Tumour Incidence Rate Alternatives and the Cause-of-Death Test for Carcinogenicity
该研究探讨了致癌性研究中死因检验与肿瘤发病率假设之间的关系，发现当模型不成立时，Peto检验可能过于宽松且功效不足。
Lagakos & Louis (1988) and Burnett, Krewski & Bleuer (1989) show that the cause-of-death test of Peto et al. (1980) for carcinogenicity studies can be viewed as the approximate score test of a one-dimensional joint hypothesis about tumour prevalences and the rates of death from tumour. We relate this joint prevalence-death rate hypothesis to hypotheses about differences in tumour incidence rates. For two-sample tests, we show that when the control group tumour incidence is low, when tumours are highly lethal, or when the alternative is local, the prevalence-death rate family of alternatives is approximately equivalent to a proportional tumour incidence rate family. More generally, the prevalence-death rate null hypothesis is always equivalent to a hypothesis of equal tumour incidence rates, and each member of the family of prevalence-death rate alternatives requires the tumour incidence rates not to cross. Using computer simulation we show that, when the model does not hold, the test of Peto et al. can be anti-conservative and have poor power. We discuss the model's implications for the interpretation of tests of regression hypotheses.