Accounting for errors in dose estimates used in studies of workers exposed to external radiation.

This paper discusses approaches for accounting for errors in dose estimates used in dose-response analyses of data from epidemiologic studies of workers exposed to external radiation and illustrates these approaches with analyses of data on workers at the Hanford site. In these analyses, estimates of the excess relative risk are corrected for bias in recorded doses as estimates of organ dose, and confidence intervals reflect uncertainty in the correction factors. For the Hanford data, these procedures did not greatly modify results for all cancer excluding leukemia, but the upper confidence limit for leukemia was increased by about 40%, a difference that is of some importance in comparing worker-based estimates and confidence intervals with estimates that serve as the basis of radiation protection standards. It is argued that aside from taking account of uncertainty in correction factors, no additional corrections are needed to address random errors resulting from variation in exposure energies and geometries. In addition, it is shown that because the larger cumulative doses, which are most influential in dose-response analyses, are the sums of large numbers of independent dosimeter readings, random errors resulting from variation in laboratory measurements are unlikely to be important for epidemiologic purposes. It is hoped that the approaches illustrated in this paper will be applied to future analyses of data from worker studies, especially combined analyses of data from several countries. Taking account of uncertainty in factors that correct for systematic bias will be especially important as uncertainty resulting from sampling variation decreases.