United States
Environmental Protection
Agency
Air And Radiation
(6601J)
EPA 402-R-99-003
May 1999
Estimating Radiogenic
Cancer
Addendum:
Uncertainty Analysis
-------
EPA 402-R-99-003
ESTIMATING RADIOGENIC CANCER RISKS
ADDENDUM: UNCERTAINTY ANALYSIS
May 1999
U.S. Environmental Protection Agency
401 M Street S.W.
Washington, DC 20460
-------
11
-------
PREFACE
In November 1997, a draft version of this report (dated October 1997) was
submitted by the Office of Radiation and Indoor Air (ORIA) to EPA's Science Advisory
Board (SAB) for review. As part of the review, the Uncertainty in Radiogenic Cancer
Risk Subcommittee (URRS) of the SAB's Radiation Advisory Committee (RAC) held
two public meetings in Washington, DC, on November 20, 1997 and March 4, 1998 to
receive briefings from ORIA staff "and interested members of the public and to discuss
the relevant issues. The SAB's report (EPA-SAB-RAC-99-008) was transmitted on
February 18, 1999, to EPA Administrator, Carol Browner, with a cover letter signed by:
Dr. Joan M. Daisey, Chair of the SAB; Dr. Stephen L. Brown, Chair of the RAC; and Dr.
F. Owen Hoffman, Chair of the URRS. The SAB approved ORIA's approach to
estimating uncertainty but suggested a number of possible improvements in the ORIA
document or, in some cases, in future assessments of uncertainties. EPA responded to
the issues raised in an April 29 letter from Robert Perciasepe, Assistant Administrator,
Office of Air and Radiation to Dr. Daisey and made a number of changes in the final
document to address SAB concerns.
This report was prepared by EPA staff members, Jerome S. Puskin, Christopher
B. Nelson, and David J. Pawel, Office of Radiation and Indoor Air (ORIA), Radiation
Protection Division (RPD). The authors gratefully acknowledge the thoughtful
comments received during the review process from the members of the URRS, F.O.
Hoffman, S.L. Brown, W. Bair, P.O. Groer, D.G. Hoel, E. Mangione, L.E. Peterson, W.J.
Schull, S.L. Simon, and A.C. Upton, as well as from additional reviewers, W.K. Sinclair,
S. Jablon, C.E. Land, D.A. Pierce, and S.S. Yaniv. We also wish to thank Jonas
Geduldig for developing the computer program, "Murky Ball," which was used to
perform many of the Monte Carlo calculations in the development of this report.
The mailing address for the authors is:
U.S. Environmental Protection Agency
Office of Radiation and Indoor Air (6602J)
Washington, DC 20460
-------
IV
-------
ABSTRACT
In 1994, EPA published a report entitled Estimating Radiogenic Cancer Risks
(EPA 402-R-93-076), which described the Agency's methodology for deriving estimates
of excess cancer morbidity and mortality due to low doses of ionizing radiation. Using
this methodology, numerical estimates of the risk per unit dose were derived for each
applicable cancer site, and for both low-LET and alpha-particle radiation.
Subsequently, small adjustments were made to the procedure used in the 1994
document, chiefly the use of more recent vital statistics. These adjustments produced a
slight increase in the estimated average risk from uniform, whole-body radiation: the
low-LET nominal estimate increased from 5.1 *10'2 Gy'1 to 5.75* 10"* Gy"1. In this
document, a method is described for estimating the uncertainties in the EPA risk
projections. The uncertainty in each site-specific (or whole-body) risk estimate is
treated as the product of several independent sources of uncertainty, e.g., sampling
errors in the epidemiologic data underlying the risk model, or uncertainty in the
extrapolation of observations at high acute doses to chronic low dose conditions. A
distribution is assigned to each source of uncertainty, which defines the probability that
the assumption employed in EPA's risk model with respect to this source of uncertainty
either underestimates or overestimates the risk by any specified amount. The joint
probability distribution for the uncertainty due to all sources combined is then
calculated using Monte Carlo techniques. A detailed uncertainty analysis is performed
for the risks from uniform, low-LET irradiation of the whole body, the lung, and the bone
marrow. For the whole body or the bone marrow, the upper limit on the 90%
confidence interval is about 2 times higher, and the lower limit is about 3 times lower,
than the respective nominal risk estimate. In the case of the lung, the upper bound is
also about a factor of 2 higher, but the lower bound is about a factor of 5 times lower,
than the nominal estimate.
V
-------
VI
-------
CONTENTS
Section Page
PREFACE iii
ABSTRACT v
TABLES AND FIGURES V ix
I. Introduction 1
II. Update of Risk Estimates 1
III. Uncertainty Analysis 2
A. Sampling Variability 3
B. Diagnostic Misclassification 5
C. Temporal Dependence 6
D. Transport of Risk Estimates from Japanese A-Bomb Survivors 7
E. Errors in Dosimetry 8
F. Low Dose (Low Dose Rate) Extrapolation 10
G. Alpha Particle RBE 15
H. Uncertainty in Whole-Body Risk 18
1. Sampling variation 18
2. Diagnostic rhisclassification 19
3. Temporal dependence 19
4. Transport of risk estimates 19
5. Errors in dosimetry 21
6. Error in the choice of DDREF 21
7. Calculation of uncertainty 21
I. Uncertainty in Incidence Estimates 22
J. Numerical Estimates of Uncertainty in Specific Organ Risks 23
1. Uncertainty in lung cancer risk estimate 24
2. Uncertainty in leukemia risk estimate 27
K. Discussion 30
REFERENCES 32
Vll
-------
VI11
-------
LIST OF TABLES
Table Title Page
1 EPA low dose, low dose rate cancer mortality risk estimates 2
2 Uncertainties due to sampling errors in the LSS 4
3 Dosimetric uncertainties 9
4 Multiplicative, NIH and GMC projections of specific organ risks 20
5 Distributions used to estimate the uncertainty in the risk of low level,
low-LET, whole body irradiation 22
6 Distributions used to estimate uncertainty in risk of lung cancer mortality
from low dose irradiation of the lung 27
7 Distributions used to estimate the uncertainty in leukemia risk
from low dose irradiation of the bone marrow 30
LIST OF FIGURES
Figure Title Page
1a DDREF uncertainty distributions 13
1b DDREF cumulative uncertainty distributions 13
IX
-------
X
-------
I. Introduction
In 1994, EPA published Estimating Radiogenic Cancer Risks ("EPA94"), which
described the Agency's methodology for calculating excess cancer morbidity and
mortality risks due to ionizing radiation (EPA 1994). For most cancer sites, a "CMC
model" was employed in which each age-, sex- and site-specific relative risk coefficient
was obtained by taking a geometric mean of the corresponding coefficients in the
"multiplicative" and "NIH" projection models derived from Japanese A-bomb survivors
Lifespan Study (LSS) data. Subsequently, the risk projections in EPA94 were updated
in light of more recent (1990) U.S. Vital Statistics (EPA 1997). In this document, a
methodology is developed for estimating the uncertainties in the EPA risk projections.
Using that methodology, quantitative estimates of uncertainty in cancer mortality risk
estimates are derived for low dose, low-LET exposures to the whole body, the lung,
and the bone marrow.
II. Update of Risk Estimates
In 1997, EPA issued Federal Guidance Report No. 13, Interim Version (FGR 13),
which provides risk estimates for chronic, low level exposures to over 150 radionuclides
through a number of pathways (internal exposure through ingestion or inhalation,
external exposure from surface and soil contamination or from submersion) (EPA
1997). [The final version of FGR 13 is expected to be published in 1999.] The
methodology in FGR 13 differs from that in EPA94 in two main respects: (1) the use of
newly recommended ICRP age-specific dosimetry models and (2) the substitution of
1990 for 1980 U.S. Vital Statistics. The discussion here focuses on the site-specific
risks per unit dose; for this purpose, only the latter of these changes is relevant.
In EPA94, 1980 U.S. baseline cancer mortality rates were used to: (1) derive the
relative risk coefficients for the NIH/GMC projection models and (2) project risk in the
U.S. population based on the site-specific relative risk coefficients. In FGR 13,
baseline cancer rates for these purposes were obtained from the 1990 U.S. Vital
Statistics. In addition, EPA's methodology for projecting risk utilizes a survival
function, which gives the probability that a person in the population will survive to any
specified age. In FGR 13 calculations, the survival function was also updated to reflect
1990 Vital Statistics. For consistency, other minor changes were made to the risk
model in FGR 13 (EPA 1997).
Table 1 lists the revised site-specific risk estimates presented in FGR 13. The
previous EPA94 values, calculated using 1980 Vital Statistics, are shown for
comparison. For the most part, the changes are minimal. In particular, the estimated
cancer mortality risk associated with uniform whole-body irradiation has increased only
slightly, from 5.09x10'* per Gy to 5.75*10"2 per Gy. The most notable change appears
in the lung cancer risk, reflecting an increase in the baseline rate between 1980 and
-------
1990. [Use of the 1990 Vital Statistics also changes cancer incidence risk estimates
slightly, the whole-body morbidity risk increasing from 7.60x10"2 to 8.46x10"2 per Gy.]
TABLE 1
EPA low dose, low dose rate cancer mortality risk estimates
Cancer Site EPA 1994 FGR 13
Esophagus
Stomach
Colon
Liver
Lung
Bone
Skin
Breast
Ovary
Bladder
Kidney
Thyroid
Leukemia
Remainder
TOTAL
9.0
44.4
98.2
15.0
71.6
0.9
1.0
46.2
16.6
24.9
5.5
3.2
49.6
123.1
509.1
11.7
40.7
104.2
15.0
98.8
0.9
1.0
50.6
14.9
23.8
5.1
3.2
55.7
149.5
575.2
III. Uncertainty Analysis
An analysis of the uncertainties in the fatal cancer risk estimate for uniform
whole-body, low-LET radiation has been published by the NCRP in its Report No. 126
(NCRP 1996). Based on its analysis, the NCRP committee arrived at an uncertainty
range of 1.5*10~2/Gy to 8.2x10"2/Gy. In some respects, the analysis and results for
whole-body, low-LET radiation here closely parallels that in NCRP No. 126. However,
we expand the framework to include estimation of uncertainties in specific organ risks,
and to include high-LET as well as low-LET radiation. A detailed analysis is provided
for lung cancer and leukemia. For other sites, probability distributions are indicated for
some sources of uncertainty, but additional analysis would be required to arrive at
quantitative uncertainty bounds on the risk coefficient. Coupled with estimates of
uncertainty in organ doses resulting from the intake of internally deposited
radionuclides, these estimates of uncertainty in organ-specific risk per unit dose could
-------
be used to calculate uncertainties in the risks from ingested or inhaled radionuclides.
However, in summing the organ-specific contributions, careful attention must be given
to possible correlations in the uncertainties for the various sites.
To quantify the uncertainties associated with the risk estimates derived here
primarily from the LSS, a methodology similar to that previously employed for
assessing the uncertainties in radon risk can be applied (Puskin 1992). First, in a
manner similar to that described by Sinclair (1993), the uncertainty in each site specific
risk estimate is treated mathematically as the product of several independent sources
of uncertainty, including: sampling variations, errors in dosimetry, and errors in medical
ascertainment with respect to the epidemiological data; the modeling of the
dependence of risk on age at exposure and time since exposure; the transport of risk
estimates from the study population to the U.S. population; the extrapolation to low
doses and dose rates; and, for high-LET radiation, uncertainty in RBE. Second, a
distribution is assigned to each source of uncertainty, which defines the probability that
the assumption employed in the model pertaining to this source of uncertainty either
underestimates or overestimates the risk by any specified amount. Finally, the joint
probability distribution for the combined uncertainty due to all sources is calculated
analytically as was done for radon or, as here, with the aid of Monte Carlo techniques.
Necessarily, subjective judgment is usually required in assigning "probability"
distributions. Where this is the case, the uncertainty will be characterized as a
"subjective confidence interval" rather than as a "confidence interval." [The term
"credibility interval" has also been used for this purpose (NIH 1985).]
A. Sampling Variability
For the statistical analysis of the LSS data, the deaths and person-years of
survival for the 75,991 individuals in the DS86 subcohort were aggregated by city, sex,
6 age (at time of bomb) categories, 7 follow-up intervals, and 10 radiation dose
intervals (Shimizu et al. 1990). Site-specific risk coefficients were calculated with a
maximum likelihood estimation method, which assumes that the number of deaths in
each group are independent Poisson variates.
Based on this analysis, Shimizu et al. have derived excess relative risk
estimates, with associated 90% confidence intervals, for a number of sites. These are
listed in Table 2. It can be seen that, for most sites, the upper and lower confidence
interval limits, U and L, respectively, are distributed nearly symmetrically about the
central estimate of the risk coefficient, as one would expect if the errors are distributed
normally about this value. [The more skewed distributions occur for organs such as
esophagus and ovary, which would ordinarily contribute only a small proportion of the
total risk from a given exposure.] Hence, for each site, i, we shall represent the
uncertainty due to sampling variation as a normal distribution with mean (Uj+Li)/2 and
-------
standard deviation (Ui-Li)/3.29. This distribution, in general, has a mean displaced
slightly (upward) from the maximum likelihood estimate but with the same 90%
confidence interval limits obtained from the statistical analysis of the data.
TABLE 2
Uncertainties due to sampling variations in the LSS
Cancer Site
Leukemia
All cancers except
leukemia
Esophagus
Stomach
Colon
Lung
Female breasf
Ovary
Urinary tract
Excess Relative Risk
per Gya
5.21 (3.83, 7.12)
0.41 (0.32,0.51)
0.58(0.13,1.24)
0.27 (0.14, 0.43)
0.85(0.39, 1.45)
0.63 (0.35, 0.97)
1.19(0.56,2.09)
1.33(0.37,2.86)
1.27(0.53,2.37)
Derived Distribution
Mean (s.d.)b
5.48(1.00)
0.42 (0.06)
0.69 (0.34)
0.29 (0.09)
0.92 (0.32)
0.66 (0.19)
1.33(0.47)
1.62(0.76)
1.45(0.56)
a Relative risk coefficient and 90% confidence interval from Shimizu et al. (1990).
b Mean and standard deviation of normal distribution used to characterize the
uncertainty in the relative risk coefficient (see text).
c Included here for completeness. EPA's breast cancer risk estimate is based on data
collected on medically irradiated North American women (EPA94: Section IV).
The uncertainties listed in Table 2 reflect the uncertainties in the age-averaged
relative risk coefficient determined over the period of epidemiologic follow-up. The
percent error in the coefficient applicable to any particular age-at-exposure group is
highly variable. Since those exposed as children have just entered their cancer prone
years, sampling uncertainties are particularly high for these survivors, who so far show
the highest excess relative risks per Gy. Consequently, when risks are projected over
a lifetime and summed across age groups to obtain population risks, the uncertainty
due to sampling error may be considerably larger than indicated. The magnitude of this
uncertainty will depend, in part, on how much each age-at-exposure group contributes
-------
to the lifetime risk, which in turn depends on the detailed characteristics of the age and
temporal projection model (see below). Thus, the uncertainties due to sampling
variations and due to temporal modeling are not really independent as assumed here.
A comprehensive assessment of these two sources of uncertainty would require an
extensive reanalysis of the LSS data.
B. Diagnostic Misclassification
Two types of diagnostic misclassification of cancer can occur: classification of
cancers as noncancers (detection error) and erroneous classification of non-cancer
cases as cancer (confirmation error). The former leads to an underestimate of the
excess absolute risk (EAR), but does not affect the estimated excess relative risk
(ERR). Conversely, the latter leads to an underestimate of the ERR but does not affect
the EAR (NCRP 1997).
Based on results from an RERF autopsy study, Sposto et al. (1992) estimated
that, due to diagnostic misclassification between cancer and noncancer causes of
death, the estimated ERR of induced cancers in the LSS population should be
corrected upward by a factor of 1.13. Using information in Sposto et al., but two
different calculational procedures, the NCRP 126 Committee found that the 90%
confidence range for the correction factor to the ERR due to confirmation error is 1.09-
1.18, or 1.095-1.156. However, for the purposes of its uncertainty analysis, the
Committee assigned a probability distribution N(1.1,0.05), which translates into a 90%
subjective confidence interval from 1.02 to 1.18.
For most solid tumors, EPA's risk model (CMC model) coefficients were obtained
by a geometric averaging of the corresponding relative risk coefficients for the
multiplicative and NIH projection models. The correction factor of 1.13 is appropriate
for the multiplicative but not the NIH model, even though the latter is also a relative risk
model. However, the NIH model is constructed in such a way that it projects
approximately the same absolute risk in the U.S. population (over a 40-y period after
exposure) as observed in the LSS. For this reason, the correction for misclassification
in the NIH model should be the same as for the EAR rather than the ERR estimates.
Based on the approach outlined by Sposto et al. (1992), Pierce et al. (1996) estimate
that the EAR should be adjusted upward by about 16% to reflect errors in diagnostic
misclassification.
In view of the information above, we estimate that the EPA projection of the risk
from whole-body irradiation should be increased by about 15% to adjust for
misclassification error. We have assigned a normal probability distribution,
N(1.15,0.06), to the correction factor for misclassification, corresponding to a 90%
subjective confidence interval from 1.05-1.25. Misclassification errors vary
considerably by cancer site, both with respect to proper identification of cancer as the
-------
cause of death and with respect to the primary site. Consequently, the assignment of
an uncertainty distribution associated with diagnostic misclassification should be
readdressed in the case of nonuniform irradiation.
C. Temporal Dependence
A substantial fraction of the estimated population risk is associated with
childhood exposures. As discussed in EPA94 (Section II.B), there is considerable
uncertainty in the estimated risk from doses received by children. First, statistical
uncertainties in the age-specific risk coefficients are generally large, especially for the
youngest age groups among the A-bomb survivors, since those individuals are now just
entering the years of life in which cancers are commonly expressed. Second, there is
suggestive evidence that for some types of solid tumors the excess relative risk due to
childhood exposures may decrease over time (Little et al. 1991).
With respect to adult exposures, a temporal fall-off in the excess relative risk of
lung cancer has been observed in radon-exposed underground miners (NAS 1988) and
in irradiated spondylitic patients (Darby et al. 1987), but there is no clear indication of
such a fall-off in the A-bomb survivors who were over age 20 at the time of the bomb
(Little and Charles 1989). For childhood exposures, Little et al. (1991) have concluded
that there is evidence of a temporal fall-off. Based on an analysis of observed temporal
trends in risk among 4 cohorts of children exposed to radiation, including those from
the LSS, they estimate that the lifetime risk for solid tumors in the (UK) population may
be 30-45% lower than projected by the constant relative risk model.
In developing its estimate of uncertainty in lifetime risk associated with temporal
projection, NCRP Report No. 126 notes that a model in which the excess relative risk
per Sv depends only on attained age appears to fit the LSS cancer mortality data as
well as a model in which the relative risk depends only on the age at exposure (but see
Little et al. 1997). The former model, proposed by Kellerer and Barclay (1992), projects
about a factor of 2 lower risk. Noting also that there is some possibility that the relative
risk may actually increase with age, the Committee suggested a triangular distribution
of uncertainty with a most likely value of 1.00 and a range from 0.50 to 1.10.
In assigning uncertainties associated with temporal projection, three classes of
cancer sites should be considered:
(1) sites for which follow-up is essentially complete, with relatively few additional
radiation induced cancers expected past the present period of epidemiological follow-
up. For this group/which might include bone sarcomas and leukemia, the uncertainty
in lifetime risk associated with temporal projection outside the period of follow-up would
be small (however, see Section J.2).
-------
(2) sites for which a constant relative risk model has been used to project risk
beyond the period of follow-up, but for which the risk coefficients are dependent on the
age at exposure. This group includes stomach, colon, lung, breast, thyroid, and
remainder sites. Most of the projected lifetime risk for these sites is associated with
exposures before age 20. As discussed above, the contribution of childhood
exposures is highly uncertain in view of the statistical limitations and possible
decreases in relative risk with time after exposure. For this group of sites, the model
appears more likely to overestimate than to underestimate the population risk. We
assign a range of 0.5-1.0 to the uncertainty factor associated with age and temporal
dependence for each of these sites except the colon. For colon, a large fraction of the
estimated population risk is from childhood exposures, but the childhood risk is based
on a very limited number of observed cancers; hence, for this organ, we assign a larger
range of uncertainty, 0.4-1.0.
For any individual site, the distribution of uncertainty is assumed to be uniform
on an arithmetic scale; e.g., in the case of the colon, it is assumed that if we consider
only the uncertainty in time projection, the nominal risk estimate should be multiplied by
a factor of x, where x is a random variable uniformly distributed in (0.4,1).
(3) sites for which a constant relative risk projection has been used, but for
which the risk coefficient reflects a single age-averaged value. This group includes
esophagus, liver, bladder, kidney, ovary, and skin. The data available on these sites
are generally sketchy and heavily weighted towards adult exposures. It is plausible
that childhood exposures may convey a higher risk than adult exposures for these
sites, as they appear to do for other sites. Consequently, the model used to project
risk, in this case, may tend to understate the population risk. Typically, the relative
risks for childhood exposures are found to be 2 to 3 times the average for adults
(Shimizu et a/. 1990). If risks for childhood exposures are similarly elevated for the
sites in question, the population risks would be increased by roughly 50%. On the other
hand, some fall-off in relative risk may occur for these sites even in the case of adult
exposures. An uncertainty factor between 0.8 and 1.5 is judged to be reasonable for
these sites; the factor is again assumed to be distributed uniformly within this interval
(on an arithmetic scale).
D. Transport of Risk Estimates from Japanese A-Bomb Survivors
As discussed in Sections II and IV of EPA94, uncertainty exists over how to
apply the results of the analysis of the Japanese A-bomb survivors to the estimation of
risk in the US population, particularly for cancer sites which exhibit markedly different
baseline rates in the two populations. Reflecting this uncertainty, EPA has adopted a
model for most sites in which the age- and sex-specific risk coefficients are a geometric
mean of the corresponding coefficients used in the multiplicative and NIH projection
models. In transporting risk across populations, the multiplicative model presumes that
-------
the excess risk will scale with the baseline cancer rate, whereas the NIH model
presumes that the excess risk is nearly independent of differences in the baseline rate.
Viewed from a mechanistic standpoint, the former presumes that radiation risks act
multiplicatively with the other risks for cancer, while the latter presumes that the
interaction is roughly additive.
For those organs where the GMC model has been employed to calculate risk,
the NIH and multiplicative projections will be used as uncertainty bounds on the
component of uncertainty associated with transporting the risk. In view of the lack of
information on how radiation interacts with other factors affecting carcinogenesis, the
distribution between these two bounds is taken to be uniform on a logarithmic scale.
[The "loguniform" distribution has the property that the probability of finding the random
variable in a small interval dx is dx/x; thus, it is weighted more towards lower values
than a distribution which is uniform over the same range on an arithmetic scale.]
It is not obvious how the uncertainties in specific organ risks associated with
transport across populations should be combined in evaluating the uncertainty in risk
due to an exposure to multiple organs. On one hand, the uncertainties in individual
organ risks could be considered as independent; on the other, as perfectly correlated.
In the former instance, the random variable defining the point on the distribution
between the multiplicative and NIH model would be sampled independently for each
site. In the latter, the same value of the random variable would be assigned to each
site. For uniform, whole-body irradiation, the transport uncertainty does not contribute
strongly to the uncertainty in total cancer risk (see Section H), and it makes little
difference which of these methods is chosen. In the absence of information to the
contrary, and for calculational convenience, the transport uncertainty will be taken to be
independent of cancer site.
E. Errors in Dosimetry
Random errors in the individual dose estimates for the A-bomb survivor
population have been estimated at 25-45% (Jablon 1971, Pierce et al. 1990, Pierce
and Vaeth 1991). The net result of such errors is to overestimate the average dose in
the high dose groups (Pierce and Vaeth 1991). As a result, for a linear fit to the data,
the slope of the dose response will be biased low by roughly 10% (Pierce et al. 1990).
More significantly, perhaps, the shape of the dose response will be distorted towards a
convex (downward) curvature; hence, a true linear-quadratic dependence may be
distorted to look linear (Pierce and Vaeth 1991). This possible distortion has been
discussed in EPA94 (Section II.B.) and is factored into the subjective estimate of
uncertainty in the DDREF presented in the next section.
Measurements of neutron activation products indicate that DS86 may seriously
underestimate neutron doses for Hiroshima survivors, the relative magnitude of the
-------
error increasing with distance from the epicenter (Straume et al. 1992). If neutron
doses have been underestimated, then a larger fraction of the radiogenic cancers
would be attributable to neutrons, and the estimated gamma ray risk would have to be
reduced. Using the tentative revised estimates of neutron flux derived by Straume et
al., and assuming a neutron RBE of 20, Preston et al. (1993) have calculated that the
gamma ray risk estimate for all cancers other than leukemia may be reduced by about
22%. Alternatively, if a neutron RBE of 10 is assumed, the estimated reduction in
gamma ray risk is about 13%. Since the ratio of organ doses to kerma doses for both
neutrons and gamma rays vary somewhat by organ, the magnitude of the estimated
error would also vary by cancer site. Hence, for a nonuniform dose distribution, the
appropriate correction may be larger or smaller than the average value calculated by
Preston et al.
NCRP Report No. 126 identified two additional sources of uncertainty relating to
the DS86 dosimetry: bias in gamma ray estimates and uncertainty in neutron RBE.
Thus, altogether, the Committee analyzed four distinct sources of error relating to DS86
dosimetry, which they took to be uncorrelated. An uncertainty distribution was ascribed
to each of the four sources of error, as summarized in Table 3.
TABLE 3
Dosimetric uncertainties
Parameter
Symbol
Parameter
Value at
Peak
Distribution
Range* or
90% Cl**
Random errors
Neutron weight
Neutron dose
Gamma-ray
free field
f(RE)
f(NR)
f(Dn)
1.1
1.0
1.1
1.1
Normal
Triangular
Triangular
Triangular
1.0 to 1.2
0.9 to 1.1
1.0 to 1.3
1.0 to 1.4
* Range of triangular distributions
** 90% confidence interval for the normal distribution
-------
The combined distribution, f(D), representing the multiplicative uncertainty in risk
due to all four sources of dosimetric uncertainty, was derived using Monte Carlo
techniques and the equation below:
f(RE)
f(D)= (1)
f(DY) x f(NR) x f(Dn)
The resulting distribution f(D) is approximately normal with a mean of 0.84 and a 90%
confidence interval from 0.69 to 1.0; the standard deviation of this normal distribution is
about 0.095. The NCRP distribution for dosimetric uncertainty is adopted here for the
analysis of the uncertainty in whole-body risk.
In estimating the error in risk associated with possible underestimation of the
neutron dose Preston et a/, used the colon dose as a surrogate for the whole-body
dose. Transmission of neutrons to some organs is significantly higher than to the colon
(see: MAS 1990, p. 195). For these organs, the relative error in the risk coefficient due
to an underestimation of neutron dose would be increased (see Section J).
F. Low Dose (Low Dose Rate) Extrapolation
Radiogenic cancer risk estimates are primarily based on observed excess
cancer deaths among A-bomb survivors receiving acute doses of 0.1 to 4 Gy. It would
appear that epidemiology can provide no direct information on the very small risks that
may arise from environmental exposures to radiation (-0.1 Gy/y). Risk estimates at low
doses and dose rates are extrapolations based upon radiobiological data and our
current theoretical understanding of radiation carcinogenesis. This extrapolation is
usually the most important source of uncertainty in estimates of risk from environmental
exposures to low-LET radiation. A detailed analysis of the issues involved in this
extrapolation has been published by the United Nations (UNSCEAR 1993).
Carcinogenesis is understood to be a multistage process in which a single cell
gives rise to a tumor, with mutation of DMA required in one or more of the steps leading
to malignancy. Since cancer is a common disease, the background rates for each of
these steps must be greater than zero, and any filtration mechanism for removing
precancerous cells must be imperfect. Hence, any exposure that increases the rate of
somatic mutations would be expected to have some risk of causing cancer. Traversal
of a single ionizing track through a cell appears to be capable of causing DMA damage
that cannot always be faithfully repaired. A dose threshold for radiation carcinogenesis
is therefore unlikely.
Studies at the molecular, cellular, tissue, and whole-animal level have
demonstrated that radiation damage increases with dose and that, at least for low-LET
10
-------
radiation, it is often greater, per unit exposure, at high doses and dose rates than at low
doses and dose rates. Qualitatively, this can be understood as a reduction in repair
efficiency at the higher doses and dose rates, either due to induction of more complex
damage or due to saturation of repair enzymes (Goodhead 1982). The reduction in
effectiveness at low doses and dose rates, relative to that observed at high acute
doses, is commonly expressed in terms of a dose and dose rate effectiveness factor
(DDREF).
The damage from low-LET-radiation at low and moderate doses is commonly
modeled as a linear-quadratic (LQ) function of dose (R= aD + p D2). At low doses, the
relationship reduces to a linear function of dose, and experiments on animals or
mammalian cells generally indicate that the relative contribution of the quadratic term is
negligible below about 0.2 Gy. In this domain, multi-track effects are presumed to be
negligible, and, as a result, the response there is expected to be independent of dose
rate. Supporting this picture are results of experiments showing an equivalence of the
slope of the dose response observed at low doses with that observed when high doses
are fractionated or delivered chronically (NCRP 1980); however, this equivalence does
not seem to be universal (UNSCEAR 1993). There are compelling reasons to believe
that the dose response for induction of mutations or cancer should be linear down in
the dose range (« 0.001 Gy) where multiple traversals of cell nuclei are rare.
However, direct evidence from epidemiology or radiobiology below about 0.01 to 0.1
Gy is lacking. Consequently, without a fuller understanding of the mechanisms
involved in radiation carcinogenesis, a significant deviation from linearity below 0.01
Gy cannot be categorically ruled out, even though the dose response derived from
epidemiological data at higher doses appears to be fairly linear.
According to the LQ model, the linear component of the dose response is
expected to be predictive of the risk at very low doses and dose rates. The DDREF, in
this view, is obtained from the ratio of the slopes calculated using linear and LQ fits to
the data, respectively. Under the assumption of a LQ dose response, the maximum
likelihood estimate for the DDREF derived from the LSS data is about 2 for leukemia
but only about 1 for solid tumors (Shimizu et al. 1990, MAS 1990). The upper bound on
the DDREF is also higher for leukemia. If the possible distorting effects of errors in
dosimetry (estimated to be roughly ±30%) are taken into account, the upper bound
estimate (95% confidence bound) for the DDREF is about 5 for leukemia and about 3
for all other cancers combined (Pierce and Vaeth 1991). However, the analysis of
Pierce and Vaeth did not consider all the potential errors in dosimetry, most notably,
the possibly large underestimation of neutrons at Hiroshima discussed above. Such an
error could have distorted the shape of the dose-response relationship for gamma rays
(Kellerer and Nekolla 1997). If neutron fluxes were to be increased by as much as
proposed by Straume (1992), then a DDREF as high as 10 might be consistent with the
LSS data (D. Preston, unpublished results).
11
-------
Studies of tumorigenesis in animals most often yield DDREFs in the range of 2
to 5 (MAS 1990). Thus, the DDREF derived from a dose-response fit to the LSS data
on solid tumors is consistent with values near, or below, the lower end of the range
derived from animal studies. There are very limited data on humans, however, which
bear directly on the question of extrapolation to low dose rates. Data on medically
irradiated cohorts indicate that fractionation of the dose has no large effect on the risk
of radiogenic thyroid cancer or breast cancer (Shore et al. 1984, Davis et al. 1989,
Howe 1992). On the other hand, the apparent absence of radiogenic lung cancers in
fluoroscopy patients receiving fractionated doses of X-rays suggest that a larger
DDREF may be applicable to the lung (Davis ef al. 1989, Howe 1992). Although the
authors claim that these findings are inconsistent with projections from the A-bomb
survivors based on a low value for the DDREF, no detailed statistical analysis is
provided from which one can infer a lower bound on the lung cancer DDREF. Such an
analysis would have to address: (1) the sampling errors in both the LSS and
fluoroscopy data; (2) the dependence of risk on age and sex; and (3) the differences in
baseline lung cancer rates between the Japanese and North American populations.
Finally, careful attention must be paid to the possible confounding influence of
tuberculosis within the fluoroscopy cohorts.
There is evidence that low dose radiation may induce or activate cellular DMA
repair mechanisms through a so-called "adaptive response," leading to speculation that
low doses may be protective against cancer (Feinendegen 1991). However, the effects
seen to date have been essentially short term; for this reason, it does not yet appear
likely that the net effect of the radiation would be beneficial (Puskin 1997). At this
point, too little is known about this adaptive response to influence estimates of risk at
low doses and dose rates. It is also theoretically possible that low dose radiation could
stimulate other protective mechanisms, e.g., programmed cell death (apoptosis). A
detailed review of possible radiation induced adaptive responses can be found in the
UNSCEAR 1994 report.
Assigning an uncertainty distribution to the DDREF requires subjective
judgement. For all cancers combined, NCRP Report No. 126 posited a piecewise
linear distribution, peaked at 2.0, and spanning the interval from 1 to 5. As a default for
most sites, and for uniform whole-body irradiation, we have adopted a distribution that
places somewhat more weight on a DDREF value close to 1 and assigns a finite
probability to DDREFs > 5. The distribution is uniform from 1 to 2, falling off
exponentially for values greater than 2. The two parts of the distribution are normalized
so that: (1) the probability density function is continuous and (2) the integrals of the
uniform and exponential portions are each 0.5. Mathematically, the probability density
for the DDREF, f(x), can then be written:
f(x) = 0.5 1 < x < 2 (2a)
= 0.5e'(x'2) x >2 (2b)
12
-------
The median (2.00) and mean (2.25) of this distribution are slightly lower than the
corresponding values based on the NCRP distribution (2.34 and 2.48). The 90%
subjective confidence interval (Cl) is 1.10-4.30, compared to 1.25-4.13 for the NCRP
distribution. The two distributions are compared in Figures 1a and 1b.
Figure 1a. DDREF uncertainty distributions.
12345
DDREF
Figure 1b. Cumulative DDREF uncertainty distributions.
13
-------
No probability is assigned here to a threshold or to a protective effect of low
dose rate radiation. Likewise, no weight is given to the possibility of a heightened
sensitivity at low doses (DDREF < 1). None of these alternatives is incompatible with
the epidemiological data, which are generally not informative about risks at very low
doses. Currently, support for these concepts from radiation biology is weak
(UNSCEAR 1993, 1994). However, newly observed effects of ionizing radiation on
cells, including the adaptive response (Olivieri etal. 1984, UNSCEAR 1994, Wolff
1996), genomic instability (Kennedy etal. 1980, Morgan etal. 1996) and the "bystander
effect" (Nagasawa and Little 1992, Deshpande etal. 1996, Lorimore etal. 1998), could
eventually lead to some fundamental revisions in the theory of radiation carcinogenesis
and estimates of risk at low doses. Our present understanding of these phenomena is
very limited, and any implications for low dose risk estimation are highly speculative
(UNSCEAR 1993, 1994; Fry etal. 1998). Consequently, the reported observations on
these effects did not influence our quantitative assessment of uncertainty.
Several reviewers of this report have recommended that, in view of what
appears to be a linear dose response in the LSS data, we should assign a finite
probability to a DDREF of exactly 1-at least for solid tumors. In our judgement,
adequate weight is already placed on DDREF values near 1. For example, 12.5% of
the total area is contained between DDREF values of 1 and 1.25: 2V2 times that for the
NCRP distribution. To further explore this issue, a Monte Carlo simulation was
performed using a modified distribution for the DDREF, which assigned a 10%
probability for a DDREF equal to 1, keeping the shape of the distribution otherwise
constant. This modification produced about a 13% increase in the upper bound
estimate of whole-body risk and about a 4% increase in the lower bound estimate (see
Section H below). Thus, the uncertainty range for the risk is not very sensitive to
inclusion of a small probability for a DDREF of 1.
It is quite likely that the DDREF varies from one cancer site to another, and for
certain sites there may be enough information to justify an alternative probability
distribution. For example, a narrower uncertainty range in the DDREF for breast
cancer may be warranted in view of the linearity of the dose response observed in
several study populations and the apparent invariance in risk with dose fractionation
(Hrubec etal. 1989, NAS 1990, Howe 1992, Tokunaga etal. 1994). A modified
uncertainty distribution may also be appropriate for leukemia, for which the dose
response in the LSS appears to be concave upward. Other sites should perhaps be
assigned different DDREF uncertainty distributions, as well. The assumption we are
making here is that, for an essentially uniform dose to all target tissues, the overall
uncertainty is dominated by the nominal distribution described above. A more careful
consideration of the DDREF uncertainty distribution may be needed for cases where
the dose is heavily concentrated in a few specific target tissues.
14
-------
G. Alpha Particle RBE
The NCRP has recently reviewed the laboratory data bearing on the issue of
RBEs for high-LET radiation (NCRP 1990). From an examination of the data on
internal emitters, the NCRP concluded that: "The effectiveness of alpha emitters is
high, relative to beta emitters, being in the range of 15 to 50 times as effective for the
induction of bone sarcomas, liver chromosome aberrations, and lung cancers. The
RBE of alpha emitters tends to increase as the dose decreases, probably mainly due
to the decreased effectiveness per .Gy of low-LET radiation at low doses and low dose
rates." Also relevant are the findings on external exposures to fission neutrons, which,
because of their comparable LET, are expected to have an RBE similar to that for alpha
particles. For neutrons, a wide range of RBE has been observed, but if one considers
only the most relevant data on tumorigenesis, the range is about 6 to 60.
EPA is generally concerned with low dose, low dose rate conditions. Under
these conditions, the low-LET risk is presumed to be reduced by a DDREF, and the
alpha particle RBE is increased by this same factor. For solid tumors other than breast,
EPA has adopted a nominal DDREF of 2 and an RBE of 20 (EPA 1994). [For the
breast, these nominal values are 1 and 10, respectively (EPA 1994).] The DDREF
adopted for solid tumors is somewhat lower than what is often observed in animal
experiments. It follows that RBEs determined from low dose extrapolation of
experimental data may be higher than what would apply to humans. Taking this
consideration into account, for solid tumor induction, we assign to the RBE an
uncertainty range of 5 to 40 (90% subjective confidence interval). [Stated another way,
the risk per unit dose for alpha radiation is estimated to be 2.5-20 times higher than
that for high acute doses of low-LET radiation.] Within this range the uncertainty is
assumed to be distributed lognormally around the geometric mean of the upper and
lower subjective confidence bounds.
Since there is a dearth of cancer sites for which there are detailed
epidemiological data relating to both high- and low-LET exposures, one cannot
generally base estimates of alpha particle RBE on human data. One apparent
exception is leukemia. The risk of leukemia induced by internally deposited radium
appears to be much lower than what one would calculate based on low-LET
epidemiological data and an RBE of 20 (NAS 1988). Possibly, the discrepancy can be
explained in terms of error in estimates of alpha particle dose to sensitive cells in the
bone marrow. For purposes of our risk assessments, we have treated this as a special
case, basing our high-LET leukemia risk estimates directly on high-LET
epidemiological data (EPA 1994).
In calculating leukemia risk from alpha irradiation, EPA employs a nominal RBE
of 1 (EPA 1994). Although experimental studies of neutron irradiated animals also
point to a low RBE (about 2 or 3) for leukemia induction (Ullrich and Preston 1987), the
15
-------
RBE of 1 was based primarily on epidemiological studies of patients injected with 232Th
(Thorotrast). In the Thorotrast studies, a clear excess of leukemia was observed,
consistent with a risk estimate of roughly 50 leukemias per 104 person-Gy (Mays et a/.
1985, MAS 1988). Numerically, this is about equal to the low-LET leukemia risk
estimate derived from the LSS, assuming a DDREF of 2 (EPA 1995), thus implying an
RBE of about 1.
Recently, some investigators have proposed a readjustment of the dosimetry in
the Thorotrast studies, which would imply an upward revision of the leukemia risk
coefficient and an RBE of 6 or 7 for alpha-particle induced leukemia (Hunacek and
Kathren 1995, RAC 1999). Based on ICRP dosimetry models, an RBE of this
magnitude would imply that the calculated leukemia risk from alpha emitters of greatest
interest - i.e., those depositing in/on the mineral phase of bone - would be comparable
to the estimated bone sarcoma risk. However, studies of radium dial painters who had
ingested 226Ra and of patients injected with 224Ra indicate that the risk of leukemia
induction by such alpha emitters is small compared to the risk of bone sarcoma (Mays
et a/. 1985, MAS 1988). These findings imply an RBE of no more than about 1 for
leukemia induction.
The alpha particle dose is quite different for the Thorotrast patients from that for
the radium exposed cohorts. The 232Th is incorporated into colloidal particles which are
suspended in the marrow rather than being deposited in the bone mineral phase.
Some of the marrow is effectively screened from alpha particles emitted from the bone
surface. The Thorotrast particles may, therefore, more effectively irradiate target cells
in the marrow, leading to a higher leukemia risk for a given average marrow dose.
In conclusion, for what are generally the most important cases of interest (i.e.,
alpha-emitting radionuclides deposited in/on the mineral bone) the risk of leukemia
appears to be small compared to the bone sarcoma risk. An effective RBE of about 1
seems to provide an upper bound on the leukemia risk. The multiplicative error
associated with the uncertainty in these cases is assigned a uniform distribution U(0,1).
We would emphasize that this reflects the uncertainty in what is basically an "effective
RBE," which factors in the nonuniform distribution of energy deposition in the bone
marrow; in no way does it imply that the target cells in the marrow are less sensitive to
high-LET radiation than to gamma rays.
Ordinarily, leukemia induction by alpha-emitters not deposited in/on bone
constitutes a relatively small fraction of the total (whole-body) risk from these
radionuclides. Nevertheless a different uncertainty distribution is appropriate. Taking
into account both the Thorotrast data and the evidence from neutron irradiated animals,
a central estimate for the RBE of 3 and a range (approximate 95% confidence interval)
from 1 to 10 seems reasonable (RAC 1999); accordingly, an RBE uncertainty
distribution LN(3,1.7) is judged to be reasonable for leukemia induction by these
radionuclides.
16
-------
Lung is another site for which there are both low-LET data (from the LSS) and
high-LET data (from radon exposed underground miners) on humans. The latter
pertain almost entirely to adult male exposures. For adult males, the two sets of data
are reasonably consistent with an RBE of roughly 10, but childhood and female
exposures among the bomb survivors appear to be associated with higher relative risks
than those for adult males (Shimizu et al. 1990). This contrasts with the BEIR IV/EPA
radon risk model, which posits no dependence of the relative risk coefficient on sex or
age at exposure. In conclusion, one cannot at this time properly assess whether or not
a single model can be used for calculating the risk of both low-LET and high-LET
(radon) exposures. Until such a model gains acceptance, one cannot unequivocally
assign a single RBE value to lung cancer induction based on the epidemiological data.
Experiments comparing beagles inhaling particulates of alpha or beta emitters
suggest an RBE of 33 to 58 for the lung (Griffith et al. 1987). However, the effect per
unit dose from the beta emitters decreased rapidly with dose in these experiments,
reflective of a high DDREF for low-LET radiation. EPA risk estimates for humans, on
the other hand, are based on a DDREF of 2. Consequently, the dog experiments-if
they are relevant to humans-are more suggestive of an overestimation of low dose,
low-LET risk than an underestimation of high-LET risk for the lung. We have,
therefore, adopted the same uncertainty distribution for RBE in the case of the lung as
for other solid tumors (i.e., a lognormal distribution with a 90% Cl from 5 to 40).
Two other sites for which there are human data on alpha-particle risk are bone
and liver, but direct human information on low-LET risks for these sites is sparse.
Consequently, for these sites, high-LET risk estimates are based directly on the human
data, whereas the corresponding low-LET risk estimates are obtained by dividing the
high-LET estimates by a nominal alpha-particle RBE of 20. In general, liver and bone
risk do not constitute a major portion of the total risk from intake of beta/photon
emitters. Thus, while the uncertainty in low-LET risk for these sites would be strongly
influenced by the uncertainty in alpha-particle RBE, these uncertainties are not
important from a practical standpoint, since they are not major contributors to the
overall uncertainty in risk from any likely low-LET exposure. [Note: Citing animal data
published by Lloyd et al. (1995), RAC (1999) assigned a 97.5% upper confidence
bound of 375 to the alpha-particle RBE for bone. Even for ingestion of a bone-seeker
like 90Sr, inclusion of this extreme upper bound RBE would only slightly perturb the
lower bound total cancer risk estimate.]
Finally, in assessing risks from short-range alpha particle radiation, attention
must be paid to possible nonuniformity in both the doses and the radiation sensitivity
within target tissues. This point has already been discussed above with respect to the
induction of leukemia by bone-seeking radionuclides. With respect to the induction of
bone cancer, too, it is necessary to evaluate the dose to the target (endosteal) cells
residing on the bone surface, which may be quite different from the average dose to the
bone (Puskin et al. 1992). Another important case is the nonuniform dose to cell
17
-------
systems within the lung delivered by radionuclides deposited in the respiratory system,
and the widely varying sensitivity of these cells to radiation. In such instances, great
care is required in extrapolating from low-LET information, where the dose is fairly
uniform within the tissue. Application of simple RBE factors derived from animal
studies without proper consideration of the dosimetric issues can easily lead to
fallacious conclusions about the magnitude of the risk and the range of uncertainty.
H. Uncertainty in Whole-Body Risk
In general, radionuclide exposures produce a nonuniform distribution of dose
within the body. The uncertainty in risk associated with the exposure depends on the
uncertainty in organ specific doses as well as the uncertainty in the risk per unit dose
for each organ irradiated. It is outside the scope of this document to assess the
uncertainty in risk estimates for specific radionuclides and exposure pathways. The
main purpose here is to provide a framework for assessing the uncertainty in organ
specific risk, for specified dose distributions. This general framework has been
applied elsewhere to assess the uncertainty in risk from ingested radon-222 in drinking
water (EPA 1993).
A commonly encountered exposure scenario is one in which there is an
approximately uniform, whole-body dose of low-LET radiation. Examples would
include: (1) external exposure to energetic x-rays or gamma rays and (2) ingestion of
Cs-137 or tritiated water, where the radionuclides are distributed fairly uniformly
throughout the body. As stated in Section II, the estimated risk from a uniform, whole-
body, low dose rate exposure is 5.75x10~2 fatal cancers Gy"1. In this section we assess
the uncertainty in this risk estimate, based on the discussion of uncertainties in organ
specific risk estimates presented above.
To quantify the overall uncertainty in the nominal estimate of risk, we treat the
various component sources of uncertainties as independent multiplicative factors, each
with its own probability distribution. For whole-body irradiation the estimate of risk
rests mainly on results from the LSS. Accordingly, this uncertainty analysis will focus
on the uncertainty in the data and model projections derived from the LSS. These
components of uncertainty include: (1) sampling variation in the LSS; (2) the model
used to project risk over time; (3) the transport of risk from the LSS to the U.S.
population; (4) errors in A-bomb dosimetry; (5) the value of the DDREF; and (6) errors
due to diagnostic misclassification.
1. Sampling variation
From Table 2, the relative standard error in the estimate of all cancers other than
leukemia is about 15%. The relative standard error for leukemia is only slightly higher;
moreover, the leukemia risk constitutes only about 10% of the risk from uniform, whole-
-------
body irradiation. Thus, it should be adequate to estimate the uncertainty due to
sampling variation based on the data for solid tumors. This source of uncertainty is
well represented by a multiplicative factor, x^ normally distributed with a mean of 1.0
and a standard deviation of 0.15.
2. Diagnostic misclassification
As outlined in Section B, the multiplicative error in the whole-body risk estimate
due to diagnostic misclassificatiorrin the A-bomb survivor study (x6) will be represented
by a normal distribution N(1.2,0.06).
3. Temporal dependence
As discussed above, some of the organ specific risk projections are dependent
on a lifetime extrapolation of very high relative risks seen among those exposed as
children. To an extent these estimates are based on poor statistics; also, there are
some theoretical reasons and empirical observations to suggest that the relative risks
will decrease over time. Hence, as discussed in Section C, the population risks may be
overestimated by about a factor of 2 or more. Conversely, for some organs, the risk
estimates are dominated by data on adult exposures, possibly leading to an
underestimate of the risks to the general population including children. Generally,
however, these organs include only ones for which the risks are fairly low, together
accounting for only 14% of the whole-body risk. Thus, if risks for these sites are
underestimated by as much as a factor of 1.5 (see Section C), the whole-body risk
would only be increased by 7%.
As a result, the error induced by temporal projection is likely to be in the
"conservative" direction. This uncertainty is characterized here by a multiplicative
factor (x2) with a probability density given by a trapezoidal function
Trpz(0.5,0.6,1.0,1.1). The probability density in this case increases linearly from zero
to a maximum as x2 increases from 0.5 to 0.6, remains constant over the interval from
0.6 to 1.0, and then decreases linearly to zero as x2 approaches 1.1.
4. Transport of risk estimates
Our analysis of this uncertainty is predicated on the assumption that the
multiplicative and NIH projections from the LSS provide upper and lower bounds on the
transport uncertainty for each cancer site. As shown in Table 4, sites differ as to which
projection is higher or lower. In general, there are little or no data to support one
projection model over the other, and the multiplicative model may be better for
projecting some organ risks, whereas the NIH model may be more suitable for others.
For some organs a correct projection may fall between the two model projections.
19
-------
TABLE 4
Multiplicative, NIH and CMC projections of specific organ risks
(deaths per 104 person-Gy)
Cancer Site
Esophagus
Stomach
Colon
Lung
Ovary
Bladder
Leukemia
Remainder
Multiplicative
9.16
12.7
185
194
26.2
30.6
54.3
185
NIH
15.7
135
62.8
52.7
8.89
19.3
50.9
146
GMC
11.7
40.7
104
98.8
14.9
23.8
55.7
149
An extreme upper (lower) bound on the "transport uncertainty" can be obtained
by choosing for each site the multiplicative or NIH projection, whichever is higher
(lower), and then summing over all sites. Less extreme estimates of the uncertainty
bounds can be obtained through a Monte Carlo procedure in which the projection for
each site is allowed to vary randomly within the limits defined by the multiplicative and
NIH projections for that site, and then summing over sites. The resulting projection will
depend on assumptions regarding the mathematical form of the uncertainty distribution
for the individual sites and possible correlations between sites (e.g., if the multiplicative
projection were known to hold for one site, this may increase the probability that it
would hold for other sites, as well).
For those sites which the GMC estimate has been adopted (i.e., esophagus,
stomach, colon, lung, ovary, bladder, leukemia, and residual), we assume that the
distribution of uncertainty is loguniform between the NIH and multiplicative projections,
and that the transport with respect to different sites is independent. A Monte Carlo
calculation then shows that the uncertainty in the whole-body risk estimate associated
with the transport of risk estimates is distributed approximately symmetrically about a
mean estimate, which is about 10% higher than the nominal estimate, and with a
standard deviation of about 12%. Accordingly, a normal distribution is assigned to this
20
-------
multiplicative uncertainty factor (x3), with a mean of 1.1 and a standard deviation of
0.12.
5. Errors in dosimetry
Based on information presented in Section E, the multiplicative uncertainty (x4)
associated with errors in dosimetry is assumed to follow a normal distribution
[N(0.84,0.11)] with mean 0.84 and standard deviation 0.11.
6. Error in the choice of DDREF
The value of the DDREF (x5) is drawn from the distribution defined by Equations
2a and 2b in Section F. The probability density is uniform for x e [1 , 2], falling off
exponentially for x>2 (see Figure 1a).
7. Calculation of uncertainty
Table 5 summarizes the sources of error considered here in estimating the
uniform whole-body risk and the assumed probability distribution for each. Treating
each of these sources of error as independent and multiplicative, but noting that the
DDREF divides rather than multiplies the risk and that a DDREF of 2 is already
incorporated into the risk estimate, a Monte Carlo calculation was carried out (see
Table 5 caption). The results showed that the uncertainty was distributed
approximately lognormally, with a median risk estimate of 4.9xlO"2 fatal cancers/Gy and
a geometric standard deviation (GSD) of about 1.66; the mean of the distribution is
5.4xlO"2Gy"1. The estimated 90% subjective confidence interval is 2.0x10~2-1.1*10~1
Thus, based on the results of the uncertainty analysis, the nominal estimate is
biased high, but only slightly. The actual risk is expected to be no more than about a
factor of 2 times higher or 3 times lower than the nominal estimate of 5.75*10~2 Gy'1.
The estimated bias mainly results from the assumed form of the temporal dependence
in our risk model: i.e., a constant relative risk. Overall, the most important uncertainties
seem to be in the temporal dependence of the risk (especially for childhood exposures)
and in the DDREF.
21
-------
TABLE 5
Distributions used to estimate the uncertainty in the risk of
low level, low-LET, whole-body irradiation
Source of Uncertainty Distribution
Sampling variation (f,) N(1.0,0.15)
Diagnostic misclassification (f2) N(1.2,0.06)
Temporal dependence (f3) Trpz(0.5,0.6,1.0,1.1)
Transport across populations (f4) N(1.1,0.12)
Errors in dosimetry (fs) N(0.84,0.095)
DDREF(f6) U(1,2):50%
EXP(>2): 50%
Notes
(1) The combined multiplicative uncertainty distribution was generated by
repeatedly sampling each of the distributions in the table and calculating
the value of (x, x2 x3 x4 xs)(2/x6)R, where the X| denote random values of the
independent variates defined by the distributions fi, and where R is the
nominal low dose, whole-body risk estimate, 5.75x10"2 fatal cancers Gy~1.
(2) For the normal distribution, the parameters in parentheses refer to the
mean and standard deviation, respectively. The trapezoidal distribution,
Trpz(a,b,c,d) rises linearly from zero to a maximum as the random variable
x, increases from a to b; remains constant for b 2; the median of the distribution is assumed to
be 2.
I. Uncertainty in Incidence Estimates
In most cases, site-specific incidence estimates reflect mortality risk
estimates divided by the cancer lethality estimate for the site in question (see
Section IV.F of EPA94). For those sites contributing most heavily to the whole-
body risk (e.g., lung and colon), the possible error in lethality is small compared to
22
-------
the uncertainty in the radiogenic cancer risk estimate for that site. One exception
is radiogenic leukemia, where the use of a lethality fraction of 99% (EPA94, Table
5) understates the cure rate achievable with modern medicine, especially for
children. As noted in Section IV.F of EPA94, this would likely produce an
overestimate of the mortality rather than an underestimate of the incidence.
Another issue relates to the estimate of skin cancer incidence. Nonfatal skin
cancers, most of which are of little clinical significance, were not included in the
estimate of incidence. If-as often occurs for external exposure to a beta
emitter-the dose is predominantly to the skin, inclusion of all these nonfatal cases
could increase the incidence estimate by up to a factor of 500 (see: EPA94,
Section IV.D). For uniform whole-body irradiation, the total cancer morbidity
estimate would be increased from 850><10~4/Gy to ISSQxIO^/Gy. A small fraction of
nonfatal radiogenic skin cancers are serious in that they require substantial
medical intervention and may result in significant residual impairment or
disfigurement. Although there appear to be no published estimates of this fraction,
it is not expected that inclusion of these serious nonfatal cases would appreciably
increase the incidence for uniform, whole-body irradiation. However, in cases
where the dose to the skin is high compared to other organs, inclusion of the
serious nonfatal cases might increase the incidence by as much as an order of
magnitude.
A comparison between Japanese A-bomb survivor incidence and mortality
data has recently been published (Ron etal. 1994). The authors conclude that:
For all solid tumors the estimated excess relative risk at 1 Sv...for
incidence...is 40% larger than the excess relative risk (ERR) based
on mortality data... For some cancer sites, the difference...is greater.
These differences reflect the greater diagnostic accuracy of the
incidence data and the lack of full representation of radiosensitive but
relatively nonfatal cancers, such as breast and thyroid, in the
mortality data.
To date, the Japanese incidence data have not been used to develop
comprehensive risk projections for other populations. It seems likely, however, that
the Japanese incidence data will increasingly serve as a basis for radiogenic
cancer risk estimates in the future.
J. Numerical Estimates of Uncertainty in Specific Organ Risks
Quantification of the uncertainties in all the site-specific cancer risk
estimates is beyond the scope of this document. However, for illustrative
purposes, we shall attempt to quantify the uncertainties in the low-LET cancer
23
-------
mortality risk estimates for two cancer sites: lung and bone marrow. These two
sites are of particular interest because they are often "critical organs": i.e., for
certain radionuclides and pathways, the highest projected organ-specific risk is
associated with one of these sites.
1. Uncertainty in lung cancer risk estimate
Sampling variation. As shown in Section A, the nominal estimate of the
ERR/Gy is 0.63 with a 90% confidence interval of 0.35 to 0.97. A simple
approximation to the distribution is given by a normal distribution with a mean of
0.66 and standard deviation 0.19. Normalizing the nominal estimate to unity, the
uncertainty is given by a multiplicative factor N(1.05,0.29).
Diagnostic misclassification. A comparison of autopsy and death
certificate diagnoses indicates that diagnostic misclassification of lung cancer
among the LSS cohort is high, especially for persons over age 75 (Ron et al.
1990). Moreover, the ERR/Sv for lung cancer incidence, as determined from
Hiroshima and Nagasaki tumor registry data between 1958-1987, is about 42%
higher than the corresponding ERR/Sv for mortality, as determined from death
certificates collected over the same time period (Ron et al. 1994). Since lung
cancer is rapidly fatal in a high percentage of cases, this would seem to indicate
that diagnostic errors are substantially perturbing the ERR estimates. The
incidence determinations are believed to be more accurate than the death
certificate information. It is therefore likely that our risk estimates, which are based
on mortality data, are biased low due to diagnostic misclassification. As a
subjective estimate of the error due to diagnostic misclassification of lung cancer,
we assign a multiplicative uncertainty factor N(1.3,0.15) to this source of error.
Temporal dependence. As discussed in Section C, an uncertainty
distribution 11(0.5,1.0) is assigned to this source.
Transport of risk estimate from LSS. As shown in Table 4, the
multiplicative and NIH model projections for radiogenic lung cancer mortality are,
respectively, 194/98.8 = 1.96 and 52.7/98.8 = 0.53 times that of the nominal
estimate. Accordingly, a multiplicative uncertainty factor LU(0.5,2.0) will be
assigned to the transport of the risk estimate from the LSS to the U.S. population
[see Section D]. This transport is thus a much larger source of uncertainty for the
lung cancer risk than for the whole-body risk [see Section H.4].
Errors in dosimetry. The analysis of uncertainty in whole-body risk
associated with errors in dosimetry incorporated four sources of uncertainty in
DS86 (see Table 3). In order to derive an uncertainty distribution for lung
dosimetry, one of the four underlying distributions, that associated with neutron
24
-------
dose, needs to be modified. For whole-body irradiation, the distribution was
triangular, with a peak at 1.10 and a range of 1.0 to 1.3. This distribution, however,
was based on dose to the colon; transmission of radiation to the lung is
considerably higher than to the colon. According to
Table 4B-1 of the BEIR V Report (MAS 1990), the transmission factors for the colon
are 0.74 and 0.19, for gamma rays and neutrons, respectively; the corresponding
values for the lung are 0.80 and 0.33.
To see the effect of transmission factors on the dosimetric uncertainty,
consider the colon dose to an arbitrary survivor. DS86 neutron doses are relatively
unimportant (MAS 1990) so we can approximate the dose as:
Dc = DCY
The corrected dose (with neutrons added) can be written:
DC' = Dcn + Dc;
where DCn is the weighted neutron dose to the colon and DCY' is the y-ray dose,
corrected for the increase in neutron flux. Neglecting the small contribution to the
Y-ray dose from n,y reactions, D^' = D^, and these equations imply:
Dc'/Dc = 1 + DCn/DCy = 1 + Xc
Similarly, for the lung, we would have:
DL7DL = 1 + DLn/DLY = 1 + XL
Using the transmission factors above, the fractional increment to the lung dose can
then be compared to that for the colon:
XL/XC= (0.33/0.19)(0.74/0.80) = 1.61
Thus, the neutron correction to the lung dose is about 61% larger than for the
colon. Scaling the uncertainty distribution based on colon dose appropriately, the
uncertainty distribution for the lung due to underestimation of neutron dose is
triangular with a range from 1 to 1.483 and peaked at 1.161.
With this modification to the distribution fs(D) in Table 5, the uncertainty
factor for lung cancer risk due to dosimetric uncertainties is found, by Monte Carlo
simulation, to be approximately normal, with a mean of 0.72 and a standard
deviation of 0.093.
25
-------
Extrapolation to low doses and dose rates. Assigning an uncertainty
distribution to the DDREF for lung cancer is problematic. On the one hand, the
dose response in the A-bomb survivor cohort shows no sign of nonlinearity
(Thompson et al. 1994). On the other hand, a study of lung cancer mortality in
Canadian tuberculosis patients receiving highly fractionated doses of radiation
(through fluoroscopy examinations) revealed no excess risk of lung cancer (Howe
1995).
The upper 95% confidence limit on the ERR/Sv in the fluoroscopy patients
was only about one-ninth the central estimate based on the A-bomb survivors and
only about one-fourth the corresponding 95% lower confidence estimate. Although
Howe concludes that the discrepancy is most likely attributable to the dose
fractionation, other factors may be important. In particular, Howe's comparison
presupposes that relative risk transports from one population to another
(multiplicative projection model). As discussed above, the NIH projection model
would reduce the projected risk for a North American population by roughly a factor
of 4. Taking this into account, the data are consistent with a DDREF of about 2. As
noted in Section F, moreover, other differences in the populations might further
contribute to the discrepancy; e.g., differences in the age distribution of the
exposed populations and possible confounding by the lung disease present in the
fluoroscopy subjects.
Thus, while the fluoroscopy data suggest a large reduction in lung cancer
risk at low dose rates, the case is not compelling. So long as the large uncertainty
in "risk transport" is factored into the analysis, it seems reasonable to adopt for
lung cancer the default DDREF uncertainty distribution, defined by Equations 2a
and 2b in Section F.
Calculation of the uncertainty in low dose lung cancer risk. Table 6
summarizes the distributions assigned to each multiplicative uncertainty factor for
lung cancer risk. Following the same Monte Carlo procedure used for whole-body
irradiation in Section H.7, we arrive at an overall 90% subjective confidence
interval of 2.0*10"3Gy1 to 2.0*10~2Gy1. The median and mean of the uncertainty
distribution are 6.7*10~3 Gy~1 and 8.3*10~3 Gy~1, respectively. Thus, it is estimated
that the actual lung cancer risk could be about a factor of 2 higher or a factor of 5
lower than the nominal estimate of 9.9*10"3 fatal cancers/Gy, a significantly wider
range than for the whole-body risk (cf. Section H.7). Primarily, the wider range
reflects the greater transport uncertainty in the case of lung cancer risk.
26
-------
TABLE 6
Distributions used to estimate uncertainty in risk of lung
cancer mortality from low dose irradiation of the lung
Source of uncertainty Distribution
Sampling variation N(1.05,0.29)
Diagnostic misclassification N(1.3,0.15)
Temporal dependence U(0.5,1.0)
Transport across populations LU(0.5,2)
Errors in dosimetry N(0.72,0.093)
DDREF U(1,2):50%
EXP(>2): 50%
2. Uncertainty in leukemia risk estimate
Sampling variation. Compared to other sites, the sampling errors for
leukemia risk are relatively small. Based on the data in Table 2, we assign an
uncertainty factor N(1.05,0.18) to sampling errors.
Diagnostic misclassification. Diagnosis of leukemia in the A-bomb
survivors has been found to be relatively accurate (Ron etal. 1994). Diagnostic
misclassification has therefore been neglected in this uncertainty analysis.
Temporal dependence. The temporal response observed in the LSS for
leukemia is complex, being dependent on city, sex, age at exposure, and type of
leukemia (Preston et al. 1994). From an examination of the trends in the data, it
would appear that relatively few excess cases are expected beyond the current
period of epidemiological follow-up (Pierce et al. 1996). In addition, since leukemia
data on the A-bomb survivors were not collected before 1950, a temporal projection
is also required for the period 0-5 years after exposure. Again, based on a
mathematical extrapolation from the observed cases, it is projected that only about
10-15% of the leukemias occurred in the A-bomb survivors during the initial 5-yr
period (Preston et al. 1994). This would suggest that the backward temporal
projection is not a major source of uncertainty. However, other studies give
conflicting results. Among irradiated cervical cancer patients, all of the excess
27
-------
leukemias were observed within the first five years post exposure (Boice et al.
1987). In the ankylosing spondylitis cohort, more than half of all excess leukemias
were observed in the first five years of follow-up (Darby et al. 1987, MAS 1990).
These results suggest that the backward extrapolation from the A-bomb survivor
data may significantly understate the risk during the first five years post exposure
(NAS 1990).
BEIR V presents four alternative temporal projection models for leukemia
(NAS 1990). The report's preferred model projects roughly twice the risk as the
other three BEIR V models or the EPA leukemia model. Computer simulations
show that this difference primarily reflects differences in projected cases within the
(5-40 y) period of epidemiologic follow-up, rather than before 5 y or after 40 y.
Thus, at least in the case of leukemia, there are appreciable temporal model
uncertainties unrelated to the projection of risk outside the period of follow-up.
In conclusion, the temporal model may significantly understate the risk
during the first five years post exposure, before data collection on the A-bomb
survivors began. There is additional uncertainty in the risk projection over
subsequent years due to uncertainties in the mathematical form of the age and
temporal response function. To account for these uncertainties, we have
subjectively assigned to the uncertainty factor for temporal projection a triangular-
shaped probability distribution, peaked at 1, and extending over the range from 0.8
to 2.
Transport uncertainty. Removing chronic lymphatic leukemias from
consideration since they are not regarded as radiogenic (NAS 1990), the overall
leukemia incidence rates for Japan and the U.S. are similar. As a consequence,
the risk estimates for the U.S. under the multiplicative and NIH projections differ
only by slightly (see Table 4). This very close agreement may be partly fortuitous,
resulting from a cancellation of differences between different types of leukemia
whose risks are being summed. A subjective uncertainty distribution LN(1.0,1.15)
has been assigned to the transport of risk from the LSS to the U.S. population.
Oosimetric errors. Like the lung, radiation transmission to the bone
marrow is considerably higher than to the colon, the dose to which was used in
calculating the dosimetric uncertainty for the whole body. The neutron
transmission factor to the bone marrow is 0.37, nearly twice that for the colon, and
0.81 fory-rays (Shimizu et al. 1989, NAS 1990). Following the procedure used for
deriving the dosimetric uncertainty distribution in the case of the lung, the
uncertainty factor associated with neutron dose is taken to be triangular with a
range from 1.0 to 1.534 and peaked at 1.178. With this replacement, the combined
uncertainty factor associated with bone marrow dosimetry is still approximately
normal, but with a mean of 0.70 and a standard deviation of 0.096.
28
-------
Extrapolation to low doses and dose rates. Pierce and Vaeth (Pierce
and Vaeth 1991) have examined data on cancer mortality among the A-bomb
survivors with the aim of determining the degree of curvature in a linear-quadratic
dose-response model that is consistent with the data. From this analysis, they
estimated the degree to which a linear extrapolation of the risk could overestimate
the risk at low doses. The magnitude of the error is expressed in terms of a "low
dose extrapolation factor" (LDEF), which is equivalent to the dose and dose-rate
effectiveness factor (DDREF). Adjusting for random errors in dosimetry of ±35%,
the maximum likelihood estimate for the LDEF was about 2.0 with respective 80,
90, and 95% one-sided confidence limits of about 3, 4.2, and 6.
Supplementary to data on the A-bomb survivors are reports of excess
leukemia in populations exposed chronically to low-LET radiation. A best estimate
of the leukemia risk in a combined cohort of nuclear workers from the U.S., UK,
and Canada is about one-half that predicted from the linear model derived from the
LSS data (Cardis et al. 1995). Although the uncertainty bounds are wide, these
results are suggestive of a DDREF of about 2. Preliminary data on leukemia
incidence among populations exposed occupationally or environmentally to
radiation from the Chelyabinsk nuclear weapon facilities in the former Soviet Union
indicate about a 3-fold reduction in risk at low dose rates (UNSCEAR 1993).
The preponderance of current data indicate that the DDREF for leukemia is
about 2 or 3, but values as high as about 6 or as low as about 1 are not excluded.
For the uncertainty analysis here, a lognormal distribution is assigned to the
DDREF, with a GM of 2.5 and a GSD of 1.5.
Calculation of uncertainty in leukemia risk. Table 7 lists the sources of
uncertainty in the leukemia risk at low doses and dose rates along with the
probability distribution assigned to each source. Again utilizing the Monte Carlo
procedure used for whole-body and lung cancer risks, an overall uncertainty
distribution for the general population leukemia mortality risk was calculated. The
90% subjective interval of the distribution is 1.7*10"3- 9.4x10"3 Gy"1; the median
and mean are 4.0x10"3Gy"1 and 4.5*10'3 Gy1, respectively. Thus, similar to the
case of uniform whole-body risk, the 90% subjective confidence interval for
leukemia risk includes values as much as 3.3 times lower or 1.7 times higher than
the nominal estimate (5.6x10"3Gy"1). The largest contributor to the uncertainty in
leukemia risk is the uncertainty in low dose extrapolation (DDREF).
29
-------
TABLE 7
Distributions used to estimate the uncertainty in leukemia
risk from low dose irradiation of the bone marrow
Source of uncertainty Distribution
Sampling variation N(1.05,0.18)
Temporal dependence T(0.8,1.0,2.0)
Transport uncertainty LN(1.0,1.15)
Dosimetric errors N(0.70,0.096)
DDREF LN(2.5,1.5)
K. Discussion
The aim of this report is to provide quantitative estimates of uncertainty in
EPA's radiogenic cancer risk estimates. Thus, we are concerned here with the
uncertainties in only one portion of a radionuclide risk assessment: the dose-
response model. For each of the cases analyzed in detail here (i.e., low-LET risks
to the whole body, to the lung, and to the bone marrow) the estimated 90%
subjective confidence interval was relatively narrow - a factor of 5 to10 from the
low to the high end of the interval. In general, however, to assess the overall
uncertainty in risk, one must also consider uncertainties in exposure and dose. For
a specific risk assessment, these latter uncertainties will often predominate.
The approach here was to treat the overall uncertainty in population risk as
a product of independent "uncertainty factors," each of which defined the possible
deviation of the nominal risk estimate from the "true" population average due to
one specific source of uncertainty. This approach greatly simplifies the analysis
and generally helps to clarify the relative importance of the various sources of
uncertainty, but there may be some resulting inaccuracies. Of particular concern in
this regard is the separation of the uncertainties due to sampling errors and those
due to age and temporal modeling. Both these types of uncertainty are magnified
for childhood exposures, which contribute a disproportionate part of the population
risk. As a consequence, treating these two sources of uncertainty as independent
may lead us to underestimate the uncertainty in the population risk. On the other
hand, the uncertainty factors pertaining to different sources may sometimes be
inversely correlated; in those cases, treating them as independent may produce
overall uncertainty ranges that are too wide.
30
-------
The greatest source of controversy in radiogenic cancer risk estimation
remains the extrapolation to low acute doses and low dose rates. Current scientific
data do not rule out the possibility that the risk per unit dose is effectively zero at
environmental exposure levels, and proponents of an effective threshold or a
protective effect of low dose radiation continue to argue their case vigorously
(Luckey 1990, Jaworowski 1995, Goldman 1996). Our judgment-based on an
examination and synthesis of information from molecular, cellular, animal, and
human studies-is that there is at this time very little support for such an effective
threshold or a protective effect, and they were excluded from consideration in
arriving at our uncertainty bounds on risk.
Another approach to the problem of estimating the uncertainty relating to low
dose rate extrapolation would be to elicit the judgment of experts as to the
probability of different DDREF values and then to perform some kind of averaging
of these results (NRC 1997). While potentially useful in bringing out a wider range
of opinions and assumptions, this method is highly resource intensive, and it is
unclear whether it would provide a "better" estimate of uncertainty than the method
used here or in NCRP Report No. 126, where a few analysts developed a
probability distribution based on a broad review of the literature and modified it in
light of comments from colleagues. Moreover, while the conclusions drawn here
reflect subjective views of the authors, they are informed by those of various expert
panels (NCRP 1997; UNSCEAR 1993, 1994; NRPB 1993, NRC 1997).
In addition, we would note that the risk estimates and uncertainty estimates
presented here are predicated on a life table population and the published U.S.
1990 Vital Statistics. Any exposed population will differ from this idealized
population. First, the age distribution will be different; e.g., the actual U.S.
population has a higher proportion of individuals in the younger age groups.
Second, the baseline cancer incidence and mortality rates, as well as life
expectancies, undergo continuous evolution, due to changes in lifestyle, medical
care, etc. As a consequence, each birth cohort will experience a different risk,
even if all cohorts are presumed to be chronically exposed at the same constant
rate over their lifetime. Obviously, the use of current baseline rates becomes
increasingly problematic as we project farther into the future. Third, baseline
cancer rates for the U.S. population, like the LSS population, must be regarded as
uncertain due to diagnostic misclassification. Such misclassification will clearly
lead to errors in projecting radiogenic cancers in an exposed U.S. population
based on a relative risk model.
Finally, we would emphasize that the risk and uncertainty estimates
contained here reflect population averages. A more specific assessment could be
carried out for each gender or age group, but information on other factors strongly
affecting risk (e.g., genetic susceptibility) may be lacking. Thus, the uncertainty in
risk for an individual may be considerably larger than for a population.
31
-------
REFERENCES
Boice JD, Jr., M Blettner, RA Kleinerman etal. Radiation dose and leukemia risk in
patients treated for cancer of the cervix. JNCI79, 1295-1311, 1987.
Cardis E, ES Gilbert, L Carpenter, et al. Effects of low doses and low dose rates of
external ionizing radiation: cancer mortality among nuclear industry workers in
three countries. Radiat Res 142, 117-132, 1995.
Darby SC, R Doll, SK Gill and PG Smith. Long-term mortality after a single
treatment course with X-rays in patients treated for ankylosing spondylitis. Br J
Cancer 55, 179-190, 1987.
Deshpande A, EH Goodwin, SM Bailey, BL Marrone and BE Lehnert. Alpha-
particle-induced sister chromatid exchange in normal human fibroblasts: evidence
for an extranuclear target. Radiat Res 145, 260-267, 1996.
EPA (Environmental Protection Agency). Uncertainty Analysis of Risks Associated
with Exposure to Radon in Drinking Water. EPA draft report, January, 1993.
EPA (Environmental Protection Agency). Estimating Radiogenic Cancer Risks. EPA
Report 402-R-93-076, Washington, DC, June, 1994.
EPA (Environmental Protection Agency). Health Risks From Low-Level
Environmental Exposure to Radionuclides. Federal Guidance Report No. 13 -
Part 1, Interim Version. EPA Report 402-R-97-014, Washington, DC, January,
1997.
Feinendegen LE. Radiation risk of tissue late effects, a net consequence of
probabilities of various cellular responses. Eur J Nucl Med 18, 740-751, 1991
Fry RJM, A Grosovsky, PC Hanawalt, RF Jostes, JB Little, WF Morgan, NL Olenick
and RL Ullrich. The impact of biology on risk assessment-workshop of the National
Research Council's Board on Radiation Effects Research. Radiat Res 150, 695-
705, 1998.
Goldman M. Cancer risk at low-level exposure. Science 271, 1821-1822, 1996.
32
-------
Goodhead DT. An assessment of the role of microdosimetry in radiobiology. Radiat
Res 91, 45-76, 1982.
Griffith WC, BB Boecker, RG Cuddihy, RA Guilmette, FF Hahn, RO McClellan, BA
Muggenburg and MB Snipes. Preliminary radiation risk estimates of carcinoma
incidence in the lung as a function of cumulative radiation dose using proportional
tumor incidence rates. In: Inhalation Toxicology Research Institute Annual Report
LMF-120, pp. 196-204, NTIS, Springfield, VA, 1987.
Howe GR. Effects of dose, dose-rate and fractionation on radiation-induced breast
and lung cancers: the Canadian fluoroscopy study. In: International Conference on
Radiation Effects and Protection, pp. 108-113, Japan Atomic Energy Institute,
Tokyo, 1992.
Hrubec Z, JD Boice Jr., RR Monson and M Rosenstein. Breast cancer after
multiple chest fluoroscopies: second follow-up of Massachusetts women with
tuberculosis. Cancer Res 49, 229-234, 1989.
Hunacek MM and RL Kathren. Alpha radiation risk coefficients for liver cancer,
bone sarcomas, and leukemia. Health Phys 68, 41-49, 1995.
Jablon S. Atomic Bomb Radiation Dose atABCC., TR 23-71, Atomic Bomb
Casualty Commission, Hiroshima, 1971.
Jaworowski Z. Stimulating effects of ionizing radiation: new issue for regulatory
policy. Regulatory Toxicol and Pharmacol 22, 172-179, 1995.
Kellerer AM and D Barclay. Age dependencies in modelling of radiation
carcinogenesis. Radiat Prot Dos 41, 273-281, 1992.
Kellerer AM and E Nekolla. Neutron versus v-ray risk estimates. Radiat Environ
Biophys 36, 73-83, 1997.
Kennedy AR, Fox M, Murphy G, and Little, JB. relationship between x-ray exposure
and malignant transformation in C3H 10T1/£ cells. Proc Natl Acad Sci USA 77,
7262-7666, 1980.
Little MP and MW Charles. Time variations in radiation-induced relative risk and
implications for population cancer risks. J Radiol ProM 1, 91-110, 1991.
Little MP, MM Hawkins, RE Shore, MW Charles and NG Hildreth. Time variations
in the risk of cancer following irradiation in childhood. Radiat Res 126, 304-316,
1991.
33
-------
Little MP, F de Vathaire, MW Charles, MM Hawkins and CR Muirhead. Variations
with time and age in the relative risks of solid cancer incidence after radiation.
exposure. J Radiol Prot 17,159-177, 1997.
Lorimore SA, MA Kahim, DA Pocock, D Papworth, DL Stevens, DT Goodhead and
EG Wright. Chromosomal instability in the descendants of unirradiated surviving
cells after alpha-particle irradiation. Proc Natl Acad Sci USA 95, 5730-5733, 1998.
Luckey TD. Radiation Hormesis. CRC Press, Boca Raton, FL, 1990.
Morgan WF, JP Day, Ml Kaplan, EM McGhee and CL Limoli. Genomic instability
induced by ionizing radiation. Radiat Res 146, 247-258, 1996.
Nagasawa H and JB Little. Induction of sister chromatid exchanges by extremely
low doses of alpha particles. Cancer Res 52, 6394-6396, 1992.
MAS (National Academy of Sciences). Health Risks of Radon and Other Internally
Deposited Alpha-Emitters (BEIRIV). National Academy Press, Washington, D.C.,
1988.
MAS (National Academy of Sciences). Health Effects of Exposure to Low Levels of
Ionizing Radiation. (BEIR V). National Academy of Sciences Press, Washington,
D.C., 1990.
NCRP (National Council on Radiation Protection and Measurements). Influence of
Dose and Its Distribution in Time on Dose-Response Relationships for Low-LET
Radiations. NCRP Report No. 64. National Council on Radiation Protection and
Measurements, Bethesda, MD, 1980.
NCRP (National Council on Radiation Protection and Measurements). The Relative
Biological Effectiveness of Radiations of Different Quality. NCRP Report No. 104.
National Council on Radiation Protection and Measurements, Bethesda, MD, 1990.
NCRP (National Council on Radiation Protection and Measurements).
Uncertainties in Risk Estimates for Fatal Cancer Induction by Low LET Radiation.
NCRP Report No. 126. National Council on Radiation Protection and
Measurements, Bethesda, MD, 1997.
NIH (National Institutes of Health). Report of the National Institutes of Health Ad
Hoc Working Group to Develop Radioepidemiological Tables. NIH Publication No.
85-2748, 1985.
34
-------
NRC (Nuclear Regulatory Commission). Probabilistic Accident Consequence
Uncertainty Analysis. Vol. 1. Late Health Effects Uncertainty Assessment.
NUREG/CR-6555, U.S. Nuclear Regulatory Commission, Washington, DC, 1997.
NRPB (National Radiological Protection Board). Estimates of late radiation risks to
the UK population. Documents of the NRPB 4(4). National Radiological Protection
Board, Chilton, Didcot, UK 1993.
Olivier! G, J Bodycote and S. Wolff. Adaptive response of human lymphocytes to
low concentrations of radioactive thymidine. Science 223, 594-597, 1984.
Pierce DA, DO Stram and M Vaeth. Allowing for random errors in radiation
exposure estimates for the atomic bomb survivor data. Radiat Res 123, 275-284,
1990.
Pierce DA and M Vaeth. The shape of the cancer mortality dose-response curve for
the A-bomb survivors. Radiat Res 126, 36-42, 1991.
Pierce DA, Y Shimizu, DL Preston, M Vaeth and K Mabuchi. Studies of the
mortality of atomic bomb survivors. Report 12, Part I. Cancer: 1950-1990. Radiat
Res 146, 1-27, 1996.
Preston DL and DA Pierce. The effect of changes in dosimetry on cancer mortality
risk estimates in the atomic bomb survivors. Radiat Res 114, 437-466, 1988.
Preston DL, D Pierce and M Vaeth. Neutrons and radiation risk: a commentary.
RERF Update 4(4), 5, 1993.
Preston DL, S Kusumi, M Tomonaga, S Izumu, E Ron, A Kuramoto, N Kamada,
H Dohy, T. Matsui, H Nonaka, DE Thompson, M Soda and K Mabuchi. Cancer
incidence in atomic bomb survivors. Part III: Leukemia, lymphoma and multiple
myeloma, 1950-1987. Radiat Res 137, S68-S97, 1994.
Puskin JS. An analysis of the uncertainties in estimates of radon-induced lung
cancer. Risk Analysis 12, 277-285, 1992.
Puskin JS. Are low doses of radiation protective? In: Health Effects of Low Dose
Radiation. Challenges of the 21st Century, pp. 211-213, British Nuclear Energy
Society, London, 1997.
Puskin JS, NS Nelson and CB Nelson. Bone cancer risk estimates. Health Phys 63,
579-580, 1992.
35
-------
RAC (Radiological Assessments Corporation). Assessing Risks of Exposure to
Plutonium. Part of Task 3: Independent Analysis of Exposure, Dose and Health
Risk to Offsite Individuals. RAC Report No. 5 CDPHE-RFP-1999-Final, 1999.
Ron E, R Carter, S Jablon. Agreement between death certificate and autopsy
diagnoses among atomic bomb survivors. Epidemiology 5, 48-56 (1994).
Ron E, DL Preston, K Mabuchi, DE Thompson and M Soda. Cancer incidence in
atomic bomb survivors. Part IV: Comparison of cancer incidence and mortality.
RadiatRes 137, 98-112 (1994).
Shimizu Y, H Kato, WJ Schull, DL Preston, S Fujita and DA Pierce. Studies of the
mortality of A-bomb survivors: 9. Mortality, 1950-1985: Comparison of risk
coefficients for site-specific cancer mortality based on the DS86 and T65DR
shielded kerma and organ doses. Radiat Res 118, 502-524, 1989.
Shimizu Y, H Kato and WJ Schull. Studies of the mortality of A-bomb survivors:
9. Mortality, 1950-1985: Part 2. Cancer mortality based on the revised doses
(DS86). RadiatRes 121, 120-141, 1990.
Sinclair WK. Science, Radiation Protection and the NCRP. NCRP L.S. Taylor
Lecture No. 17. National Council on Radiation Protection and Measurements,
Bethesda, MD, 1993.
Sposto R, DL Preston, Y Shimizu and K Mabuchi. The effect of diagnostic
misclassification on non-cancer and cancer mortality dose response in A-bomb
survivors. Biometrics 48, 605-617, 1992.
Straume T, SD Egbert, WA Woolson, RC Finkel, PW Kubik, HE Gove, PSharma
and
M Hoshi. Neutron discrepancies in the DS86 Hiroshima dosimetry system. Health
Phys 63, 421-426, 1992.
Tokunaga M, CE Land, S Tukoka, I Nishimori, M Soda and S Akiba. Incidence of
female breast cancer among atomic bomb survivors, 1950-1985. Radiat Res 138,
209-223, 1994.
Ullrich RL and RJ Preston. Myeloid leukemia in male RFM mice following
irradiation with fission spectrum neutrons or y rays. Radiat Res 109, 165-170,
1987.
36
-------
UNSCEAR (United Nations Scientific Committee on the Effects of Atomic
Radiation). Sources and Effects of Ionizing Radiation. 1993 Report to the General
Assembly, with Scientific Annexes. United Nations, New York, 1993.
UNSCEAR (United Nations Scientific Committee on the Effects of Atomic
Radiation). Sources and Effects of Ionizing Radiation. 1994 Report to the General
Assembly, with Scientific Annexes. United Nations, New York, 1994.
Wolff S. Aspects of the adaptive response to very low doses of radiation and other
agents. Mutat Res 358, 135-142, 1996.
37
------- |