Publication Description
The complexities of long‐term tumorigenicity assays are discussed and statistical approaches suggested. Account must be taken of how tumors are discovered, and a reasonable distinction might be made between self‐evidencing tumors (palpable tumors and rapidly lethal tumors) and those identified only on necropsy. Tumor autolysis can pose difficulties. The nature of the experimental design must be taken into account‐the study can be completely randomized, or may involve litter‐matching, or circumstances may require that all members of a litter be treated alike. The use of a logrank procedure or a closely equivalent Mantel‐Haenszel procedure can provide a unifying approach, but modification needs to be made so as to take into account‐data on non‐self‐evidencing tumors. Censoring plays a role, whether because of eventual study termination or because losses to competing risks are treated as censored observations. Where there are several non‐zero dosage levels, the analysis can be modified according to the dosage metric used. Multiplicity of statistical tests can arise whether because a number of questions are asked or because a multiplicity of tumor sites have to be considered. Some of the kinds of questions raised have been addressed in earlier papers by various of the present authors, to which the reader is referred. Specifically new material is given on use of the jackknife so as to take into account that the unit of observation is the litter or even the group of litters caged together. Various problems were found to attend the application of the jackknife in such situations.