DOI 10.1007/s00411-012-0441-x
CONTROVERSIAL ISSUE
Abstract Radiation-related risks of cancer can be transported from one population to another population at risk, for
the purpose of calculating lifetime risks from radiation
exposure. Transfer via excess relative risks (ERR) or excess
absolute risks (EAR) or a mixture of both (i.e., from the life
span study (LSS) of Japanese atomic bomb survivors) has
been done in the past based on qualitative weighting. Consequently, the values of the weights applied and the method
of application of the weights (i.e., as additive or geometric
weighted means) have varied both between reports produced
at different times by the same regulatory body and also
between reports produced at similar times by different regulatory bodies. Since the gender and age patterns are often
markedly different between EAR and ERR models, it is
useful to have an evidence-based method for determining the
relative goodness of fit of such models to the data. This paper
identifies a method, using Akaike model weights, which
could aid expert judgment and be applied to help to achieve
consistency of approach and quantitative evidence-based
results in future health risk assessments. The results of
applying this method to recent LSS cancer incidence models
L. Walsh (&)
Federal Office for Radiation Protection, Department of Radiation
Protection and Health, Ingolstadter Landstr. 1,
85764 Oberschleissheim, Germany
e-mail: lwalsh@bfs.de
L. Walsh
The Faculty of Medical and Human Sciences,
University of Manchester, Manchester, UK
U. Schneider
Vetsuisse Faculty, University of Zurich, Zurich, Switzerland
U. Schneider
Radiotherapy Hirslanden AG, Aarau, Switzerland
123
123
Zamax
MD; e; aSa=Se da
eL
2a
M D; e; a ERR D; e; a ma;
2b
2c
j1
123
Model
type
Deviance
Number of
optimized
parameters
AIC
Weights:
1 - w1
(ERR)
w1
(EAR)
All solid
ERR
14,741.0
14
14,769.0
0.98
EAR
14,748.4
14
14,776.4
0.02
ERR
3,048.7
12
3,072.7
0.85
EAR
3,052.2
12
3,076.2
0.15
Thyroid
Results
The results of applying this technique to the sites considered here using the original models of Preston et al. (2007),
from the computer files on the RERF website, are shown in
Table 1. It can be seen, from the last column of Table 1,
that the relative EAR weighting, (w1), by cancer incidence
site is zero for breast, 0.11 for colon, 0.12 for all solid and
thyroid, 0.19 for lung, 0.22 for bladder, 0.25 for liver and
0.79 for stomach.
Table 2 shows the results of applying this technique to
the sites considered here with the optimized models where
the fit parameters that corresponded to p [ 0.1 in the original Preston et al. (2007) models were removed and the
models re-optimized. The weights obtained with the choice
of p [ 0.1 are very similar to those obtained if p [ 0.05 is
Model
type
Deviance
Number of
optimized
parameters
AIC
Weights:
1 - w1
(ERR) w1
(EAR)
All solid/
lss07solmod.log
Thyroid/
lss07siteahs.log
Fem. Breast/
lss07sitemod.log
Stomach/
lss07sitemod.log
Colon/
lss07sitemod.log
Liver/
lss07sitemod.log
Lung/
lss07sitemod.log
Bladder/
Lss07sitemod.log
Leukemia/
DS02can.loga
ERR
EAR
ERR
EAR
ERR
EAR
ERR
EAR
ERR
EAR
ERR
EAR
ERR
EAR
ERR
EAR
ERR
EAR
14736.0
14739.9
3038.0
3042.0
3295.2
3307.1
8943.6
8941.0
4561.2
4565.3
4795.2
4797.4
5326.6
5329.5
2418.1
2420.6
2258.7
2254.8
19
19
20
20
13
13
19
19
19
19
19
19
19
19
19
19
22
22
14774.0
14777.9
3078.0
3082.0
3321.2
3333.1
8981.6
8979.0
4599.2
4603.3
4833.2
4835.4
5364.6
5367.5
2456.1
2458.6
2302.7
2298.8
0.88
0.12
0.88
0.12
1
0
0.21
0.79
0.89
0.11
0.75
0.25
0.81
0.19
0.78
0.22
0.12
0.88
In this table w1 and 1 - w1 correspond to the weights in Eq. (2c). The named
files, from which the deviance and number of fit parameters were taken, are
available from www.rerf.or.jp
a
Note for leukemia: only an EAR model was published by Preston et al.
(2004), the deviance value for the ERR model was calculated in Walsh and
Kaiser (2011)
123
Fem.
Breast
Stomach
Colon
Liver
Lung
Bladder
ERR
3,300.0
3,314.0
EAR
ERR
3,312.3
8,949.8
8
13
3,328.3
8,975.8
0
0.07
EAR
8,944.6
13
8,970.6
0.93
ERR
4,562.5
13
4,588.5
1.0
EAR
4,576.9
13
4,602.9
0.0
ERR
4,800.7
10
4,820.7
0.92
EAR
4,803.6
11
4,825.6
0.08
ERR
5,330.4
13
5,356.4
0.97
EAR
5,339.6
12
5,363.6
0.03
ERR
2,423.6
11
2,445.6
0.82
EAR
2,426.6
11
2,448.6
0.18
Leukemia
ERR
2,670.9
10
2,690.9
1.0
Kaiser
and
Walsh
(2012)
EAR
2,677.7
12
2,701.7
Model
type
Central risk
estimate (1SE)
Preston et al.
(2007)
Central risk
estimate (1SE)
with optimized
model
All solid/
ERR
0.467 0.044
0.461 0.043
51.85 4.97
lss07solmod.log
EAR
51.63 4.98
Thyroid/
ERR
0.58 0.26
0.67 0.26
lss07siteahs.log
EAR
1.23 0.51
1.37 0.46
Fem. Breast/
ERR
0.88 0.21
0.86 0.20
lss07sitemod.log
EAR
9.26 1.58
9.45 1.57
Stomach/
ERR
0.336 0.080
0.340 0.074
lss07sitemod.log
EAR
9.52 2.44
9.97 2.08
Colon/
lss07sitemod.log
ERR
EAR
0.53 0.16
7.95 2.15
0.57 0.15
8.80 2.13
Liver/
ERR
0.30 0.13
0.41 0.12
lss07sitemod.log
EAR
4.23 1.66
3.77 1.33
Lung/
ERR
0.81 0.16
0.90 0.15
lss07sitemod.log
EAR
7.55 1.68
8.71 1.35
Bladder/
ERR
1.24 0.41
1.15 0.37
Lss07sitemod.log
EAR
3.23 1.16
2.45 0.75
The ERR is given per unit dose (1 Gy) and the EAR is given as the
number of cases per 10,000 personyears per Gy
Discussion
In radiation-related cancer risk assessment for a subpopulation at risk, one is often required to transfer the risk
obtained from the LSS of atomic bomb survivors to the
actual subpopulation at risk. Due to a process of decision
making based on expert judgement, the numerical values of
weights to apply to ERR and EAR models, when used in
the calculation of lifetime cancer risks, vary both between
reports produced at different times by the same body and
also between reports produced at similar times by different
bodies.
In transporting risk estimates from Japan to the U.S.A.,
BEIR V (1990) assumed a multiplicative model. In contrast
to this, BEIR VII/phase2 (2006) applied a weight, wB7 of
0.7 for the estimate obtained using ERR transport for sites
other than breast, thyroid, and lung, and a complementary
weight of 0.3 for the estimate obtained using EAR transport. This choice was justified in BEIR VII/phase2
(Chapter 10), by acknowledging that there is somewhat
greater support for relative risk than for absolute risk
transport. However, the BEIR VII/phase 2 (2006) weighting was done on a logarithmic scale. The LAR values were
calculated separately based on preferred EAR and ERR
models and then combined using a weighted geometric
(1-wB7)
mean, whereby LARB7 = LARwB7
. The
ERR LAREAR
BEIR VII report acknowledges that the choice of wB7
values clearly involves subjective judgment. This geometric mean (GM) approach is not consistent with Eq. (4)
and the current literature on MMI (some of which is cited
in the materials and methods section of this paper).
The EPA (1994) report also adopted a GM approach
stating that this reflects a judgment regarding the distribution of uncertainty associated with the transportation of
risk. However, the later EPA (2011) report (see also
Pawel and Puskin 2012) made two points against the
weighted GM approach stating, First, it is difficult to
123
123
123
123
Number of
incident thyroid
cancers
w1
(EAR)
1 - w1
(ERR)
Age at
exposure
471
0.15
0.85
All ages
169
0.32
0.68
C15 years
56
0.11
0.89
\15 years
309
1.0
\15 years
LSS,
FU:19581987
LSS,
FU:19581987
Chicago tonsils
(MRH)
60
1.0
\15 years
38
0.88
0.12
\15 years
0.5
0.5
B18 years
232
1.0
B18 years
232
1.0
B18 years
Chernobyl,
Model 3
(dose*age in
1986)
232
1.0
B18 years
232
1.0a
B18 years
Tinea capitis
Thymus
Chernobyl (Jacob
et al. 2006)
L
Chernobyl,
Model 1
1089
(dose)
L
Chernobyl,
Model 2
(dose*gender)
Chernobyl,
Model 4
(dose*calendar
year)
The studies marked with a superscript R are cited in and taken from
the pooled study by Ron et al. (1995), where the goodness of fit
parameters required for the calculation of the weights presented here
have been taken from Table 6 of Ron et al. (1995). The models
marked with a superscript L are taken from the study by Likhtarov
et al. (2006), where the goodness of fit parameters required for the
calculation of the weights presented here have been taken from
Tables 6 and 7 of Likhtarov et al. (2006) for models 14 of that study
which are for a baseline plus linear doseresponse (model 1), whereby
the interaction of dose with either gender (model 2), age in 1986
(model 3) or calendar year period (model 4) was also given. Followup (FU) period for the LSS is given
a
Conclusion
In the transport of radiation risks from one population (i.e.,
the LSS of atomic bomb survivors) to another population at
risk, transfer via ERR or EAR or a mixture of both has
been performed in the past in a qualitative and inconsistent
way. This paper identifies an approach that could aid expert
judgment and could be applied, with the cautions given, to
help achieve consistency of approach and evidence-based
results and therefore contribute to future health risk
assessments.
However, it is important to state that definitive conclusions, regarding the appropriate method for transporting
cancer risks, are limited by a lack of knowledge in several
areas. Such areas include, but are not limited to, unknown
factors and uncertainties in biological mechanisms and
genetic and environmental risk factors for carcinogenesis,
uncertainties in radiation dosimetry and insufficient statistical power and/or incomplete follow-up in data from
radio-epidemiological studies. It is also particularly
important to acknowledge that the generalization and
interpretation of radiation effect estimates based on the
LSS cancer data, when projected to other populations, are
particularly uncertain for cancer sites where considerable
differences exist between site-specific baseline rates in the
LSS and the other populations of interest.
Acknowledgments Dr. Walsh would like to thank Prof. Richard
Wakeford (University of Manchester, U.K.) for drawing her attention
to the current situation regarding the subjective nature of the
weighting of ERR and EAR models for the calculation of lifetime
risks of cancer from radiation exposure. The authors would also like
to thank Prof. Donald A. Pierce, Dr. Elke A. Nekolla, Dr Charles
Land, Dr. Roy Shore, Dr Michaela Kreuzer, Dr. Jan C. Kaiser and
Dr. Peter Jacob for useful discussions. They would also like to thank
two anonymous reviewers for their careful considerations and for
providing many interesting points for the discussion section. This
work was partly supported by the seventh framework program of the
European Union, FP-7-EU-ANDANTE (Multidisciplinary evaluation
of the cancer risk from neutrons relative to photons using stem cells
and the analysis of second malignant neoplasms following paediatric
radiation therapy). This work makes use of the data obtained from the
Radiation Effects Research Foundation (RERF) in Hiroshima, Japan.
RERF is a private foundation funded equally by the Japanese Ministry
of Health and Welfare and the US Department of Energy through the
US National Academy of Sciences. The conclusions in this work are
those of the authors and do not necessarily reflect the scientific
judgement of RERF or its funding agencies.
References
Akaike H (1973) Information theory and an extension of the
maximum likelihood principle. In: Petrov BN, Caski F (eds)
Proceedings of the second international symposium on information theory. Akademiai Kiado, Budapest, pp 267281
Akaike H (1974) A new look at the statistical model identification.
IEEE Trans Autom Control 19(6):716723
123
123
123