Statistical designs in combination toxicology: a matter of choice.

Combination toxicological studies with statistically designed experiments do not often elaborate on the qualities of the design. Therefore, it may seem that the design used is the only one to meet the objectives of the study. This article demonstrates, on the contrary, that there are often several possibilities. Three recent toxicological experiments, featuring economy of experimentation, robustness against inhomogeneity of experimental material and increasing precision of results, respectively are used for illustration. The alternatives differ statistically in the number of distinct effects for which information can be obtained or in the safeguards to counter both inhomogeneity of material and incorrect models. Practically they differ in complexity of experimental conduct. It is concluded that the choice between the alternatives is a result of balancing practical complexity of the design against informational gain. Continuous co-operation between toxicologists and statisticians is needed to strike the balance again and again.