—h o > So on 9 © Cumulative fraction 20, and was assumed to pertain| to every individual in the modeled exposed population. The local food items appearing in Talfle 20 were divided into three types (and thell indicated corresponding sampling periods werelassumed): pork-related items (n, = 12 y1), chicHen-related o o rh items (nm, = 52 y!), and other items 0.5 1 1.5 (nz = 182.5 y"1). 2 Model-uncertainty (i.e., missppcification Meandaily intake (Bq/kg-d) Figure 16. Sample distribution of interindividual variability in daily intake of 137Cs per unit body weight based on survey data for 34 adult Ujelang females (bold), fit to a lognormal distribution (light) with mean = 0.447 Bq kg! d-1 and a geom. stand. dev. = 2.274. Watson tests (Stephens, 1970; Pearson and Hartley, 1972). We used this lognormal distribution as the basis of our model of variability in (R)= 365{Ry) for a hypothetical Rongelap population of arbitrary size N. The distribution has a corresponding CV with respect to modeled variability equal to gp = 0.9821. Uncertainty due to random dietary sampling associated with daily 137Cs intake for any given individual about that individual’s mean daily level (presumed constant for each individual) was estimated under the assumptions stated above that food imports are available and that local foods of type j are randomly and independently sampied nj; times per year from among Rongelap sources, using LLNL-model-diet assumptions discussed previously, along with the information summarized in Table 20 about predicted amounts and measured inter-sample variability of 137Cs in different food items local to Rongelap. For this analysis, the activities associated with the items listed in this table (accounting for ~99% of 137Cs intake associated with local foods) were scaled to correspond to an assumption that these items comprise 100% of the local-food_diet. Each corresponding CV, YRy = OR, / (Ry), with respect to presumed dietary sampling error was assumed to be the measured value appearing in column 6 of Table error) was estimated directly from] the data shown in Figure 3 relating LLNL thodel-diet predictions assuming imported foods are available, and corresponding BNL measurements of whole-body dose amongdifferent famples of Marshallese people tested during the period 1977-1983. The mean of the six megsured- to predicted-burden ratios shown is 125 + 0.37 (differing insignificantly from 1, p > P).16 by Ttest). Based on these data, an uncerta[nty-CV of ~40% was assumed, and model unceftainty for the LLNL model diet assuming impofted foods are available, was characteriz@d as a corresponding lognormally distributefi factor B with expectation 1 and SD, = 1.47. Predicted population risk I (her@ taken as the numberof fallout-induced cancey fatalities) necessarily depends on the size, and age distribution of the population involved in any Rongelap resettlement. To reasonably estimate _I, it was assumed that N = 500 In a 1995 resettlement, wherein 40% of this population would be exposed for 70 y (i.e., be present upon birth) and the remainder (of adultB of 40-y average age) for 30 y. Calculation of J was by the method of Bogen and Spear (1987)] treating [ as compound-Poisson-distributed [with an uncertain parameter (population-average dose), here approximated as 500W D(Lifetim®), where W is an uncertain risk-per-unit dose dbnversion - factor and D(Lifetime) was assumed ]to be the weighted functional (not stochastic) qverage of D(70) and D(30) using 0.4 and 0% as the respective weights. For this purpose, (70) was taken to be 1.63 D(30) based on the corresponding LLNL/ICRP model predictions (Table 1). Based on the BEIR V (NRC, 1990) predictioh of total cancer (leukemia + nonleukemia) fatplities for 47