Recent uses of biological data for the evaluation of A-bomb radiation dosimetry.

Random errors in the DS86 radiation dose estimates used in the analysis of A-bomb survivor data are recognized to have an important impact upon estimates of the risk of late effects such as cancer. Little however is known for certain concerning the distribution of such random errors. This paper gives an overview of recent work at the Radiation Effects Research Foundation (RERF) using multivariate analysis of biological data, including acute effects of radiation exposure, late effects (eg leukemia mortality) and stable chromosome aberrations, for the purpose of evaluating the extent of random error in the estimation of individual doses using DS86. The emphasis here is on analyses of apparent association between biological endpoints, in light of a dosimetry error model framework proposed recently by Pierce et al. Analyses performed to date appear to be consistent with the view that lognormal random dosimetry errors with a standard deviation of 40% or greater of true dose may exist in DS86. Association between radiogenic outcomes in A-bomb survivors, after adjustment for DS86 estimated dose level, has been detected for such widely varying pairs of outcomes as mutant T-cell frequencies and chromosome aberrations, epilation and leukemia mortality, and epilation and chromosome aberrations. The motivation for examining association between pairs of biological endpoints has usually been to determine the extent to which radiation sensitivity varies between individual survivors. Recognizing, however, that random error in dose estimates results in apparent association between biological outcomes is crucial to interpreting studies, such as these, which use data on multiple biological endpoints. To go one step further, in situations where there is a prior knowledge about the biological plausibility of such associations in outcome data the amount of association between radiogenic outcomes (remaining after adjustment for estimated dose), to the extent that they are greater than that assumed to be reasonable, is an important potential source of information concerning the magnitude of random errors in the DS86 dose estimates.