Statistical Properties of a Model-Free Approach to Low-Dose Extrapolation

Estimates of risk associated with exposure to low doses of carcinogenic substances are generally obtained by linear extrapolation of laboratory results obtained at higher doses at which risks can be estimated directly. In this paper a model-free approach to linear extrapolation (MFX) is described, and compared with the linearized multistage model (LMS) using computer simulation. In general, estimates of low dose risk based on MFX and the LMS are somewhat comparable. In some cases, notably convex dose response curves with low background, estimates of risk based on MFX were typically a factor of two-fold or more higher than those based on LMS. In these cases, however, MFX was more likely to satisfy the nominal coverage probability than was LMS.