Statistical power considerations show the endocrine disruptor low-dose issue in a new light

The endocrine disruptor field has been vexed by difficulties in reproducing various claims of effects at unusually low doses. In previous analyses, variations in control responses from experiment to experiment and problems with observing effects in positive controls have been identified as possible explanations of the resulting impasse.

In this article, we argue that both of these viewpoints fail to take sufficient account of the problems that exist in estimating low effects and low-effect doses. We have carried out post hoc power analyses on selected published data to illustrate that claims of low-dose effects (or their absence) are often compromised by insufficient statistical power of the chosen experimental design.

We demonstrate that low-dose estimates such as the no observed adverse effect levels derived from statistical hypothesis-testing procedures are dependent on the specific experimental conditions used for testing. Thus, below the statistical detection limit of the experiment, the presence of effects can neither be proven nor ruled out. Common practice is to attempt to establish "doses without effect." However, low-dose estimations in the endocrine-disruptor field could be improved if decisions regarding the toxicologic effect size of relevance formed the starting point of testing procedures. Statistical power considerations could then reveal the resources necessary to demonstrate effect magnitudes of concern.