Events and Seminars : 2012 Seminars

Predictive Bayesian Dose-Response Assessment in a Multiplicative Correlated World

The probabilities of everyday outcomes follow a bell-shaped curve much of the time.  However, many other outcomes, such as the magnitudes of hurricanes, solar flares, oil spills, pathogen counts in drinking water, and, apparently, human illnesses, have highly skewed, L-shaped distributions.  That is, the vast majority of such outcomes are very small, and the rare events quite large.  In this talk we will consider how such distributions emerge in these complex systems.  Recognition of the multiplicative correlated nature of these processes will explain the L-shaped distribution, and reveal an emergent, attracting form of the dose-response function for chemicals, mixtures, pathogens, and potentially medicines.  Then, the use of predictive Bayesian techniques to address parameter uncertainty, due to the sparse and noisy nature of dose-response data, will be explained.  Use of the emergent function in the predictive Bayesian framework will then be shown to produce a predictive dose-response function that is broader when available information is limited, and vice versa.  In this way, the unconditional probability of illness is obtained from available information, as a consistent (absolute) assessment of pathogenic, toxic and, potentially, medicinal health effects.

Click here to view slides