Biomedical Engineering Reference
In-Depth Information
distribution of the model's parameters, which is subsequently updated
via fitting to (i.e., simulation of) additional experimental data that have
not been used to specify the prior distribution, in a stepwise manner. As
new data sets are added, one or more at a time, their contributions to
parameter uncertainty/variability and bias reduction can be assessed
objectively in terms of the extent to which resulting posterior mean
parameter estimates shift away from their prior values and their
posterior variance estimates shrink or widen.
The power of the MCMC approach in this Bayesian estimation
context is its capacity to achieve convergence to final parameter values
even when the PBPK model and its parameter covariance structure are
extremely complex and the data available for model calibration are very
sparse. Interestingly, USEPA is already quite familiar with the compel-
ling advantages of the MCMC approach to PBPK model estimation
since it recently (1) cosponsored a state-of-science monograph that
included multiple applications of this approach to trichloroethylene
(USEPA, 2000) and (2) published a new PBPK model for benzene with
parameters estimated using MCMC methodology (Yokley et al., 2006).
Since the Rogers et al. (1993) data set is crucial to the methanol risk
assessment process, particularly for the 1000, 2000, and 5000 ppm
exposure groups, it is important that these data receive the greatest weight
when other data sets are included in the fitting process. Indeed, it might
well be appropriate to select the mouse prior parameter distribution to
reproduce, as closely as possible, the Rogers et al. (1993) mean blood
methanol levels 7 hours after exposure onset. Then, as other data sets (e.g.,
those fromDorman et al. (1995) and Perkins et al. (1995)) are brought into
consideration, their contribution, if any, tobias and uncertainty/variability
reduction can be systematically and objectively determined.
6.7 WHEN HUMAN DATA ARE AVAILABLE, SHOULDN'T
THEY BE UTILIZED IN MAKING AN OBJECTIVE
COMPARISON OF MODEL-SPECIFIC PREDICTIONS?
Many of the data sets employed by Battelle in human model parameter
estimation were also employed by Bouchard et al. (2001) for either
