A Bayesian Hierarchical Method for Fitting Multiple Health Endpoints in Toxicity Studies

• Activity Number: 121
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 12:00 PM to 1:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Taeryon Choi*+ and Mark J. Schervish and Ketra Schmitt and Mitchell Small
• Companies: Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: Bayesin hierarchical model ; perchlorate ; dose-response curve ; posterior predictive distribution

Abstract

Bayesian hierarchical models are built to fit multiple health endpoints from dose-response studies of a toxic chemical, perchlorate. In particular, several models are developed to fit data from the Springborn 90-day study of Springborn laboratory Inc. (1998). We propose empirical models to fit the data based upon a mechanistic model derived from the assumed toxicological relationships between dose and the various endpoints. The model building is compatible with the tentative mode-of-action for perchlorate toxicity proposed by the EPA. We use logistic regression models to estimate the probabilities of histopathology endpoints and multivariate regression models for hormone data. All of the models are estimated in a fully Bayesian framework, and predictions about each endpoint in response to dose are simulated based on the posterior predictive distribution. A hierarchical model that tries to exploit possible similarities between different combinations of sex and sacrifice date allows us to produce more stable estimates of dose-response curves. Finally, we investigate whether or not we can make use of any information from the Caldwell et al. (1995) study using alternative approaches.