You are here:
Continuous Toxicological Dose-Response Relationships Are Pretty Homogeneous (Society for Risk Analysis Annual Meeting)
Citation:
Setzer, Woodrow AND W. Slob. Continuous Toxicological Dose-Response Relationships Are Pretty Homogeneous (Society for Risk Analysis Annual Meeting). Presented at Annual Meeting of the Society for Risk Analysis, Arlington, VA, December 06 - 10, 2015. https://doi.org/10.23645/epacomptox.5077717
Impact/Purpose:
Dose-response modeling has been plagued by the fact that we generally do not know the true functional form for the dose-response relationship. Thus, for instance, for benchmark dose computation, multiple models are typically fit to datasets, and a procedure used to pick the best-fitting one. This can lead to the appearance of less uncertainty about the relationship than is really presence when standard statistical methods are used to estimate uncertainty. An alternative is to either use non-parametric methods to estimate the dose-response, or average multiple models, using the variation in fits to estimate the uncertainty. This work strongly suggests that continuous toxicological endpoints generally fall into a narrow range of possible shapes which are well-described by relatively simple models that are already in use. This means that uncertainty about the true model is negligible in the experimental range, and can be quantified using standard tools for quantifying the uncertainty of parameter estimates.
Description:
Dose-response relationships for a wide range of in vivo and in vitro continuous datasets are well-described by a four-parameter exponential or Hill model, based on a recent analysis of multiple historical dose-response datasets, mostly with more than five dose groups (Slob and Setzer, 2014). Furthermore, the estimated shape parameters for the sigmoid models fall within narrow ranges that depend on endpoint and whether the study was in vivo or in vitro. Based on this work, we suggest that the bulk of model uncertainty for continuous endpoints can be covered by parameter uncertainty in Hill or four-parameter exponential models. Using the observed regularity of shape parameters as prior information in Bayesian fits allows sigmoid models to be fit even to datasets with inadequate numbers of doses, or suboptimal dose placement. Additionally, among-dose-group variability was evident over all the datasets. This renders conventional goodness-of-fit tests unreliable, and may unduly influence BMD estimates unless an appropriate error model which includes a hierarchical error component is used. We demonstrate some examples of fitting such a model to simulated and real datasets. This abstract does not necessarily reflect U.S. EPA policy.
URLs/Downloads:
DOI: Continuous Toxicological Dose-Response Relationships Are Pretty Homogeneous (Society for Risk Analysis Annual Meeting)
SETZER-SLOB_ABSTRACT_V2A.PDF (PDF, NA pp, 60.021 KB, about PDF)
SETZER-SLOB_SRA2015_FINAL.PDF (PDF, NA pp, 542.341 KB, about PDF)