Table VIII gives the unadjusted SMR (in % and L%), the value of the score test statistic based on all of the data, and the estimated dose-response coefficient based on the main effects model with exponential relative risk using the ''low dose" data. In this low dose analysis all cells in the ADS in the highest (640mSv+) dose group were omitted so that once an individual's cumulative dose exceeded 640 mSv he was censored. This choice was based on the results of the regression diagnostics ---see Fig. AII in the Appendix---and the fact that a radiation worker would be very unlikely to receive a cumulative lifetime dose above 640mSv. There are 41 cells with 220 persons years (total person years is 603,365) and 3 cancer deaths in the highest dose group (see Appendix Table AIV). The regression diagnostics in the Appendix (which are based on Eq. 3 using all data) suggest these few cells with cumulative dose values that exceed 640 mSv have a large influence on the estimate of the dose-response coefficient for the exponential relative risk model. For the X10/Y12 ever cohort about 0.1 percent of the workers exceeded this value. (It may be more appropriate to think of this as a low to medium dose analysis.)
The effect of the high dose group is to decrease the trend estimate when the exponential model is used, since most of the 41 cells have no events and high leverage values. For all cancer causes the estimate obtained using all the data is lower than that obtained when cells for the highest dose group are omitted. The low dose estimate (1.59 per Sv) is three times larger than the estimate using all the data (0.49 per Sv). Fig. 2A shows the relative risk estimates for all cancer cancer for each dose group (ten year lag). These estimates were obtained using the ten level factor XG in the main effects model and are therefore ''adjusted" for factors B, S, L, IG, F, and age through use of the external rates (see Appendix). Fig. 2A shows that the exponential dose-response is reasonable in the low dose region, but that it does not provide a good description of the relation between risk and dose over the range of doses in this study. Note that while the relative risk for the highest dose group is lower than that predicted from the exponential model fitted over the lower doses, it is roughly in line with the linear ERR model.
The score test values in column 6 of Table VIII are based on all the data and the LRT values in column 9 were obtained with the cells in the highest does category excluded. When all the data are used the results for all cancer and lung cancer show the strongest association between external dose and deaths certified to these causes (see column 6 of Table VIII). When the high dose data are excluded there is a decrease in strength of the association for all cancer, and a large decrease occurs for lung cancer (see LRT statistics column 9 of Table VIII). The opposite pattern is observed for digestive system cancer and emphysema. Estimates of the relative risk by dose group and those derived from the exponential relative risk and ERR models are shown graphically for lung cancer in Fig. 2B and for digestive cancer in Fig. 2C. For digestive cancer there were no deaths in the highest dose group so the relative risk was estimated for the last two groups combined. The dose values used for the graphical displays in Figs. 2 and 3 are the person-year weighted average of the doses associated with each stratum in the ADS (see the Appendix) for a given dose category. For prostate cancer there were no deaths in the highest two dose categories so the relative risk at the highest dose value in Fig.2D is a combined estimate for the three highest dose groups. The dose-response coefficient for prostate cancer is positive but the score test and LRT do not show a strong association with dose (see row 6 of Table VIII). Fig. 2D shows that the risk is increased for workers with external doses greater than zero (recall that the internal referent group is based on cumulative lifetime occupational dose with a ten year lag). An ad hoc analysis yields a relative risk estimate of 70.4L% (SE = 29.8) for those with any external radiation exposure versus those with zero dose. This ad hoc result was obtained using the main effects model with an indicator variable for external radiation exposure. The unadjusted SMR for the ten year lag zero dose group is -19.8L% (SE = 21.8), and for those with dose greater than zero the unadjusted SMR is 16.4L% (SE = 13.4).
Fig. 3A shows relative risk versus doses for all cause mortality. Figs. 3B-3D show similar results for diseases of the circulatory system, nonmalignant respiratory disease, and all external causes of death. death. Dose-response coefficients, score test statistics, and LRT statistics are given in Table VIII.