A Bayesian approach for reducing uncertainty in physiologically based pharmacokinetic models
A. Roy, Y.-C. Chien, C.P. Weisel, M. Ouyang, P.G. Georgopoulos (EOHSI, UMDNJ - R.W. Johnson Medical School and Rutgers University)
Estimation of parameters in mechanistic environmental and biological process models is often impractical using classical estimation techniques due to the typically complex structure of these models and the large number of structural and constitutive parameters that appear in their formulation. Many of the limitations of classical parameter estimation techniques can be circumvented by employing Markov Chain Monte Carlo (MCMC) simulation, an empirical Bayesian methodology. MCMC methods generate probability distributions rather than point estimates for model parameters, and incorporate prior information on model parameters together with information contained in the data. MCMC methods are particularly suited for reducing uncertainty in the parameters of mechanistic environmental and/or biological models as reasonable prior distributions can usually be specified for parameters in mechanistic models based on physical constraints.
In this paper, the MCMC simulation methodology is applied to the estimation of the parameters of a physiologically based pharmacokinetic (PBPK) model for tetrachloroethylene using an extensive set of data on a single individual. The PBPK model describes the exhaled breath concentrations measured in an individual, following controlled and field low environmental level exposure to tetrachloroethylene. Bayesian inference using MCMC simulation resulted in decreases of more than 10% in the uncertainty of 6 of 15 model parameters.