A survival-adjusted quantal response test for comparing tumour incidence rates

Summary.  The paper presents a case-study of skin fibromas among male rats in the 2-year cancer bioassay of methyleugenol that was conducted by the US National Toxicology Program (NTP). In animal carcinogenicity experiments such as this one, tumour rates are often compared with the Cochran–Armitage (CA) trend test. The operating characteristics of the CA test, however, can be adversely affected by survival differences across groups and by the assumed dose metric. Survival-adjusted generalizations of the CA test have been proposed, but they are still sensitive to the choice of scores that are assigned to the dose groups. We present an alternative test, which outperforms the survival-adjusted CA test which is currently used by the NTP to compare incidence rates. Simulated data from a wide range of realistic situations show that the operating characteristics of the test proposed are superior to those of the NTP's survival-adjusted CA test, especially for rare tumours and wide logarithmic spacings of the dose metric.