Reconsidering sufficient and optimal test design in acute toxicity testing.

In dose-response analysis, regression analysis and hypothesis testing are the main tools of choice. These methods, however, have specific requirements for the design of acute toxicity experiments. To produce meaningful results, both approaches require a constant exposure concentration over the duration of the test, and regression analysis makes an additional demand for at least two doses with partial mortality at the end of the test. These requirements, however, result from the limitations of the statistical techniques, which only use the observations at the end of the test. In practice, most standard protocols for acute testing prescribe that observations are made at several points in time (often daily). In this contribution, I demonstrate how dynamic modelling can make use of this information to produce robust estimates of LC50 as function of time, with confidence intervals, from data sets that violate the requirements for standard dose-response analysis. This form of modelling invites an entirely different, more flexible, view on experimental design, which could lead to a more efficient use of test animals and, at the same time, a stronger support for environmental risk assessment as well as the science of ecotoxicology.