Combining Estimates of Potency and Benchmark Dose from Multiple Studies. R. W. Setzer and A. B. Lowit, US Environmental Protection Agency
While the normal situation in dose-response assessment for environmental agents is a paucity of usable data, we are sometimes embarrassed with the riches of multiple datasets for estimating potency or benchmark dose for the same endpoint. In such cases, often the variability of estimates among studies exceeds that predicted from the within-study variability. This extra variability can be attributed to methodological differences among laboratories, differences in the animals used, or any of the other small, undiagnosed sources of variability that often appear in studies of this sort. Where multiple studies exist, an honest assessment of benchmark dose or potency should account for the uncertainty of the estimate induced by the variability of estimates across studies. We have approached the problem with hierarchical statistical models, assuming that the variability among studies is random. We illustrate one approach to the problem, the "global two-stage method" (Davidian, M. and Giltinan, D., "Nonlinear Models for Repeated Measurement Data", 1995) by estimating both potency and a benchmark dose using data on cholinesterase inhibition due to an organophosphate pesticide. In this example, there are multiple sets of studies, each set containing several dose-response studies. A dose-response model was fit to each dose-response study. That model was used to compute an estimate of potency (and standard error) and a benchmark dose for a 10% reduction in cholinesterase activity (with standard error). The individual estimates were then combined to give an estimate of potency and benchmark dose that takes into account the variability among studies within sets of studies, and among sets of studies. Finally, we assess the coverage probabilities of the resulting confidence intervals with Monte Carlo simulation. The approach described here is generally applicable, and would be particularly useful for combining estimates of benchmark doses from multiple studies. This is an abstract of a proposed presentation and does not necessarily reflect EPA policy.
Go to . . .
2001 SRA Annual Meeting Table of Contents
2001 SRA Annual Meeting Author Index
Main Abstracts Menu Page
RiskWorld Home Page