Perspective: common errors in dose–response analysis and how to avoid them

Abstract

Abstract Dose–response experiments are conducted to determine the toxicity of chemicals on organisms. The relationship between dose and response is described by different statistical models. The four‐parameter log‐logistic model is widely used in pesticide sciences to derive biologically relevant parameters such as ED 50 and resistance index (RI). However, there are some common errors associated with the calculation of ED 50 and RI that can lead to erroneous conclusions. Here we discuss five common errors and propose guidance to avoid them. We suggest (i) all response curves must be fitted simultaneously to allow for proper comparison of parameters across curves, (ii) in the case of nonparallel curves absolute ED 50 must be used instead of relative ED 50 , (iii) standard errors or confidence intervals of the parameters must be reported, (iv) the e parameter in asymmetrical models is not equal to ED 50 and hence absolute ED 50 must be estimated, and (v) when the four‐parameter log‐logistic model returns a negative value for the lower asymptote, which is biologically meaningless in most cases, the model should be reduced to its three‐parameter version or other types of model should be applied. The mixed‐effects model and the meta‐analytic approach are suggested as appropriate to average the parameters across repeated dose–response experiments. © 2021 Society of Chemical Industry

Topics & Concepts