Statistical issues in the analysis of the long-term carcinogenicity bioassay in small rodents: an empirical...

Data on 49 randomly selected studies from the NCI/NTP Carcinogenesis Bioassay Program were reanalyzed using four statistical decision rules to classify substances as either being negative or falling into one of three categories indicating increasing evidence of oncogenicity. The data available for analysis were the crude marginal counts of numbers of animals with specified lesions, as well as the number of animals surviving at the time the studies were terminated. Statistical analysis was based primarily on the Cochran-Armitage test for linear trend in proportions, with and without the use of historical control information. If only concurrent controls were used, classifications of carcinogenicity obtained in between 34 and 57% of the studies, depending on the decision rule used. The incorporation of historical control information into the Cochran-Armitage test statistic led to almost universal findings of carcinogenicity. The data base assembled here was used to estimate false negative and false positive rates for each of the four decision rules.