Robust statistical evaluation of tumor incidences in long-term rodent carcinogenicity studies: the reformulated poly-k trend test

Mortality-adjusted tumor rates in long-term carcinogenicity rodent bioassays are commonly evaluated by means of the poly-k=3 Armitage trend test. However, this assumes exactly a linear dose-response curve and the Weibull parameter of k=3 for all tumor sites. These unrealistic assumptions can be circumvented by multiple testing across multiple possible dose-response shapes, multiple Weibull parameters, multiple effect sizes, multiple correlated tumors as well as pairwise and trend tests using the multiple marginal models approach. Based on data examples, different multiple tests are demonstrated using the CRAN R packages multcomp, tukeytrend, coin, MCPAN and multfisher.