Exposure-response assessment methods have shifted towards more quantitative approaches, with health risk assessors exploring more statistically driven techniques. These assessments, however, usually rely on one critical health effect from a single key study. Categorical regression addresses this limitation by incorporating data from all relevant studies – including human, animal, and mechanistic studies – thereby including a broad spectrum of health endpoints and exposure levels for exposure-response analysis in an objective manner. Categorical regression requires the establishment of ordered response categories corresponding to increasingly severe adverse health outcomes and the availability of a comprehensive database that summarizes all data on different outcomes from different studies, including the exposure or dose at which these out-comes are observed and their severity. It has found application in the risk assessment of essential nutrients and trace metals. Since adverse effects may arise from either deficient or excess exposure, the exposure-response curve is U-shaped, which provides a basis for determining optimal intake levels that minimize the joint risks of deficiency and excess. This article provides an overview of the use of categorical regression fit exposure-response models incorporating data from multiple evidence streams. An extension of categorical regression that permits the simultaneous analysis of excess and deficiency toxicity data is presented and applied to comprehensive databases on copper and manganese. Future applications of cat-egorical regression will be able to make greater use of diverse data sets developed using new approach methodologies, which can be expected to provide valuable information on toxic responses of varying severity.
Keywords: categorical regression risk assessment; joint excess and deficiency model; logistic regression; nonsymmetric U-shaped dose-response curve.