Use of Tumour Lethality to Interpret Tumorigenicity Experiments Lacking Cause‐Of‐Death Data

SUMMARY A popular statistical test for comparing control and exposed groups in tumorigenicity experiments requires, for each animal found to have a tumour at death, an indication of whether the tumour caused death. Two other popular tests do not require cause‐of‐death information but assume that the tumour being investigated is either instantly lethal or non‐lethal. For lack of alternatives, these two tests are also routinely used for tumours of intermediate lethality because cause‐of‐death information is not available in most experiments. However, when the risks of death from non‐tumour causes in the control and exposed groups differ, both tests are biased for tumours of intermediate lethality and can give very different indications about the tumorigenicity of the compound being tested. to help to resolve some of the problems with the interpretation of the non‐lethal and rapidly lethal tests, we derive an incomplete data analogue of the cause‐of‐death test for situations where the cause is not available, and with examples illustrate its use in the interpretation of tumorigenicity experiments.