Accounting for bias in dose estimates in analyses of data from nuclear worker mortality studies

Although dose estimates used in epidemiologic studies of nuclear workers are far superior to exposure measures used in many epidemiologic studies, they are subject to several types of bias. These include bias resulting from practices used to measure and record very low doses and from underestimation of dose from neutrons. They also include bias resulting because available dose estimates do not include dose from internally deposited radionuclides, occupational dose received after workers have terminated employment at the facility of interest, or dose from background or medical exposures. These biases can potentially distort dose-response analyses and lead to biased risk estimates and confidence limits. This paper uses data on workers at the Hanford site to investigate the possible effects of these biases on dose-response analyses of leukemia and of all cancer except leukemia. This is accomplished by conducting analyses based on a variety of assumptions regarding the nature and magnitude of these biases. Results were not modified greatly (as compared to those based on the dose as recorded) for any of the 40 sensitivity analyses presented, including some based on fairly extreme assumptions. However, analyses that excluded workers with potential for dose from internal depositions and from neutrons did increase both the risk estimate and upper confidence limits; it may be desirable to emphasize this approach in future presentations. It is recommended that similar sensitivity analyses be performed with data from other worker studies, addressing the specific dosimetry problems that have been identified in those populations.