Biomedical Engineering Reference
In-Depth Information
the unwitting introduction of small, but systematic, and possibly
critical, errors.
The most critical deficiency of the visual optimization technique
employed by Battelle is readily apparent in Table 6.1 of the Battelle
(2006) report, the summary final model parameter values: no standard
errors of estimate are provided. It is thus impossible to quantify the bias,
uncertainty, and/or variability associated with the Battelle (2006)
model's final parameter estimates. Some estimates, such as those for
body weights, may well be very precise and accurate, while others, such
as the human K
m
and V
max
estimates, may have residual uncertainties
and/or biases that span orders of magnitude. Visual optimization fails
to provide such critical information regarding bias, uncertainty, and/or
variability.
Importantly, the human inhalation data that were used in model
calibration by Battelle show no evidence of the saturation of methanol
metabolism. Thus, only the ratio of V
max
to K
m
will bewell specified (c.f.,
Sweeney et al., 2004). In such circumstances, methanol metabolism is
most efficiently represented by pseudo-first-order kinetics, as was
employed by Bouchard et al. (2001). Other data, such as those cited in
Perkins et al. (1995), are required to specify the human values of K
m
and
V
max
individually. TheBattelle (2006) report's final humanK
m
estimate of
12mg/l is exactly the same as itsmouse estimate, and, as notedpreviously,
this is altogether inconsistent with the human andmonkey values reported
in the published scientific literature. This suggests strongly that the visual
optimization technique employed by Battelle (2006) was simply
incapable of generating reliable estimates of the key parameters (V
max
and K
m
) that together specify the human metabolism of methanol.
Faced with a bewildering array of data from multiple sources
involving multiple routes of exposure, a more formal and systematic
approach to parameter estimation than that provided by visual optimi-
zation is required, and the approach best suited to this complex task is
Bayesian Markov chain Monte Carlo (MCMC) modeling (c.f., Gelman
et al., 1996; Bernillon and Bois, 2000). With the MCMC approach,
initial PBPK model parameter estimates, including estimated standard
errors, are extracted from the peer-reviewed scientific literature. This
information is then used to construct a prior multivariate probability
