You are here:
Relating Data and Models to Characterize Parameter and Prediction Uncertainty
Citation:
SETZER, R. W. Relating Data and Models to Characterize Parameter and Prediction Uncertainty. Presented at Continuing Education Course at the Society of Toxicology Annual Meeting, Baltimore, MD, March 15, 2009.
Impact/Purpose:
Monte Carlo simulation is typically used to propagate the uncertainty at individual steps to the final prediction, but it must be structured appropriately to maintain the important distinction between uncertainty and variability. Finally, I discuss the role of sensitivity analysis in identifying important sources of uncertainty.
Description:
Applying PBPK models in risk analysis requires that we realistically assess the uncertainty of relevant model predictions in as quantitative a way as possible. The reality of human variability may add a confusing feature to the overall uncertainty assessment, as uncertainty and variability are often confused (and confounded). This lecture will discuss the nature of uncertainty and variability with specific reference to the use of PBPK models in risk assessment. It is useful to consider the steps in model development separately: development in an animal experimental model → extrapolation to humans and parameterization of human model → combination with human variability model → prediction of relevant human dose metric. I discuss approaches for evaluating the contribution each step makes to the overall uncertainty. Monte Carlo simulation is typically used to propagate the uncertainty at individual steps to the final prediction, but it must be structured appropriately to maintain the important distinction between uncertainty and variability. Finally, I discuss the role of sensitivity analysis in identifying important sources of uncertainty.