Statistical Considerations in Threshold Identification through Toxicological Experiments

抄録

Toxicologists are accustomed to using methods of hypothesis testing in order to recognize the existence of target toxicity, on the basis of experimental data. However, in principle, it is impossible to identify the threshold via hypothesis testing in toxicological experiments since the probability of false-negative decisions cannot be determined in the context of hypothesis testing. When a mechanism for producing a threshold is hypothesized from a toxicological (or biological) perspective, and it is mathematically formulated as a dose-response relationship, statistics may be helpful in evaluating the existence (or non-existence) of the threshold, that is, in addressing the threshold identification problem. Consequently, it is important to select a model from a particular set of mathematical dose-response functions. Although the Akaike information criterion (AIC) or its modification is often used as a criterion index in the selection of the optimum model, their effectiveness in the threshold identification problem have not been validated. In these circumstances, the determination of a practical threshold using in vitro experiments may be an alternative to the identification of a “true” threshold, provided an appropriate in vitro assay affords a large scale experiment. Although quite difficult, it is important to maintain the homogeneity of experiments throughout the entire procedure. Only after establishing the experimental skills, the statistics may provide appropriate statistical methods that satisfy the requirements of toxicologists.