A versatile trend test for the evaluation of tumor incidences in long-term carcinogenicity bioassays
For the evaluation of carcinogenicity bioassays a new trend test is proposed which is based on a maximum of arithmetic, ordinal, and logarithmic regression scores as well as the Williams-type contrasts for either crude proportions or more appropriate poly3-estimates for the tumor-by-time relationships. This test provides an almost appropriate power for most shapes of dose-response relationships (including for possible downturn effect at high(er) dose(s)), common signs of significance (p-value, confidence limits) and the information on the probable shape. Related software is easily available within the CRAN-packages tukeytrend, MCPAN, multcomp.
READ FULL TEXT