Extrapolation in human health hazard characterization: a probabilistic approach

Abstract

A classical deterministic risk assessment often uses conservative, worst-case assumptions to estimate the possible health risk in humans. When such an assessment shows an unacceptable human health risk, a more realistic risk assessment may be needed to estimate the actual health impact in the population. A probabilistic approach accounts for variability and uncertainties in a risk assessment. This results in a better understanding of the actual risk of adverse health effects in humans. As a part of the risk assessment the goal of a risk characterization is to derive a human limit value (HLV). The HLV may be derived in a probabilistic way when probabilistic extrapolation factors (EFs) are available. The aim of this thesis is to provide risk assessors with EF distributions based on data and assumptions that can be validated rather than an arbitrary number of ten, the justification of which is based on general acceptance only. Such EF distributions may be used to perform probabilistic risk assessments. In this thesis interspecies and subchronic-to-chronic EF distributions were obtained based on No-Observed-Adverse-Effect-Level ratios as well as on benchmark dose ratios using the same dose-response data. It is concluded that using allometric scaling to body weight with a scaling power of 0.7 in addition to a species-independent interspecies EF distribution is the appropriate way to perform interspecies extrapolation. An interspecies EF distribution, which can be quite precisely estimated from pairs of test species such as rat vs. mouse, could actually be used as a surrogate for the animal-human EF distribution. In this way the problem of scarcity of human data can be circumvented. Using deterministic values for in the hazard characterization implies a piling up of worst-case assumptions. The probability of simultaneous occurrence of various worst-case assumptions is smaller than that of a single worst-case situation to occur. Therefore, the more extrapolation steps are taken into account, the higher the level of conservatism associated with multiplying deterministic values. The piling-up of worst-case assumptions can be avoided by using a probabilistic approach. In this method the benchmark dose and each EF is considered uncertain and characterized as a random variable with a distribution. The uncertainty can be propagated using Monte Carlo simulation, yielding a distribution. At this stage a conservative value is identified, if a deterministic HLV is needed. To complete the risk assessment, the probabilistically derived HLV could be compared to a human exposure level, in the usual way. However, if exposure has also been assessed using probabilistic methods, it is preferable to maintain the uncertainty in the HLV, and combine it with the distribution of the exposure, resulting in an uncertainty distribution for the risk of adverse effects in the (sensitive) human population.