Investigating Hierarchical Statistical Methods for Physiologically Based Pharmacokinetic ModelsEPA Grant Number: U915179
Title: Investigating Hierarchical Statistical Methods for Physiologically Based Pharmacokinetic Models
Investigators: Collins, Amy S.
Institution: North Carolina State University
EPA Project Officer: Carleton, James N
Project Period: October 1, 1997 through September 1, 2000
Project Amount: $102,000
RFA: STAR Graduate Fellowships (1997) RFA Text | Recipients Lists
Research Category: Fellowship - Mathematics , Academic Fellowships , Environmental Statistics
The objective of this research project is to use a statistical hierarchical approach to current physiologically based pharmacokinetic (PBPK) models to develop a model that most accurately accounts for variation in existing data to allow a more precise extrapolation to humans. Although PBPK models exist for a variety of chemicals, the U.S. Environmental Protection Agency (EPA) has been cautious when using these models to assess human risk. This caution is appropriate due to the lack of statistically sound procedures in model techniques. Although most commercial simulation and optimization software performs curve fitting, more sophisticated statistical tools are now required from the modelers, as focus is shifting to uncertainty and variability assessment.
The usual model approach has been to obtain fitted estimates for unknown parameters by using maximum likelihood techniques. Yet, the validity of estimates of fitted parameters depends on the assumed error model. Used to derive the LLF, the error model in SimuSolv® is not appropriate because it confounds two sources of variation: intraindividual variability, which is variation among measurements within a given individual, and interindividual variability, which is random variation among individuals. If sources of variation are not consideredtaken into proper account, misleading estimates of the parameters and uncertainty in those estimates may result. These two variation components can be consideredtaken into appropriate account in a statistical hierarchical or staged model. I used the PBPK models for tert-amyl methyl ether (TAME) and tert-amyl alcohol (TAA) to contrast inference based on the incorrect error model implemented in SimuSolv® and inference based on a hierarchical model. To fit a hierarchical model, maximization of a more complex objective function than that found in SimuSolv® is required. Using the NLMEM macro, SAS® software is very efficient in implementing hierarchical model fitting for PBPK models.