All dose-response analyses were restricted to the subcohort of white males employed at X-10 or Y-12 (see Materials and Methods). In previous studies with follow-up through 1984 the dose-response analyses were based on white males (N = 8318, with 346 cancer deaths) employed only at X-10 [22,23]. The addition of Y-12 and multiple facility workers increased the cohort size to 28,347 with 1,038 cancer deaths. The methods used in this study differ in several ways from those in our previous analysis of the X-10 only group. First, there were ten dose groups equally spaced on a logarithmic scale (instead of eight equally spaced on a linear scale with a width of 20mSv). Second, the person-year weighted average dose was used in each cell of the ADS. In the previous analysis the median value of all of the person-years in all of the cells in the highest (greater than 120 mSv) dose category was used as the value of D for all cells in the highest dose group, and the interval mid-point was used for all of the other cells. Third, external rates and a multiplicative main effects model were used to describe the baseline rates, and internal radiation monitoring status and facility factors were included in the model. In the previous dose-response analyses external rates were not used and the baseline rates were described with a parametric model -- see the Appendix for further discussion. Fourth, in previous dose-response analyses deaths where cancer was a ''contributing cause" were included. Only underlying causes of death were used in the current study. Fifth, all of the previous dose-response analyses were limited to the exponential relative risk model.
In all dose-response analyses the potential biases associated with time-related factors and time dependent exposures were dealt with through the use of an internal control group based on birth cohort, age at risk, and length of employment. Different approaches to dealing with time dependent variables are possible---see e.g.[12,26] and alternative approaches to dealing with these and other variables in mortality studies of nuclear industry worker in the U.S.  and the United Kingdom [27,28,29,30] have been presented.