Statistical design and analysis of mutation studies in transgenic mice.

We have been working on identifying sources of variability in data from transgenic mouse mutation assays in order to develop appropriate statistical methods and designs for routine studies. Data from our lab and elsewhere point to the presence of significant animal-to-animal variability, which must be taken into account in statistical hypothesis tests. Here, the usual Cochran-Armitage (CA) test for trend in mutant frequencies, which takes the transgene as the experimental unit, and a generalized Cochran-Armitage test (GCA), which takes the animal as the experimental unit, are contrasted in computer simulations that help to quantify the differences between these statistical tests. The simulations report the statistical power of each test to detect treatment group differences, and their type I error rates. We find in general that the GCA test performs poorly compared to the CA test when it is appropriate to take the transgene as the experimental unit, and the study also uses a small number of animals. However, the CA test performs poorly in small group-size studies when the animal is the appropriate experimental unit. Extensions of the computer simulations allow for identification of cost-effective experimental designs. The results emphasize that the benefits of using additional animals in these mutation studies can be realized without substantial increases in costs. Here we illustrate the methods for liver studies in our lab. These methods can be used to derive optimal experimental designs for any combination of spontaneous mutant frequency and animal-to-animal variability.