Introduction: The use of human data to calibrate and validate a physiologically based pharmacokinetic (PBPK) model has the clear advantage of pertaining to the species of interest, namely humans. A challenge in using these data is their often sparse, heterogeneous nature, which may require special methods. Approaches for evaluating sources of variability and uncertainty in a human lung dosimetry model are described in this study.
Methods: A multivariate optimization procedure was used to fit a dosimetry model to data of 131 U.S. coal miners. These data include workplace exposures and end-of-life particle burdens in the lungs and hilar lymph nodes. Uncertainty in model structure was investigated by fitting various model forms for particle clearance and sequestration of particles in the lung interstitium. A sensitivity analysis was performed to determine which model parameters had the most influence on model output. Distributions of clearance parameters were estimated by fitting the model to each individual's data, and this information was used to predict inter-individual differences in lung particle burdens at given exposures. The influence of smoking history, race and pulmonary fibrosis on the individual's estimated clearance parameters was also evaluated.
Results: The model structure that provided the best fit to these coal miner data includes a first-order interstitialization process and no dose-dependent decline in alveolar clearance. The parameter that had the largest influence on model output is fractional deposition. Race and fibrosis severity category were statistically significant predictors of individual's estimated alveolar clearance rate coefficients (P < 0.03 and P < 0.01-0.06, respectively), but smoking history (ever, never) was not (P < 0.4). Adjustments for these group differences provided some improvement in the dosimetry model fit (up to 25% reduction in the mean squared error), although unexplained inter-individual differences made up the largest source of variability. Lung burdens were inversely associated with the miners' estimated clearance parameters, e.g. individuals with slower estimated clearance had higher observed lung burdens.
Conclusions: The methods described in this study were used to examine issues of uncertainty in the model structure and variability of the miners' estimated clearance parameters. Estimated individual clearance had a large influence on predicted lung burden, which would also affect disease risk. These findings are useful for risk assessment, by providing estimates of the distribution of lung burdens expected under given exposure conditions.