Improving the power of long term rodent carcinogenicity bioassays by adjusting the experimental design

Since long term rodent carcinogenicity studies are used to test a very large number of potential tumor endpoints, finding a balance between the control of Type 1 and Type 2 error is challenging. As a result, these studies can suffer from very low power to detect effects of regulatory significance. In the present paper, a new design is proposed in order address this problem. This design is a simple modification of the existing standard designs and uses the same number of animals. Where it differs from the currently used designs is that it uses just three treatment groups rather than four, with the animals concentrated in the control and high dose groups, rather than being equally distributed among the groups. This new design is tested, in a pair of simulation studies over a range of scenarios, against two currently used designs, and against a maximally powerful two group design. It consistently performs at levels close to the optimal design, and except in the case of relatively modest effects and very rare tumors, is found to increase power by 10%-20% over the current designs while maintaining or reducing the Type 1 error rate.