Relative risk regression models were fitted to the data with the AMFIT program (30), which calculates excess relative risk from internal comparisons of cumulative dose and disease status. Specifically, the expected number of incident cases was described by a linear excess relative risk model Njλj(1 + βdj)