Using analytic models for risk-based responses to pathogenic agents in the environment

Abstract

In the absence of quantitative risk-based information, responses to pathogenic agents in the environment, especially bioterrorism agents, can be guided by unrealistically low levels of risk tolerance (i.e. 0), as seen after the 2001 anthrax mail attacks.A novel and abbreviated risk framework links a series of three analytical models to inform a response decision after an intentional release of Bacillus anthracis. The first two analytic models are presented here in detail. A decision model establishes an actionable risk threshold, which can then be translated into a dose using a dose-response model. A Bayesian hierarchical approach is used to develop the dose-response model using extant data. Environmental concentrations which produce the actionable exposure dose are then calculated based on a third, fate-and-transport model, which was published elsewhere by Hong et al 2010.In the first model, a decision tree outlines the costs and benefits of the relevant response decisions. Sensitivity analysis is then conducted to establish a decision threshold based on actionable risk. The dose-response model used to translate this risk into dose relies heavily on experiments using surrogate hosts and microbial strains, as is the case with many infectious agents. Treatment of uncertainty and extrapolation of the results of dose-response studies therefore present a number of modeling challenges The Bayesian hierarchical model was developed to estimate generalized infectivity and susceptibility for a meta-analysis of available data sets. This modeling approach allows for quantitative estimates of uncertainty when predictions are made for unobserved species (i.e. humans) or strains. The dose-response model was first applied to a rich, U. S. Army data set of historical Bacillus anthracis studies. The results of the analysis produced multiple benchmark values for decision makers. The hierarchical approach was then applied to the meta-analysis of three different pathogen data sets, typical of those found in the literature – Bacillus anthracis, Cryptosporidum species, and Francisella tularensis. Model comparison using the deviance information criterion (DIC) was conducted to evaluate hierarchical and individual forms of the exponential and beta-Poisson dose response models. For these pathogens, the DIC indicated preference for fitting individual models to each data set. Model selection for the meta-analysis was dependent on the characteristics of the data sets for each pathogen.%%%%Ph.D., Environmental Engineering – Drexel University, 2010