Comparing Human Health Risk Values: An Analysis of Science Policy Choices

This dissertation research project entailed three separate yet related projects. Toxicity tests are widely used to set "acceptable" levels of chemical exposure. Different organizations have identified a base set of tests specifying a mix of endpoints, durations, and species to be tested. A specific test and endpoint is chosen as the basis for calculation of human health risk values (HHRV) like reference doses (RfDs). In the first study, we empirically evaluated the data and choices made in setting acute and chronic RfDs for 352 conventional pesticides. The results suggest that for Acute, Acute-Female Specific, and Chronic RfDs one test is used far more than others. Ninety-six percent of the 116 Acute Female-Specific RfDs relied on a developmental toxicity test and 78% of Chronic RfDs used the chronic bioassay. Tests in rats were used far more often than other species in all RfD calculations. For all types of RfDs a total uncertainty factor (UF) of 100 was most common although values as low as 1 and as high as 3000 were seen. These results provide insights not only into the science policy frameworks used but also into ways toxicity testing and risk assessment may be streamlined and made more efficient. The goal of the second and third studies was to systematically evaluate the choices made in deriving an HHRV for a given chemical by different organizations, specifically those from the U.S. Environmental Protection Agency, Health Canada, RIVM (Netherlands), and the U.S. Agency for Toxic Substances and Disease Registry. In the second study, we developed a methodological approach for comparing both the HHRVs and the specific choices made in the process of deriving an HHRV across these organizations. Chemicals included in these studies were primarily industrial chemicals or drinking water contaminants (~83% of chemicals evaluated), as well as a relatively small number of canceled pesticides (~9%) and currently registered pesticides (~8%). Overall, across the 96 unique chemicals and 171 2-way organizational comparisons, the HHRV agreed approximately 26% of the time. A qualitative method for identifying the primary factors influencing these HHRV differences was also developed, using arrays of HHRVs across organizations for the same chemical. The primary factors identified were disagreement on the critical study and differential application of the total UF across organizations. Of the cases where the total UF was the primary factor influencing HHRV disagreement, the database UF had the largest impact. In the third study, we further evaluated these 171 organizational comparisons, developing a quantitative method for identifying the factors to which HHRV agreement (that is, when both organizations considering the same data set the identical HHRV values) is most sensitive. To conduct this analysis, a Bayesian Belief Network (BBN) was built using expert judgment, including the specific science policy choices analysis made in the context of setting an HHRV. Based on a sensitivity of findings analysis, HHRV agreement is most sensitive to the point of departure value, followed by the total UF, the critical study, critical effect, animal model, and point of departure approach. This analysis also considered the specific impacts of individual UF, with the database UF and the subchronic-to-chronic UF being identified as primary factors impacting the total UF differences observed across organizations. The results of the sensitivity of findings analyses were strengthened and confirmed by frequency analyses evaluating which choices most often disagreed when the HHRV and the total UF disagreed. Limitations of this analysis include the small number of organizational comparisons (N=171), with future work planned to further evaluate the BBN model using additional comparisons and case testing using expert analysis.