The use of historical data for identifying biologically unimportant but statistically significant results in genotoxicity assays.

The definition of a negative result is a problem in genetic toxicology. Here we suggest that a result may be considered biologically unimportant (negative) if it falls within the limits of variation usually found in the negative controls of the particular test. To determine 'usual' variation, we have set 95% confidence limits on three indices of variation, calculated from historical values for duplicate negative control data from several genotoxicity tests. These tests showed four characteristic types of response and the appropriate index of variability was determined for each. Where there was little test-to-test variation in true mean (micronucleus test and metaphase analysis), confidence limits set on the overall distribution of negative controls were the best index of variability. In other assays (Ames, yeast and mouse lymphoma), there was considerable test-to-test variation so that differences between, or ratios of, the members of control duplicates were the preferred measure of variability. This approach can define what is biologically unimportant in terms of the test system. However, no inference can be drawn as to potential importance. Thus the main use is the removal of the positive 'label' from statistically significant results which fall within the usual range of spontaneous variation for the assay under consideration.