Table 9-1 shows vital status percentages and the availability of cause-of-death information by analysis group. There are no large differences in the percentage assumed alive between Project SHAD (Shipboard Hazard and Defense) participants and controls within exposure group, and with the exception of group B, the proportion assumed alive is roughly three-quarters for both participants and controls. Subjects with only date of death or fact of death include those whose death occurred before 1979 (for whom we were unable to obtain causes) and represent 4–5 percent of participants or controls across all groups.

Table 9-2 shows the availability of follow-up information by exposure group and Project SHAD participation status. Mortality follow-up was done by matching both the Beneficiary Identification and Records Locator Subsystem (BIRLS) file, using military service number as well as Social Security number (SSN), and the National Death Index (NDI) file, using only SSNs. Because we consider mortality follow-up that relied on NDI to be virtually complete, only study subjects with SSN can be considered well followed, evidenced by the fact that the crude death rate among subjects with SSNs (22.7 percent) was roughly eight-fold higher than among subjects without SSNs (2.8 percent). Table 9-2 shows that in all but group B controls, the percentage of not-well-followed subjects (i.e., the sum of the first two columns) is less than 6 percent. Subsequent mortality analyses will therefore be done using all subjects and then only subjects with SSNs; although the latter omits some known deaths identified by BIRLS without SSNs, it does mean that all such subjects will have been searched for in the National Death Index. The results of these two analyses of all-cause mortality are shown in Tables 9-3 and 9-4, respectively.

Table 9-3 shows the results of proportional hazards analyses of total mortality by exposure group with all subjects included, regardless of completeness of mortality follow-up. All analyses were adjusted for age, race, and pay grade, but only in group B were there sufficient Marines to adjust also for service branch. There were no statistically significant differences in all-cause mortality between Project SHAD participants and controls, although in group B the hazard ratio (HR) was 1.25 (95 percent confidence interval [CI] 0.99–1.60). The effect of age was statistically significant in all groups, while race was significant only in group B, in which nonwhites had significantly higher mortality than whites; in all other groups, all-cause mortality was lower among nonwhites. Officers had significantly lower mortality than enlisted personnel in all groups, except for group C, and Marines in group B had significantly higher mortality than Navy personnel in group B.

Table 9-4 shows the results of the same proportional hazards analysis of all-cause mortality, including only those with SSN in whom mortality follow-up was assumed to be most complete. With the exception of group B, the results of this analysis are much the same as shown in Table 9-3. In group B, the hazard ratios associated with race and pay grade are no longer statistically significant.

Table 9-5 shows the results of proportional hazards ratio analyses of selected cause-specific mortality end points for only subjects with SSNs. Group A participants had a statistically significantly higher hazard ratio than

controls for death because of cardiovascular disease. Group B participants had statistically significantly higher hazard ratios for cancer and cardiovascular deaths. Although group B participants had higher death rates than controls for many of the selected mortality outcomes in Table 9-5, most differences were not statistically significant, due to the relatively small number of deaths. Finally, comparing all Project SHAD participants versus all controls, heart disease deaths showed a statistically significant increase.

As explained in Chapter 8, standardized mortality ratios (SMRs) are used to compare the number of observed deaths in a cohort with the number of expected deaths in the U.S. general population of the same age, race, and sex. An SMR value of 100 indicates that the number of observed deaths equals the number expected. Table 9-6 shows SMRs for subjects with SSNs by analysis group for various causes of death. Because we did not have

causes for deaths prior to 1979, the follow-up period for calculation of SMRs begins in 1979. The causes of death in Table 9-6 mirror those in Table 9-5, except that diabetes has replaced endocrine disease.

All-cause SMRs are all close to 100, save for group A controls and group B participants, indicating that overall mortality in these analysis groups is close to that of the U.S. population. However, all-cause SMRs for all participants and all controls combined are slightly, but statistically significantly, greater than 100. Although the cancer mortality SMR is slightly above 100 in almost all groups, it is statistically significant only for all controls and group A controls. Most of the excess cancer deaths were due to lung cancer, which might be attributable to smoking, and several participant and control group SMRs for nonmalignant respiratory disease are above 100, but not statistically significantly different from 100. SMRs for heart disease deaths are all close to 100, with the exception of group B participants. Deaths due to external causes tended to have low SMRs, and were statistically significantly lower among all participants.