Projected fete cancer nsks @C E Lano er av

from seven through 120 of Byg(a), 1€, 2[Le x

Byi(@)] In thecase of thyroid cancer, that sum 1s 1 177%

(the value for the US SEERpopulation) trmes the ratio
of the ASW rate per 100,000 for native Hawanans to that

for the US SEER population

1177% X110X757'=1 726%
Supposmg an exposure at age 12 y and, therefore,
survival until that age, the appropriate US life table for
calculatmg lifetime excess risk 1s obtamed from Fig 6 by
setting to one the probabilites of survival to ages 1
through 12 y and dividing each of the Fig 6 hfe table
probabilities of survival to ages 13, 14, etc , by the Fig
6 probability of survival to age 12 y Then, using the
revised fe table survival probabilties for subsequent
ages, L(13), L(14), etc , the hfetme excess rate associ-

ated with exposure at age 12 y 1s calculated as the sum
over a = 13 through 110,

ass

[L(@) X Byala) X ERRyq(12, @)]

Monte Carlo methods use pseudo-random numbers
to generate realizations from eachof the assumed uncertamty distributions descnibmg particular uncertam components The randomly sampled values are mtroduced m
the excess risk equations, and realizations of the excess

numberof cancers are produced The collection of values
for the excess numberof cancers, obtamed by repeating
the process for many iterations, 1s analyzed to estrmate
the mean, median, and uncertainty interval for the excess

number of cancers The uncertamty distributions of
several mput parameters used in the present analysis
(eg, DDREF) are based partly on expert judgment
regarding the appropriateness of the available data about
that parameterfor the radiation exposures in the Marshall
Islands, rather than strictly on statistical analysis of those
data, and therefore the term “uncertamty mterval” 1s used

mstead of “confidence interval” which mvolves only
staustical uncertamty Given a sufficient number of
iterations, Monte Carlo methods are accurate and, com-

d4)

Here, the notation ERRyq(12, 2) 1s required because of

the tme-dependent, uncertam latent period discussed
above under “Latent period ” In fact, the notation above

1s somewhat overly simplified because radiation dose
varied by calendar year m each location, the above
formulation should be understood as corresponding to a
given calendar year as well
In contrast to the BEIR VII dose-response models
for leukema, stomach cancer, colon cancer, and the

group of solid cancers other than thyroid and nonmelanoma skm cancer, ERR for thyroid cancer does not

depend upon attamed age (Table 3) It does, however,
depend uponlatent period, mcreasing from zero withm the
first 2 y after exposure to tts full value 8 y after exposure A

reasonable rough calculation, used here for illustration
purposes, of hifetrme risk (excess absolute sk, or EAR)

associated with a 004 Gy thyroid dose at age 12 y 1s
obtamed by muloplymg ERR,q(12, 20), whichis distributed
as lognormal with GM = 0 0805 and GSD = 3 40 (mean =
0170), by the ferme baseline risk at age 16 y, which 1s
1172% The product 1s lognormal with GM = 000094,
GSD = 340, and mean = 00020

Computational approach
Each factor used m the calculation of the excess
number of cancers has an associated uncertamty, mclud-

img radiation doses, parameter values of dose-response
models, DDREF values and other adjustments of dose-

related nsk The uncertainty of each component was
described using probabihty distribution functions (e g , in
Table 3), and Monte Carlo methods were used to prop-

agate these uncertamties

209

pared to first-order analytical methods hike those used by
the BEIR VII committee, have defimte practical advan-

tages for handlmg any magmtude of uncertainty, distnbutions of any shape, and for dealmg with large numbers

of correlated, uncertain parameters
For this report, 200 estimates of radiation doses

from external sources and 200 from mternal sources were
generated for each of 25 population groups of Marshallese [including a Rongelap control group, see BNL (1958)
for a discussion of control subjects], 24 atolls and islands

where exposure took place, 23 calendar years of exposure, 100 possible exposure ages (treating anyone exposed at ages 100 y or older as havmg been exposed at
age 99 y), and 5 target organs or groups of organs The
details of the populations, atolls and islands, and exposure years are given m Simon et al 2010a,it should be
noted that cancer risks are only estrmated for the Marshallese population, and, therefore, exclude the US

military weather observers who were exposed at
Rongertk External and mternal doses were assumed to
be perfectly correlated within each combinationof place
and year of exposure and were assumed to be uncorrelated between years and between atolls Similarly, 200
realizations were generated for the risk per umt dose for
each possible combmation of sex, exposure age and
subsequent attamed age, and cancer type The Monte
Carlo estrmated doses and risks per umt dose were
combed together with the number of mdividuals in
each exposure age group, each sex, and eachatoll to
obtain two hundred estimates of the predicted numberof
cancers, from which means, medians, and 90% uncer-

tamty mmtervals were generated for each cancer type and
selected population The estrmated numbers of cancers
were then summedto obtam totals for desired groups of

Select target paragraph3