The predictivity of animal bioassays and short-term genotoxicity tests for carcinogenicity and non-carcinogenicity to humans.

The successful use of surrogate tests to predict whether a chemical may be carcinogenic to humans requires that the tests be both sensitive (few false negatives) and specific (few false positives). To assess specificity, results for non-carcinogens must be compared. Although no chemicals have been definitively shown not to cause cancer in humans, we have identified 29 chemicals for which some evidence of non-carcinogenicity exists in evaluations by the International Agency for Research on Cancer. Twenty of these probable non-carcinogens have been tested for rodent carcinogenicity in animal bioassays; 19 were positive and only one was negative, indicating that the specificity of animal bioassays is low. The sensitivity of animal bioassays, however, is very high: all definite human carcinogens adequately tested were positive. Most short-term tests which measure genotoxicity or transformation also had low specificity; however, four tests gave predominantly negative results for probable human non-carcinogens as well as predominantly positive results for definite human carcinogens. These results are based on comparison of small numbers of chemicals, but do suggest the need for more investigation of the relationships of genotoxicity and rodent carcinogenicity to carcinogenicity in humans.