thyroid examiner. Approximately 95% of this 173-person cohort complied with these examinations. Excellent agree- Risk Assessment The absolute risk coefficient for thyroid neoplasia is expressed as the number of excess nodules/thyroid dose/ ment was obtained between the two examiners (87% observed agreement, excess nodules are the observed minus expected nodules and the thyroid dose is expressed in grays (1 Gy =100 rad). The most recent estimates for the mean RESULTS Demographic Characteristics of Cohort A mean of 64% of the populations of years at risk/1 million persons, where dose of radiation to the thyroid giand in Marshall Islanders are 21 Gy (2100 rad) for Rongelap Islanders and 2.80 Gy (280 rad) for Utrik Islanders.” To calculate a single absolute risk coefficient for both of these populations, previous studies used the following information: a mean thyroid dose of 8 Gy (800 rad), the mean number of years at risk (18), the observed number of nodules (46), and the expected number ofnodules (16) for the combined population of 251 Rongelap and Utrik Islanders. The calculation for the expected number of nodules was based on a prevalence of 6.3% for atolls assumed unexposed to fallout.” We determined a new value for the prevalence of nodules in unexposed Marshall Islanders. To calculate a new absolute risk coefficient, we used the expected number of nodules determined with this new prevalence value as well as the above information concerning mean dose, mean years at risk, and observed nodules for the original 251 Rongelap and Utrik Islanders. Internal Validity Since all thyroid examinations were performed by a single investigator (T.E.H.), it was important to validate these observations. A substudy was designed that compared, in a masked fashion, results of the author's physical ex- amination of the thyroid gland with results of the physical examination by an expert in thyroid disease. A group of 173 individuals whom the author had examined during the previous two years was asked to participate in this study. Approximately 50% ofthese individuals had previously had normal thyroid examination results and were randomly selected from northern and southern atolls. The remaining 50% had nodular thyroid abnormalities. Each of the 173 people was examined separately by an experienced thyroid examiner from the University of Washington, Seattle. The second examiner had no prior knowl- edge of the author's previous examina- JAMA, Aug 7, 1987—Vol 258, No. 5 kappa = .80). the 13 primary atolls was screened, with a range of 33% to 96% (Table 2) As discussed in the “Methods” section, selected screening examinations were performed on Majuro and Kwajalein atolls to find those individuals who had lived on any of the 13 primary atolls at the time of the 1954 BRAVOtest. Because nearly a third of Kwajalein Atoll was screened,it was added to the other 13 primary atolls for the subsequent analyses, making a total of 14 study atolls. Prevalence of Thyroid Nodularity Of the 7266 persons screened, 2273 were alive at the time of the BRAVO test and were residing on one of the 14 study atolls on March 1, 1954 (Table 3). Exposure to the short-lived radioio- dines “J, *I, “1, and “I was therefore possible in this group. Since these isotopes have half-lives of eight days or less, exposure to radioiodines from the BRAVOtest fallout was not possible in persons born after 1954. The numbers of people with solitary thyroid nodules (mean estimated size, 2.1 em), previous thyroidectomyfor a thyroid nodule, total thyroid nodules, and the prevalence of thyroid nodules for the reconstructed 1954 population appear in Table 3. For the 12 atolls previously thought unexposed to fallout radiation, the prevalence of nodules ranged from 0.9% to 10.6% (Fig 2). If these atolls were not exposed to radioiodines from the BRAVOtest, we would expect, in the absence of other risk factors for thyroid nodularity, to see the sameprevalenceof thyroid nodulesin all the atolls. To test this hypothesis, we performed a ,* analysis. The results reject the null hypothesis that no difference exists in the prevalence of thyroid nodules among these 12 atolls O¢ = 23.45, df=11, P<.025). Predictors of Risk for Thyroid Neoplasia To better understand the wide variation in rates of thyroid nodules, we thyroid dose estimates fdr people living on these 12 atolls are lacking, the dis- tance of each atoll from the Bikini test site was selected as a proxy for the dose of radioiodine received by the thyroid gland. Weighted linear regression using the age-adjusted prevalence of nodules by atoll of residence in 1954 as the dependent variable shows a highly sig- nificant inverse linear relationship with distance from Bikini (r= — .65, P<.002) (Fig 3). Although northern atolls used in previous studies as a source for unex- posed controls were found to have a prevalence ofthyroid nodules of6.3%,"” the prevalence of nodules in our study continues to decrease to less than 1% as the distance from the site ofthe BRAVO test increases. We believe a better esti- mate for the prevalence of thyroid nodules in unexposed . Marshallese to be 2.45%, the mean prevalence of the two southernmostatolls. To examine risk at the level of the individual, we used logistic regression analysis, in which the presence or absence of a thyroid nodule was the de- pendent variable. Not only distance but also age and sex, 6 (the angle from 0° latitude), and the product of 6 and distance were all significant contributors to the logistic model (Table 4). The addition of inverse distance terms or higher order polynomial distance terms was not significant. The odds ratios obtained from the regression coefficients show that the probability of a thyroid nodule develop- ing in a female is 3.7 times higher than that in males (Table 4), a finding consistent with those of other studies of thyroid exposure.“ The oddsratio for distance is 0.33 per 100 miles from the test site, and for 6, 0.59 for every 10°. In other words, the probability of a nodule decreases ap- proximately threefold for every 100 miles farther from Bikini and twofold for every 10° going east to west in a clockwise direction. Figure 4 shows the fitted logistic model for males and females, given mean age, with the actual prevalence data plotted. Again, as seen with linear regression, the probability decreases as the distance from Bikini increases. To betterillustrate the interaction of distance and 98, we developed a set of probability contours on the map of the Marshall Islands using the logistic model with all five variables. We set the variable sex equal to females and the variable age equal to the mean age of females. For each of seven fixed probabilities between .5 and .01, the distance was calculated for possible values of 8. The valuesof6 selected were between 0° Thyroid Neoplasia—-Hamilton et al > 633 — — —_—— (masked to his previous examination) when there was disagreement between his results and those of the visiting performed multivariate analysis. Since — were born, and their residence history for their first five years was noted. tions. In addition, Dr Hamilton repeated examination of any individual - - her 1954 residence in their native language. Individuals who were too young to remember the 1954 thermonuclear BRAVO test were asked where they