A bootstrap analysis of four in vitro short-term test performances.

The present analysis is aimed at estimating the confidence intervals of a number of association measures that describe the relationships of 4 in vitro short-term tests with rodent carcinogenicity, as well as with each other. The measures considered were: sensitivity, specificity and accuracy of the short-term tests with respect to chemical carcinogens, and performance dissimilarity indices (Hamming distances). The analysis refers to Salmonella, mouse lymphoma L5178Y cell mutation, chromosomal aberrations and sister-chromatid exchanges in Chinese hamster ovary cells, and is based on the data generated in the frame of the U.S. National Toxicology Program (NTP). It exploits the properties of a statistical technique, called bootstrap, to derive from only one sample of chemicals the variability intervals of the associations that the biological systems (mutagenicity assays and rodent carcinogenicity) would show in the 'universe' of the chemical compounds. The combination of the bootstrap technique with multivariate statistical methods pointed to a remarkable robustness and reliability of the information derived from the NTP data base, and provided descriptive insights into the data.