Why dose-response relationships are often non-linear and some consequences.

Given the dependence of many risk assessments on the assumption of linearity of dose-response relationships in human populations, we analyze the circumstances likely to lead to non-linearity and test our hypothesis of the high prevalence of non-linearity by examination of recent literature. METHODS. The analysis of Bross, based on whether irradiated cells die or can manifest malignancy, leads him to generalize that if a single exposure can have one of two (or more) countervailing outcomes, non-linearity of dose-response will result. We list four other common mechanisms which would have similar effects: symptom-stimulated withdrawal from exposures to respiratory irritants; certain aspects of the "healthy worker effect" (especially its obverse-withdrawal from the work force due to illness); selected consequences of the competing risks of long-term disease; and shift in relative strength among multiple independent variables. We then examine recent literature to see how often reports of linear, monotonic, and non-monotonic dose-response relationships occur and discuss the likelihood of countervailing alternate outcomes in selected examples. RESULTS. Non-linear and linear relationships are about equally frequent. Under circumstances where countervailing outcomes are probable, dose-response relationships should be non-linear and often are. These conditions may also lead to non-linear difference equations, which may manifest "chaotic" attributes. CONCLUSIONS. Regulations and policies cannot be routinely derived on the basis of extrapolating linear dose-response relationships for human populations. Although our analysis is oriented principally to epidemiology, similar considerations apply to toxicological studies as well.