Mortality experience was evaluated for the approximately 40,000 U.S. Navy personnel who participated in Operation CROSSROADS, a 1946 atmospheric nuclear test series that took place in the Bikini Atoll in the Marshall Islands. To judge whether that mortality experience was influenced by CROSSROADS participation, those personnel were compared to a control group assembled to be similar to the participants in all ways (age, paygrade, military service, time of service, location of service) possible except for Operation CROSSROADS participation.

A roster of CROSSROADS participants was assembled and provided to the Medical Follow-up Agency (MFUA) by the Nuclear Test Personnel Review (NTPR) program of the Defense Nuclear Agency.¹ A validation study by MFUA examining other sources of information regarding participant status found that the final roster captured between 93 and 99 percent of the military personnel who participated in Operation CROSSROADS. The mortality data gathered from Department of Veterans Affairs (VA) records were validated by sample comparisons with other national data sources. By the study cut-off date, 31 December 1992, 31.3 percent of the participants and 30.8 percent of the comparison cohort were known to have died. Cause of death was available for 86.3 percent of the participants and 89.3 percent of the controls.

Adjusting for remaining differences between the cohorts in distributions of age and paygrade, we compared, using proportional hazards analysis, the survival times of the two groups. Because available dosimetry data were not considered suitable for epidemiologic analysis, we based this study on exposure surrogate groups. We looked at three principal causes of mortality: all-cause, all-cancer, and leukemia, hypothesizing that increases in the latter two could result from radiation exposure. For descriptive purposes we also present comparisons between participants and the comparison group for 44 other disease categories. Findings stated in this report follow.

• Among Navy personnel, the primary analysis group for this study, we found that participants at the CROSSROADS nuclear test experienced higher mortality than a comparable group of nonparticipating military controls. The increase in all-cause mortality was 4.6 percent (relative risk [RR] = 1.046, 95% confidence interval, 1.02$1$2$3.074) and was statistically significant (p < 0.001).² For malignancies, the elevation of mortality was lower—RR = 1.014 (0.9$1$2$3.068)—and was not statistically significant (p = 0.26). Similarly, leukemia mortality RR was elevated to 1.020 (0.7$1$2$3.39), but not significantly (p = 0.90) and by less than all-cause mortality. The increase in all-cause mortality did not appear to concentrate in any of the disease groups we considered. Of the 44 other specific cancers and disease categories we examined, there were no statistically significant increases in mortality. The overall elevation of mortality rate ratios for malignancies and leukemias in the participants were not statistically significant and, in fact, were lower than for many other causes of death.
• Navy mortality due to all malignancies and leukemia did not vary substantially among our exposure surrogate groups (i.e., those who boarded target ships after a detonation vs. those who did not, and those enlisted personnel who had an Engineering & Hull occupational specialty vs. those in other specialties).
• Participants who boarded target ships were thought to be more highly exposed than the rest of the participant group. Relative to the controls (nonparticipating comparison group), boarding participants experienced a 5.7 percent increase in all-cause mortality, RR equal to 1.057 (1.01$1$2$3.10), p = 0.0093, whereas the nonboarders (less exposed participant group) experienced a 4.3 percent increase (RR = 1.043 [1.01$1$2$3.073], p = 0.0028). Aside from all-cause mortality, risks for boarding participants did not significantly exceed those for controls for any of the disease categories, and risks relative to controls were similar for boarding and nonboarding participants. The increase in risk for all-malignancies among the participants was 2.6 percent (RR = 1.026 [0.9$1$2$3.12], p = 0.55) for boarders and 1 percent (RR = 1.010 [0.9$1$2$3.068], p = 0.73) for nonboarders. For leukemia, the increase in mortality risk for boarders was, 0.7 percent (RR = 1.007 [0.6$1$2$3.66], p = 0.98) and for nonboarders, 2.4 percent (RR = 1.024 [0.73$1$2$3.422], p = 0.89). In all cases the 95% confidence intervals overlap, suggesting the difference between boarders and nonboarders could well be due to chance.
• Those Navy participants holding an Engineering & Hull (E&H) occupational specialty were thought to be more highly exposed to radiation than their non-E&H counterparts. However, the E&H participants had essentially the same risk of mortality from all causes as non-E&H participants (RR = 0.99 [0.9$1$2$3.038], p = .81). For all malignancies and leukemias, the rate ratios were somewhat higher, 1.051 [0.9$1$2$3.14] and 1.51 [0.9$1$2$3.44] respectively, but both could be attributed to chance (p = 0.25 and 0.088 respectively). Risk ratios for leukemia and malignancies among E&H controls showed a similar elevation relative to non-E&H controls, suggesting that a factor specifically associated with CROSSROADS was not likely to have been the cause.

These findings do not support a hypothesis that exposure to ionizing radiation was the cause of increased mortality among CROSSROADS participants. Had radiation been a significant contributor to increased risk of mortality, we should have seen significantly increased mortality due to malignancies, particularly leukemia, in participants thought to have received higher radiation doses relative to participants with lower doses and to unexposed controls. We did not observe any such effects. We note, however, that this study was neither intended not designed to be an investigation of low-level radiations effects, per se, and it should not be interpreted as such.

In comparing the findings and methods employed in this study with those of other investigations of atomic veteran mortality, we have identified a possible self-selection bias in the participant cohort: participants who died of a disease (particularly cancer) may have been more likely than healthy participants to have identified themselves to the NTPR, and hence become a part of this study. Such a bias would have resulted in an apparent increase in death rates among the participants. We do not have data with which to make a good quantitative estimate of this potential bias. However, the roster of participants is nearly complete, and mortality from all malignancies and leukemia was lower, not higher, than the increase in all-cause mortality. These factors suggest that a self-selection bias was not entirely responsible for the finding of increased all-cause mortality in study participants.

We believe that the elevated risk of all-cause mortality in CROSSROADS participants relative to a comparable military comparison group is probably the result of two factors. The first is an unidentified factor, other than radiation, associated with participation in, or presence at, the CROSSROADS test. The second is a self-selection bias within the participant roster. However, the relative contributions of these two explanations cannot be accurately determined within available resources for this project.