Uncertainty in estimating exposure using a toxicokinetic model. The example of 2,3,7,8-tetrachlorodibenzo-p-dioxin

This paper deals with sources of uncertainty in the use of a minimal physiological toxicokinetic model to obtain dose estimates for a dose-response analysis of cancer in an occupational cohort. Toxicokinetic models make it possible to construct exposure parameters that are more closely related to the individual dose than traditional measures of exposures to toxic agents. However, the process introduces a wide array of sources of uncertainty. Selecting a model structure to describe the kinetics of a toxic agent implies necessarily making simplifications and assumptions that influence the range of applicability of the model. Once a model has been selected, the value of certain model parameters (constants) must be assigned, for example, from anthropometric data. The question then arises of how sensitive the model predictions are to variations in the values of these constants. Other model parameters, typically those describing the kinetics of the agent, are next estimated from actual data. There may be limitations in the data concerning, for example, sparseness (too few observations per subject) or missing values. The methods used for parameter estimation carry their own set of assumptions that need to be appropriate to the situation at hand. In summary, the dioxin example is used to characterize the sources of uncertainty at different levels, such as model structure, methods and data used for parameter estimation, estimation of occupational exposure, and imputation of missing values in exposure indices derived from the kinetic model.