The assessment of low-dose carcinogenicity.

The assessment of carcinogenic risk from very low doses of substances to which the population is exposed depends mainly on animal experiments. A variety of statistical models has been proposed for extrapolation to low doses from the much higher doses administered to animals. The most conservative approach uses a multistage model, and is almost equivalent to linear extrapolation. Cornfield has argued against a conservative approach, but linear extrapolation may also be close to the 'best' solution. Species conversion and specification of allowable risks are very difficult problems, and some form of cost-benefit analysis seems to be required.