Are BMD-Ratios Informative?: The Influence of Experimental Design and Analysis. K. Brand, P. Catalano, J. Hammitt, L. Rhomberg, and J. Evans, Department of Environmental Health, University of Washington, Seattle, WA; and Department of Biostatistics, Harvard School of Public Health (HSPH), Boston, MA and Department of Health Policy and Management, HSPH and Department of Environmental Health, HSPH
Ratios taken between direct and proxy toxicity estimates, obtained from animal bioassays, are commonly used to characterize underlying extrapolation relationships (e.g., animal to human, subchronic to chronic, maternal to developmental, or etc.). Inferences of interest include both the average difference between the proxy and direct measure and the interchemical heterogeneity of this difference. Appropriate inference should consider measurement error and potential censoring processes which can cause bias and imprecision. Such errors and their dependence upon underlying truth will affect the informativeness of datasets of BMD ratios. To examine these issues, we use simulation to ‘calibrate’ the ability of BMD ratios to reveal hypothetical relationships. Of particular interest is the dependence of informativeness on the experimental designs and BMD estimation protocols which were used in practice to obtain each BMD estimate. Our principal concerns for experimental design and estimation protocols, are, respectively, the placement, or centering, of the test doses, and the protocols used for summarizing unconventional bioassay outcomes (e.g., an outcome showing nonsignificant dose response trend). These two issues are given special attention as we mathematically model the process by which BMD ratios are ‘generated’ in practice. Our results show that inferences are subject to non-negligible imprecision (finite sample error), and potentially dramatic biases. The extent and nature of the errors is strongly dependent upon dose centering, and estimation protocols. It is also significantly dependent on exogenous factors (e.g., number of animals used in the bioassays, background rates of response, and dose response shape), and can depend on the nature of the underlying truth. Although results improve under conditions of idealized dose centering, non-negligible errors persist; inferences should account and/or correct for such errors. Implications for regulatory toxicology are discussed.
Funded by the Harvard Center for Risk Analysis.
Go to . . .
1999 SRA Table of Contents
1999 SRA Author Index
Main Abstracts Menu Page
RiskWorld Home Page