
Krestinina LY, DL Preston, EV Ostroumova, MO Degteva, E Ron, 0V Vyushkova
et al. 2005. Protracted radiation exposure and cancer mortality in the Techa
River Cohort. Radiat Res 164: 602-611.
148
-------
Krestinina LY, DL Preston, FG Davis, S Epifanova, E Ostroumova E Ron, A
Akleyev. 2010. Leukemia incidence among people exposed to chronic radiation
from the contaminated Techa River, 1953-2005. Radiat Environ Biophys 49: 195-
201.
Krewski D, JH Lubin, JM Zielinski, M Alavanja, VS Catalan, RW Field et al. 2005.
Residential radon and the risk of lung cancer: A combined analysis of 7 North
American case-control studies. Epidemiol 16: 137-145.
Land CE. 1980. Estimating cancer risks from low doses of ionizing radiation.
Science 209: 1197-1203.
Land CE, WK Sinclair. 1991. The relative contributions of different organ sites to
the total cancer mortality associated with low-dose radiation exposure. In: Risks
Associated with Ionising Radiations, Annals of the ICRP 22 (1).
Larson MG. 1984. Covariate analysis of competing-risks data with log-linear
models. Biometrics 40: 459-469.
Leenhouts HP, MJP Brugmans, M Andersson, HH Storm. 2002. A reanalysis of
liver cancer incidence in Danish patients administered Thorotrast using a two-
mutation carcinogenesis model. Radiat Res 158: 597-606.
Lehnert BE, EH Goodwin, A Deshpande. 1997. Extracellular factor(s) following
exposure to alpha particles can cause sister chromatid exchanges in normal
human cells. Cancer Res 57: 2164-2171.
Lewis EB. 1963. Leukemia, multiple myeloma, and aplastic anemia in American
radiologists. Science 142: 1492-1494.
Lewis KG, MA Weinstock. 2004. Nonmelanoma skin cancer mortality (1988-
2000). The Rhode Island follow-back study. Arch Dermatol 140: 837-842.
Little JB, H Nagasawa, T Pfenning, H Vetrovs. 1997. Radiation-induced genomic
instability: delayed mutagenic and cytogenetic effects of X-rays and a-particles.
Radiat Res 148: 299-307.
Little MP. 2002. Absence of evidence for differences in the dose-response for
cancer and non-cancer endpoints by acute injury status in the Japanese atomic-
bomb survivors. Int J Radiat Biol 78: 1001-1010.
Little MP, DG Hoel, J Molitor, JD Boice Jr., R Wakeford, CR Muirhead. 2008.
New models for evaluation of radiation-induced lifetime cancer risk and its
uncertainty employed in the UNSCEAR 2006 report. Radiat Res 169: 660-676.
149
-------
Little MP, EJ Tawn, I Tzoulaki, R Wakeford, G Hildebrandt, F Paris et a/. 2008b.
A systematic review of epidemiological associations between low and moderate
doses of ionizing radiation and late cardiovascular effects and their possible
mechanisms. Radiat Res 169: 99-109.
Lloyd EL, D Hodges. 1971. Quantitative characterization of bone. A computer
analysis of microradiographs. Clin Orthop Relat Res 78: 230-250.
Lobrich M, N Rief, M Kuhne, M Heckmann, J Fleckenstein, C Rube, M Uder.
2005. In vivo formation and repair of DMA double-strand breaks after computed
tomography examinations. PNAS 102: 8984-8989.
Loucas BD, R Eberle, SM Bailey, MN Cornforth. 2004. Influence of dose rate on
the induction of simple and complex chromosome exchanges by gamma rays.
Radiat Res 162: 339-349.
Lubin J, E Ron (1998). Excess relative risk and excess absolute risk estimates
for pooled analysis of thyroid cancer following exposure to external radiation,
memorandum to Jack Schull, Chair, NRC Committee on Exposure of the
American People to 1-131 from Nevada Atomic-Bomb Tests: Implications for
Public Health (National Cancer Institute, Rockville, Maryland).
Lubin J, DW Schafer, E Ron, M Stovall, RJ Carroll. 2004. A reanalysis of thyroid
neoplasms in the Israeli tinea capitis study accounting for dose uncertainties.
Radiat Res 61: 359-368.
Lunn DJ, A Thomas, N Best, D Spiegelhalter. 2000. WinBUGS - a Bayesian
modeling framework: concepts, structure, and extensibility. Stat Comput 10: 325-
337.
Lyng FM, CB Seymour, C Mothersill. 2000. Production of a signal by irradiated
cells which leads to a response in unirradiated cells characteristic of initiation of
apoptosis. Brit J Cancer 83: 1223-1230.
Markova E, N Schultz, IY Belyaev. 2007. Kinetics and dose-response of residual
53BP1/v-H2AXfoci: Co-localization, relationship with DSB repair and clonogenic
survival. Internat J Radiat Biol 83: 319-329.
Marples B, MC Joiner. 1993. The response of Chinese hamster V79 cells to low
radiation doses: Evidence of enhanced sensitivity of the whole cell population.
Radiat Res 133: 41-51.
Mays CW, MP Finkel. 1980. RBE of a-particles vs. (3-particles in bone sarcoma
induction. Proc. IRPA5: 401-405.
150
-------
Mays CW, H Speiss. 1972. Bone sarcoma incidence vs. alpha particle dose, pp.
409-430 in: Radiobiology of Plutonium, BJ Stover and WSS Jee, eds. Salt Lake
City: The J.W. Press.
Mays CW, RE Rowland, AF Stehney. 1985. Cancer risk from lifetime intake of Ra
and U isotopes. Health Phys 48: 635-647.
Mitchell SA, G Randers-Pehrson, DJ Brenner, EJ Hall. 2004. The bystander
response in C3H 10T% cells: the influence of cell-to-cell contact. Radiat Res 161:
397-401.
Mole RH. 1990. Childhood cancer after prenatal exposure to diagnostic X-ray
examinations in Britain. Br J Cancer 62: 152-168.
Monson RR, B MacMahon. 1984. Prenatal X-ray exposure and cancer in
children, pp. 197-206 in: Radiation Carcinogenesis: Epidemiology and Biological
Significance, JD Boice, Jr. and JF Fraumeni, Jr., eds., New York: Raven Press.
Morgan WF, JP Day, Ml Kaplan, EM McGhee, CL Limoli. 1996. Genomic
instability induced by ionizing radiation. Radiat Res 146: 247-258.
Mothersill C, CB Seymour. 1998. Cell-cell contact during gamma irradiation is not
required to induce a bystander effect in normal human keratinocytes: evidence
for release during irradiation of a signal controlling survival into the medium.
Radiat Res 149: 256-262.
Muirhead CR, JA O'Hagan, RGE Haylock, MA Phillipson, T Wilcock GLC
Berridge, W Zhang. 2009. Mortality and cancer incidence following occupational
radiation exposure: third analysis of the National Registry for Radiation Workers.
BrJ Cancer 100: 206-212.
Nair RR, B Rajan, S Akiba, P Jayalekshmi, MK Nair, P Gangadharan etal. 2009.
Background radiation and cancer incidence in Kerala, India-Karanagappally
cohort study. Health Phys 96: 55-66.
NAS (National Academy of Sciences). 1988. Health Effects of Radon and Other
Internally Deposited Alpha-Emitters (BEIR IV). Washington, DC: National
Academy Press.
NAS (National Academy of Sciences). 1990. Health Effects of Exposure to Low
Levels of Ionizing Radiation. (BEIR V). Washington, DC: National Academy
Press.
NAS (National Academy of Sciences). 1999. Health Effects of Exposure to
Radon. BEIR VI. Washington, DC: National Academy Press.
151
-------
MAS (National Academy of Sciences). 2006. Health Risks from Exposure to Low
Levels of Ionizing Radiation. BEIR VII Phase 2. Washington, DC: National
Academy Press.
NCHS (National Center for Health Statistics). 2004. Bridged-Race Population
Estimates, United States, 1990 - 2004, by Age Groups. Compiled from the April
1, 2000 resident population developed by the Bureau of the Census in
collaboration with the NCHS - accessed from the CDC WONDER On-line
Database on June 3, 2006.
NCI (National Cancer Institute) 2011. Fast Stats: An Interactive Tool for Access
to SEER Cancer Statistics. Surveillance Research Program, National Cancer
Institute, http://seer.cancer.gov/faststats. (Accessed on 1-4-2011)
NCI (National Cancer Institute) 2005. DevCan: Probability of Developing or Dying
of Cancer Software, Version 6.4.1; Statistical Research and Applications Branch,
National Cancer Institute. http://srab.cancer.gov/devcan (Accessed on 5-6-2010)
NCRP (National Council on Radiation Protection and Measurements). 1980.
Influence of Dose and Its Distribution in Time on Dose-Response Relationships
for Low-LET Radiations. NCRP Report No. 64. Bethesda, MD: National Council
on Radiation Protection and Measurements.
NCRP (National Council on Radiation Protection and Measurements). 1985.
Induction of Thyroid Cancer by Ionizing Radiation. NCRP Report No. 80.
Bethesda, MD: National Council on Radiation Protection and Measurements.
NCRP (National Council on Radiation Protection and Measurements). 1990. The
Relative Effectiveness of Radiations of Different Quality. NCRP Report No. 104.
Bethesda, MD: National Council on Radiation Protection and Measurements.
NCRP (National Council on Radiation Protection and Measurements). 2001.
Evaluation of the Linear-Nonthreshold Dose-Response Model for Ionizing
Radiation. NCRP Report No. 136. Bethesda, MD: National Council on Radiation
Protection and Measurements.
NCRP (National Council on Radiation Protection and Measurements). 2008. Risk
to Thyroid from Ionizing Radiation. NCRP Report No. 159. Bethesda, MD:
National Council on Radiation Protection and Measurements.
Nekolla EA, M Kreisheimer, AM Kellerer, M Kuse-lsingschulte, W Gossner, H
Speiss. 2000. Induction of malignant bone cancers in radium-224 patients: risk
estimates based on improved dosimetry. Radiat Res 153: 93-103.
Nikjoo H, DT Goodhead. 1991. Track structure analysis illustrating the prominent
role of low-energy electrons in radiobiological effects of low-LET radiations. Phys
Med Biol 36: 229-238.
152
-------
Nikjoo H, S Uehara, WE Wilson, M Hoshi, DT Goodhead. 1998. Track structure
in radiation biology: theory and applications. Int J Radiat Biol 73: 355-364.
Nikjoo H, L Lindborg. 2010. RBE of low energy electrons and photons. Phys Med
Biol 55: R65-R109.
NIOSH (National Institute for Occupational Safety and Health). 2002. NIOSH-
Interactive RadioEpidemiological Program (NIOSH-IREP), Technical Documen-
tation. Cincinnati, OH: National Institute for Occupational Safety and Health.
Accessed 6/2006 at: http://www.cdc.gov/niosh/ocas/pfs/irep/irepfnl.pdf.
Ostroumova E, B Gagniere, D Laurier, N Gudkova, L Krestinina, P Verger et al.
2006. Risk analysis of leukemia incidence among people living along the Techa
river: a nested case-control study. J Radiol Prot 26: 17-32.
Pawel D, D Preston, D Pierce, J Cologne. 2008. Improved estimates of cancer
site-specific risks for A-bomb survivors. Radiat Res 169: 87-98.
Pierce DA, GB Sharp, K Mabuchi. 2003. Joint effects of radiation and smoking on
lung cancer risk among atomic bomb survivors. Radiat Res 159: 511-520.
Pierce DA, M Vaeth, Y Shimizu. 2007. Selection bias in cancer risk estimation
from A-bomb survivors. Radiat Res 167: 735-741.
Pierce DA, M Vaeth, JB Cologne. 2008. Allowance for random dose estimation
errors in atomic bomb survivor studies: a revision. Radiat Res 170: 118-126.
Portess Dl, G Bauer, MA Hill, P O'Neill. 2007. Low-dose irradiation of nontrans-
formed cells stimulates the selective removal of precancerous cells via inter-
cellular induction of apoptosis. Cancer Res 67: 1246-1253.
Pottern LM, MM Kaplan, PR Larsen, JE Silva, RJ Koenig, JH Lubin et al. 1990.
Thyroid nodularity after irradiation for lymphoid hyperplasia: A comparison of
questionnaire and clinical findings. J Clin Epidemiol 43: 449-460.
Preston DL, JH Lubin, DA Pierce, ME McConney. 1993. Epicure Users Guide.
Hirosoft International Corporation, Seattle, WA, 1993.
Preston DL, E Ron, S Yonehara, T Kobuke, H Fujii, M Kishikawa et al. 2002a.
Tumors of the nervous system and pituitary gland associated with atomic bomb
radiation exposure. J Natl Cancer Inst 94: 1555-1563.
Preston DL, A Mattsson, E Holmberg, R Shore, NG Hildreth, JD Boice Jr. 2002b.
Radiation effects on breast cancer risk: a pooled analysis of eight cohorts. Radiat
Res 158: 220-235.
153
-------
Preston DL, DA Pierce, Y Shimizu, HM Callings, S Fujita, S Funamoto, K
Kodama. 2004. Effects of recent changes in atomic bomb survivor dosimetry on
cancer mortality risk estimates. Radiat Res 162: 377-389.
Preston DL, E Ron, S Tokuoka, S Funamoto, N Nishi, M Soda, K Mabuchi, K
Kodama. 2007. Solid cancer incidence in atomic bomb survivors: 1958-1998.
Radiat Res 168: 1-64.
Preston DL, H Cullings, A Suyama, S Funamoto, N Nishi, M Soda et al. 2008.
Solid cancer incidence in atomic bomb survivors exposed in utero or as young
children. JNCI: 100: 428-436.
Priest ND, DG Hoel, PN Brooks. 2006. Relative toxicity of chronic irradiation by
45Ca beta particles and 242Cm alpha particles with respect to the production of
lung tumors in CBA/Ca mice. Radiat Res 166: 782-793.
Puskin JS. 2008. What can epidemiology tell us about risks at low doses? Radiat
Res 169: 122-124.
Raabe 0, SA Book, NJ Parks. 1983. Lifetime bone cancer dose-response
relationships in beagles and people from skeletal burdens of 226Ra and 90Sr.
Health Phys. 44: 33-48.
Ramsey ML. 2006. Basal Cell Carcinoma. Accessed at eMedicine website:
www.emedicine.com/derm/topic47.htm, last updated May 9, 2006.
Redpath JL, RJ Antoniono. 1998. Induction of an adaptive response against
spon-taneous neoplastic transformation in vitro by low-dose gamma radiation.
Radiat Res 149: 517-520.
Richardson D, H Sugiyama, N Nishi, R Sakata, Y Shimizu, E Grant et al. 2009.
Ionizing radiation and leukemia mortality among Japanese atomic bomb
survivors: 1950-2000. Radiat Res 172:368-382.
Ries LAG, D Harkins, M Krapcho, A Mariotto, BA Miller, L Clegg et al. (eds).
2006. SEER Cancer Statistics Review, 1975-2003. National Cancer Institute.
Bethesda, MD, http://seer.cancer.goV/csr/1975 2003/ based on November 2005
SEER data submission, posted to the SEER website, 2006.
Ries LAG, JL Young, GE Keel, MP Eisner, YD Lin, M-J Horner (editors). SEER
Survival Monograph: Cancer Survival Among Adults: U.S. SEER Program, 1988-
2001, Patient and Tumor Characteristics. National Cancer Institute, SEER
Program, NIH Pub. No. 07-6215, Bethesda, MD, 2007.
Ries LAG, D Melbert, M Krapcho, DG Stinchcomb, N Howlader, MJ Horner et al.
(editors). SEER Cancer Statistics Review, 1975-2005. National Cancer Institute.
154
-------
Bethesda, MD, http://seer.cancer.goV/csr/1975 2005/, based on November 2007
SEER data submission, posted to the SEER website, 2008.
Roesch WC. 1987. Radiation Effects Research Foundation and the National
Academy of Sciences (U.S.). US-Japan Joint Reassessment of Atomic Bomb
Radiation Dosimetry in Hiroshima and Nagasaki. Minami-ku, Hiroshima:
Radiation Effects Research Foundation.
Romanenko AY, SC Finch, M Hatch, JH Lubin, VG Bebeshko, DA Bazyka et al.
2008. The Ukrainian-American study of leukemia and related disorders among
the Chornobyl cleanup workers from Ukraine: III. Radiation risks. Radiat Res
170:711-720.
Ron E, B Modan, JD Boice Jr., E Alfandary, M Stovall, A Chetrit, L Katz. 1988.
Tumors of the brain and nervous system after radiotherapy in childhood. NEJM
319: 1033-1039.
Ron E, B Modan, D Preston, E Alfandary, M Stovall, JD Boice Jr. 1989. Thyroid
neoplasia following low-dose radiation in childhood. Radiat Res 120: 516-531.
Ron E, B Modan, D Preston, E Alfandary, M Stovall, JD Boice. 1991. Radiation-
induced skin carcinomas of the head and neck. Radiat Res 125: 318-325.
Ron E, JH Lubin, RE Shore, K Mabuchi, B Modan, L Pottern et al. 1995. Thyroid
cancer after exposure to external radiation: a pooled analysis of seven studies.
Radiat Res 141: 259-277.
Ron E, B Modan, DL Preston, E Alfandary, M Stovall, JD Boice. 1989. Thyroid
neoplasia following low-dose radiation in childhood. Radiat Res 120: 516-531.
Ron E, DL Preston, M Kishikawa, T Kobuke, M Iseki, S Tokuoka etal. 1998. Skin
tumor risk among atomic-bomb survivors in Japan. Cancer Causes and Control
9,393-401.
Rossi HH, AM Kellerer. 1986. The dose rate dependence of oncogenic trans-
formation by neutrons may be due to variation of response during the cell cycle
(Letter to the Editor). Int J Radiat Biol 47: 731-734.
Rothkamm K, M Lobrich. 2003. Evidence for a lack of DNA double-strand break
repair in human cells exposed to very low x-ray doses. PNAS 100: 5057-5062.
Sasaki MS, T Nomura, Y Ejima, H Utsumi, S Endo, I Saito et al. 2008.
Experimental derivation of relative biological effectiveness of A-bomb neutrons in
Hiroshima and Nagasaki and implications for risk assessment. Radiat Res 170:
101-117.
155
-------
Schneider AB, E Ron, J Lubin, M Stovall, TC Gierlowski. 1993. Dose-response
relationships for radiation-induced thyroid cancer and thyroid nodules: Evidence
for the prolonged effects of radiation on the thyroid. J Clin Endocrinol Metab 77:
362-369.
(SEER 2009a). Surveillance, Epidemiology, and End Results (SEER) Program
(www.seer.cancer.gov) SEER*Stat Database: Incidence - SEER 13 Regs
Research Data, Nov 2009 Sub (1992-2007)
-------
Shore RE, M Moseson, X Xue, Y Tse, N Harley, BS Pasternack. 2002. Skin
cancer after X-ray treatment for scalp ringworm. Radiat Res 157: 410-418.
Smith PG, R Doll. 1981. Mortality from cancer and all causes among British
radiologists. Br J Radiol 49: 187-194.
Sokolnikov ME, ES Gilbert, DL Preston, NS Shilnikova, W Khokhryakov, EK
Vasilenko, NA Koshurnikova. 2008. Lung, liver, and bone cancer mortality in
Mayak workers. Int J Cancer 123: 905-911.
Stewart A, J Webb, D Hewitt. 1958. A survey of childhood malignancies.
BrMedJ 1: 1495-1508.
Tapio S, V Jacob. 2007. Radioadaptive response revisited. Radiat Environ
Biophys46: 1-12.
Thompson DE, K Mabuchi, E Ron, M Soda, S Tokunaga, S Ochikubo et al. 1994.
Cancer incidence in atomic bomb survivors. Part II: Solid tumors, 1958-1987.
Radiat Res 137 (Suppl.): S17-S67.
Tronko MD, GR Howe, Tl Bogdanova, AC Bouville, 0V Epstein, AB Brill et al.
2006. A cohort study of thyroid cancer and other thyroid diseases after the
Chornobyl accident: thyroid cancer in Ukraine detected during first screening.
JNCI 98: 897-903.
Tubiana M, A Aurengo, D Averbeck, A Bonnin, B Le Guen, R Masse et al. 2005.
Dose-Effect Relationships and the Estimation of the Carcinogenic Effects of Low
Doses of Ionizing Radiation. Paris: Academy of Medicine (Paris) and Academy of
Science. Joint Report No. 2.
Tucker MA, GJ D'Angelo, JD Boice Jr, LC Strong, FP Li, M Stovall et al. 1987.
Bone sarcomas linked to radiotherapy and chemotherapy in children. N Engl J
Med 317: 588-593.
Ullrich RL, RJ Preston. 1987. Myeloid leukemia in male RFM mice following
irradiation with fission spectrum neutrons or gamma-rays. Radiat Res 109:165-
170.
UNSCEAR (United Nations Scientific Committee on the Effects of Atomic
Radiation). 1993. UNSCEAR 1993 Report to the General Assembly with
Scientific Annexes. New York: United Nations.
UNSCEAR (United Nations Scientific Committee on the Effects of Atomic
Radiation). 2000a. UNSCEAR 2000 Report to the General Assembly with
Scientific Annexes. Volume I: Sources. New York: United Nations.
157
-------
UNSCEAR (United Nations Scientific Committee on the Effects of Atomic
Radiation). 2000b. UNSCEAR 2000 Report to the General Assembly with
Scientific Annexes. Volume II: Effects. New York: United Nations.
UNSCEAR (United Nations Scientific Committee on the Effects of Atomic
Radiation). 2008. UNSCEAR 2006 Report to the General Assembly with
Scientific Annexes. Volume I. Effects of Ionizing Radiation. New York: United
Nations.
Van Kaick G, A Dalheimer, S Hornik, A Kaul, D Liebermann, H Luhrs et al. 1999.
The German Thorotrast study: recent results and assessment of risks. Radiat
Res 152:864-871.
Vu BL, F de Vathaire, A Shamsaldin, MM Hawkins, E Grimaud, C Hardiman et al.
1998. Radiation dose, chemotherapy and risk of osteosarcoma after solid
tumours during childhood. Int J Cancer 77: 370-377.
Wang J-X, JD Boice Jr, B-X Li, J-Y Zhang, JF Fraumeni Jr. 1988. Cancer among
medical diagnostic x-ray workers in China. J Natl Cancer Inst 80: 344-350.
Weinstock MA. 1994. Epidemiologic investigation of nonmelanoma skin cancer
mortality: the Rhode Island follow-back study. J Invest Dermatol 102:65-95.
Wick RR, EA Nekolla, W Gossner, AM Kellerer. 1999. Late effects in
ankylosing spondylitis patients treated with 224Ra. Radiat Res 152: S8-S11.
Wick RR, EA Nekolla, M Gaubitz, TL Schulte. 2008. Increased risk of
myeloid leukaemia in patients with ankylosing spondylitis following
treatment with radium-224. Rheumatology 47: 855-859.
Wood SN. 2006. Generalized Additive Models: An Introduction with R. Chapman
and Hall: New York.
Young RW, H Hasai. 2005. Executive Summary, pp. 8-41 in: Reassessment
of the Atomic Bomb Radiation Dosimetry for Hiroshima and Nagasaki:
Dosimetry System 2002. Volume 1. RW Young and GW Kerr, eds.
Hiroshima: Radiation Effects Research Foundation.
158
-------
GLOSSARY
Absorbed dose: The energy deposited by ionizing radiation per unit mass of
tissue irradiated. It can be expressed in units of gray (Gy) or milligray
(mGy) where 1 Gy = 1000 mGy.
Adaptive response: A reduced response to radiation induced by a prior dose.
Alpha particle (a-particle): A particle consisting of 2 protons and 2 neutrons
emitted from a decay of certain heavy atomic nuclei; a type of high-LET
radiation.
Apoptosis: Programmed cell death.
BCC: Basal cell carcinoma.
Baseline cancer rate: The cancer mortality or incidence rate in a population in
the absence of the specific exposure being studied.
Bayesian: A statistical approach in which probability reflects the state of
knowledge about a variable, often incorporating subjective judgment.
BEIR VII: A National Research Council Report, Health Risks from Exposure to
Low Levels of Ionizing Radiation. BEIR VII. Phase 2.
Beta particle (p-particle): An electron emitted from a decay of an atomic
nucleus; a type of low-LET radiation.
Bystander effect: A change in a cell due to irradiation of a nearby cell.
Confidence interval (Cl): A range of values calculated from sample
observations that are believed, with a particular probability to contain the
true parameter value. Upper and lower values of a Cl are called
confidence limits. A 90% Cl implies that if the estimation process were
repeated many times, about 90% of the intervals would contain the true
value. The 90% probability refers to the properties of the interval and not
the parameter itself.
Confounder: In an epidemiological study, a factor that is associated with both
the exposure and outcome of interest and thereby distorts or masks the
true effect of the exposure.
159
-------
Credible interval: In Bayesian statistics, credible intervals are used instead of
confidence intervals to describe a range of parameter values which
contain the true value with a particular probability. A 90% credible interval
is an interval which contains the quantity of interest with posterior
probability of 90%.
Dose and dose-rate effectiveness factor (DDREF): A factor used to account
for an apparent decrease in the effectiveness of low-LET radiation in
causing a biological end-point (e.g., cancer) at low doses and dose rates
compared with observations made at high, acutely delivered doses.
Dose effectiveness factor (DEF): A factor estimated from the LQ model to
account for a decrease in the effectiveness of low-LET radiation in causing
a biological end-point (e.g., cancer) at low doses compared with that at
high acute doses.
Dose equivalent: A weighted sum of absorbed doses of different types of radi-
ation, measured in units of sieverts (Sv). The ICRP recommended values
for the weighting factors wr are: 1.0 for photons and electrons, 10 for
fission neutrons, and 20 for a-particles. Thus, for low-LET radiation, the
dose equivalent in Sv is numerically equal to the absorbed dose in Gy,
whereas for a-particles an absorbed dose of 1 Gy corresponds to 20 Sv.
Dose rate effectiveness factor (DREF): A factor used to account for an
apparent decrease in the effectiveness of low-LET radiation in causing a
biological end-point (e.g., cancer) at low dose rates compared with high
dose rates.
Double strand break (DSB): DMA damage in which a break extends over both
strands of the double helix.
Electron volt (eV): The customary unit of energy for all ionizing radiations'. 1 eV
is equivalent to the energy gained by an electron passing through a
potential difference of 1 volt. 1 keV = 1000 eV; 1 MeV = 1,000,000 eV.
EPA: U.S. Environmental Protection Agency.
Excess absolute risk (EAR): The rate of disease in an exposed population
minus that in an unexposed population. Also termed "attributable risk."
Excess relative risk (ERR): The fractional increase in the rate of disease in an
exposed population compared to that in an unexposed population. The
ERR is equal to the RR-1.
Gamma rays (y-rays or gamma radiation): Photons of nuclear origin similar to
X- rays but usually of higher energy. A type of low-LET radiation.
160
-------
Genomic instability: An enhanced rate of spontaneous genetic change in a cell
population.
Geometric mean (GM): The GM of a set of positive numbers is the exponential
of the arithmetic mean of their logarithms.
Geometric standard deviation (GSD): The GSD of a lognormal distribution is
the exponential of the standard deviation of the associated normal
distribution.
Gray (Gy): Unit of absorbed dose (1 Gy = 1 joule/kg).
High-LET radiation: Radiation, such as neutrons or a-particles, producing
ionizations densely spaced on a molecular scale (e.g., LET > 10 keV/um).
HPA: Health Protection Agency of the United Kingdom.
ICRP: International Commission on Radiological Protection. An independent
international organization providing recommendations and guidance on
radiation protection.
Ionizing radiation: Any radiation capable of removing electrons from atoms or
molecules as it passes through matter, thereby producing ions.
kVp (kV): Kilovolt potential - refers to the potential difference between the
electrodes of an X-ray tube. For example, the output of a 200 kVp X-ray
tube will consist of photons with a range of energies up to 200 keV.
LET: Average amount of energy lost per unit track length of an ionizing charged
particle.
Life Span Study (LSS): RERF's long term epidemiological study of health
effects in the Hiroshima and Nagasaki atomic bomb survivors.
Life table: A table showing the number of persons who, of a given number born
or living at a specified age, live to attain successively higher ages,
together with the number who die in each interval.
Lifetime attributable risk (LAR): The LAR approximates the probability that an
individual will develop (die from) cancer associated with an exposure. It
includes incident cases (deaths) that would have occurred later in time
without the exposure.
Likelihood: In statistics, this refers to the probability of a set of observations
given values for a set of parameters.
161
-------
Linear no-threshold (LNT) model: Dose-response for which any dose greater
than zero has a positive probability of producing an effect. The probability
is calculated from the slope of a linear (L) model or from the limiting slope,
as the dose approaches zero, of a linear-quadratic (LQ) model.
Linear (L) model: A model in which the probability of an effect (e.g., cancer) is
expressed as being proportional to the dose.
Linear-quadratic (LQ) model: A model in which the probability of an effect (e.g.,
cancer) is expressed as the sum of two terms - one proportional to the
dose, the other to the square of the dose. In the limit of low doses and low
dose rates, the quadratic term can be ignored.
Low-LET radiation: Radiation, such as X-rays, y-rays or electrons, producing
sparse ionizing events on a molecular scale (e.g., LET < 10 keV/um).
Lognormal distribution: A distribution in which the logarithm of a randomly
distributed quantity has a normal distribution.
Mortality (rate): the frequency at which people die from a specific cause (e.g.,
lung cancer), often expressed as the number of deaths per 100,000
population per year.
NCRP: National Council on Radiation Protection and Measurements. A Council
commissioned to formulate and disseminate information, guidance, and
recommendations on radiation protection and measurements.
NIOSH: National Institute for Occupational Safety and Health.
Orthovoltage X-rays: X-rays produced by generators in the 200-500 kV range.
Orthovoltage X-ray sources of about 200-250 kV have been extensively
used as irradiators in radiobiology.
Photon: A quantum of electromagnetic energy. Energetic photons in the form of
X-rays or y-rays can ionize atoms or molecules in a medium upon which
they are incident.
Posterior probability distribution: In Bayesian inference, posterior distributions
are probability distributions that incorporate all that is known about a
set of random quantities or parameter values, after obtaining information
from empirical data. The posterior distribution for a parameter value is
proportional to the product of the prior distribution and the likelihood.
162
-------
Prior probability distribution: In Bayesian inference, prior distributions are
probability distributions summarizing information about a set of
parameters that is known or assumed, prior to obtaining further
information from empirical data.
Radiation effectiveness factor (REF): A quantity comparing the cancer causing
potency in humans of a specified type of radiation relative to some
standard.
Radiation Effects Research Foundation (RERF): A joint Japan-U.S. research
organization, based in Hiroshima and Nagasaki, for studying the health
effects of radiation on the atomic bomb survivors.
Radiation risk: The increased probability of a cancer (or cancer death) due to a
given dose of radiation.
Relative biological effectiveness (RBE): The relative effectiveness of a given
type of radiation in producing a specified biological effect compared to
some reference radiation. For purposes of this document, the reference
radiation is generally taken to be low dose y-rays.
RBEM: The maximal limiting value of the RBE for a high-LET radiation attained in
the limit of low doses.
Relative risk (RR): The rate of disease in an exposed population divided by that
in an unexposed population.
Relative survival: Net survival measure representing cancer survival in the
absence of other causes of death.
Risk coefficient: The increase in the annual incidence or mortality rate per unit
dose: (1) absolute risk coefficient is the increase in the incidence or
mortality rate per unit dose; (2) relative risk coefficient is the fractional
increase above the baseline incidence or mortality rate per unit dose.
SCC: Squamous cell carcinoma.
SEER: Surveillance, Epidemiology, and End Results. A data base of cancer
statistics collected from registries throughout the U.S.
Sievert (Sv): Unit of dose equivalent. In the BEIR VII analysis of the A-bomb
survivor data, the dose equivalent was calculated from the absorbed y-ray
and neutron doses, assuming a radiation weighting factor of 10 for
neutrons.
163
-------
Stationary population: A hypothetical population in which the relative number of
people of a given age and gender is proportional to the probability of
surviving to that age.
Uncertainty: A term used to describe the lack of precision and accuracy of a
given estimate.
Uncertainty distribution: A mathematical expression defining the relative
probabilities of different values for an estimated quantity.
Uncertainty factor: A random factor by which an estimate or projection deviates
from its "true" value due to a specific source of uncertainty such as
DDREF or risk transport.
UNSCEAR: United Nations Scientific Committee on the Effects of Atomic
Radiation. A UN committee that publishes reports on sources and effects
of ionizing radiation.
WLM: Working level months, a measure of radon decay product exposure.
X radiation or X-rays: Energetic photons usually produced by bombarding a
metallic target with fast electrons in a high vacuum. The potential (kVp)
difference between the target (cathode) and the collecting plate (anode)
limits the maximum energy of X-rays produced. "Orthovoltage" X-rays of
200-250 kVp have been commonly used as a source of photons for
experiments in radiobiology. Diagnostic X-rays employed in medicine are
typically in the 50-150 kVp range, except for mammography, where the
typical voltage is about 30 kVp.
164
------- |