Uncertainty distributions for cancer potency factors: Combining epidemiological studies with laboratory bioassays - The example of acrylonitrile

For eight chemicals or chemical mixtures with clear positive epidemiological evidence of carcinogenicity by inhalation (acrylonitrile, arsenic, benzene, beryllium, cadmium, chromium VI, coke oven emissions, and nickel), the United States Environmental Protection Agency (USEPA) uses that evidence to obtain a single best estimate of cancer potency factor. The methods used have so far been ad hoc, because of the differences in published studies, although there are common factors. In every case, the uncertainties involved in the various stages of analysis are qualitatively acknowledged, and often quantitatively estimated, but no formal attempt has been made to propagate the uncertainties. I here provide a detailed case study for acrylonitrile that (a) incorporates all estimates of uncertainty mentioned by the USEPA in their analysis and propagates that uncertainty to produce an uncertainty distribution; (b) updates the USEPA analysis to incorporate more recent epidemiological data from the same study used in their analysis. For most of the materials known to be carcinogenic to humans (through epidemiologic evidence), there are also available cancer bioassays performed on laboratory animals. If the procedures used for estimating human carcinogenic potencies from laboratory animal bioassays are to be believed in cases where there are no human epidemiological data, their evidence should also be used where there is epidemiological evidence. A consistent method of incorporating the results of both epidemiological studies and laboratory animal bioassays into a single probability distribution for a human cancer potency is here detailed, using acrylonitrile as an example for which there is positive epidemiological data. The methods are sufficiently general to allow the incorporation of any combinations of positive and negative bioassay and epidemiological data.