A guide to the statistical analysis of long-term carcinogenicity assays

Statistical methods for analyzing data from chronic carcinogenicity bioassays are reviewed. When treatment groups do not show marked differences in tumor-free death rates, methods that compare the crude proportions of animals that develop tumors can be recommended as valid and powerful comparisons of age-specific tumor incidence rates. When the age at tumor onset can be observed, techniques of survival analysis such as the logrank test can adjust for differing tumor-free mortality rates. If tumors are strictly nonlethal, the age-adjusted comparison of tumor prevalence among dying animals compares age-specific tumor incidence rates. When tumors are occult or of intermediate lethality and when tumor-free death rates differ substantially, a variety of techniques are available, but most are computationally intensive and rely either on the collection of additional data or on additional assumptions that may not be met.