Biomedical Engineering Reference
In-Depth Information
human model calibration or validation (c.f., Osterloh et al., 1996;
Sedivic et al., 1981; Batterman et al., 1998). As noted previously,
the Bouchard et al. (2001) model is more linear and simpler than the
Battelle (2006) model, yet it accurately predicts human blood methanol
levels arising from inhalation exposures up to the highest concentration
(800 ppm) that has been tested. Given the excellent performance of the
Bouchard et al. (2001) human model, it is not clear that the Battelle
(2006) human model is in any way superior. An objective comparison of
predictions by the Battelle (2006) and Bouchard et al. (2001) human
models using human blood methanol data would go a long way toward
answering this important question.
6.8
SUMMARY OF LESSONS LEARNED
Except for the first, each of the preceding sections of this chapter began
by posing a critical question that was then addressed in considerable
detail using the Battelle (2006) and Bouchard et al. (2001) models of
methanol disposition in mice and humans in exemplary fashion. Here,
the answers to these questions, which comprise our “lessons learned,”
are summarized:
The mouse is uniquely sensitive, relative to all other species
tested, to the toxic effects of methanol, due principally to its
limited metabolic capability for methanol. If PBPK modeling
suggests otherwise, trust the data; suspect the model.
If model predictions are discordant with those of other validated
models and/or experimental data, suspect the model and search
for errors in its parameterization.
Check to confirm that the numerical values assigned to model
parameters are consistent with those in the published literature.
During the model fitting and parameter estimation process,
inspect simulation results carefully for the possibility of system-
atic bias. Attempt to discover its cause.
“Visual optimization” should be abandoned as a model fitting
technique; it should be replaced by a Bayesian Markov chain
Monte Carlo (MCMC) modeling approach.
