Tests for Differences in Tumor Incidence Based on Animal Carcinogenesis Experiments

Abstract

In animal carcinogenesis studies, statistical methods that compare the production of occult tumors should compare tumor incidence rates. In general, the incidence rate cannot be identified even approximately without both survival and sacrifice information. If information from frequent sacrifices is available, however, a function that approximates the incidence rate can be identified without assumptions about the relationship between tumor incidence and death. Some earlier proposals that based comparisons on tumor prevalence rates and other functions also compared tumor incidence rates under special conditions, but these tests can be biased when the conditions are violated. Tests that compare functions approximating tumor incidence rates can be developed using survival/sacrifice information, but these tests may be seriously inefficient in special cases in which sacrifices are not necessary to identify the tumor incidence rate.

Key Words:

• Survival experiments
• Survival/sacrifice experiment
• Tumor prevalence
• Log-rank tests
• Mantel–Haenszel test
• Identifiability
• Competing risks