In toxicity testing, each animal may have to be viewed as a surrogate for millions of people and should be examined as thoroughly as a human patient. However, there are many differences between human diagnosis and animal toxicity testing. Human diagnosis is primarily based on anamnesis, symptoms, and utilization of a huge database of diseases. However, in animal toxicity testing, clinical and anatomical pathology data are usually a primary source of toxicity information, even though the positive endpoints are generally not known in advance and the number of positive toxicity endpoints may be numerous. This situation will generate at least 2 practical problems in clinical pathology testing: (1) how to preselect test items without precise knowledge of toxicity endpoints and (2) how to handle multiple data sets for toxicity detection. The latter includes issues of inflation of the overall false-positive rate and multicomparison problems. A "disease" called "significantosis" and a concept of integrated interpretation of multiple biologically related items to avoid false-positive judgments and unnecessary censoring of meaningful outlier data are briefly discussed. In general, toxicity tests are quite exploratory and the endpoints are unknown and multiple, so the procedures for data interpretation should be determined on a case-by-case basis. Construction of toxicity entity-oriented databases may be a requirement for further refinement of toxicity study interpretation.
Refinement, Reduction, and Replacement of Animal Toxicity Tests by Computational Methods