From column 10 of Table III it is clear that monthly workers have much lower death rates than nonmonthly workers. The all cause death rates are lower by 41.5 L% for both white males and females---this corresponds to a relative risk of exp(-.415) = 0.66 for monthly versus nonmonthly workers. The log relative risk estimates for length of employment are all positive---see column 11 of Table III. This may reflect a more hazardous environment and/or lower SES of short term workers. The facility effects in columns 4 through 8 of Table III are estimates of the SMRs (in L% units) for each facility at the middle of follow-up (1965-69) for nonmonthly long term workers, and provide a direct comparison with the external referent population. These estimates are internally ''adjusted" for SES, length of employment, and calendar time; and for attained age through the use of the external rates. For example, for all cause mortality for white males employed only at TEC the facility effect is -5.9L%, and the estimated SMR is exp (-.059) = .94. An estimate of the log relative risk with one facility as the referent facility is obtained as the difference of the two facility effect estimates. Using the facility effect estimates for all cause mortality for white males (see line 1 columns 4-8 of Table III ) with X-10 as the referent group leads to:
X-10 TEC Y-12 K-25 MULT Effect Diff (L%) 0.0 20.9 9.7 25.8 20.8 Relative Risk 1.0 1.23 1.10 1.29 1.23 .This internal comparison shows that white males employed only at TEC or K-25 and multiple facility workers have higher death rates than similar workers employed only at X-10 or Y-12. The large value of the LRT statistic in column 9 of the Table III---LRT = 101.0, 4 degrees of freedom (df)---indicates that the facility differences are not due to large numbers of deaths. Evaluation of the corresponding estimates for white females (see row 1 of the bottom panel of Table III) reveals a similar pattern. The facility effect differences for all cause mortality for white females (using X-10 as the referent group) are:
X-10 TEC Y-12 K-25 MULT Effect Diff (L%) 0.0 24.6 -23.0 41.0 -40.3 Relative Risk 1.0 1.28 0.79 1.51 1.50 .As was noted earlier, the facility effect estimates for females are possibly biased downward as the result of under ascertainment of vital status, but the internal comparisons are valid if failure to determine vital status is the same across facility groups.
Due to the potential for vital status ascertainment problems and the small size of the cohort of nonwhite Oak Ridge workers (3,909 male and 2,826 females),(see Table 3 from Watkins et al.) the results in Table IV have been limited to a few selected causes. The results for nonwhite Oak Ridge workers in Table IV follow the same pattern that was observed for white Oak Ridge workers. Monthly workers have much lower death rates than nonmonthly workers, and short term workers have higher rates than long term workers. For nonwhite females there were so few monthly workers, that estimates of the relative risk for this factor could not be calculated.
Table V shows the results of the main effects analyses for selected cancer causes for white males. Death rate differences among the facilities are large enough and based on a sufficient number of deaths to yield large LRT statistics for lung cancer (LRT = 21.6, 4 df), leukemia (LRT = 10.7, 4 df) and other lymphatic tissues (LRT = 12.3,4 df). The facility effect estimates indicate that lung cancer is low at X-10 (-34.8L%), and that other lymphatic tissue rates are low for X-10 (-92.8L%) and highest for Y-12 (71.9L%). The leukemia estimate is lowest at Y-12 (-64.5L%) and is high for X-10 (61.7L%).