Keywords: Mixed effects models, multi-site studies
The Center for Veterinary Medicine (CVM) has consistently used mixed models to evaluate veterinary drugs. Animal drugs are developed to address animal production enhancement as well as for therapeutic needs. The mixed model methods often used at CVM are anchored in agricultural experimental design, including animal science, agronomy, and horticulture. The concept of site as a random effect is the basis for making broad-based inferences important in evaluating drugs over a diverse population. Other random effects generating mixed models within a site are blocks and multiple subjects in an experimental unit.
The decision to include site as a random effect in the mixed model has the biggest impact on CVM. Because animals are assigned to treatment within site, it is appropriate to include site and site interactions in the statistical model. CVM considers sites to be a random effect and includes site and associated terms as random effects in the statistical model. This strategy has its basis development of models used for breed enhancement, therapeutic use, and nutritional requirements in animals at locations (sites) across the country. The inclusion of site as a random effect has as a goal to provide a broad-based evaluation that allows inference about drug effectiveness to most animals receiving a drug regardless of the their location (weather effects drug response), breeds, feed (pasture or processed).
R. A. Fisher introduced the concept of blocking for agricultural experiments. The purpose of blocking was to identify, estimate, and remove from the hypothesis test a source of variability that had an impact on the evaluation of the effect of the treatment on the primary variable. The use of blocking on animal characteristic, introduces the simplest mixed model used at CVM. A very simple experiment could include treatment as a fixed effect and blocks as a random effect.
As with any drug, an approval must be based on the conditions of use. For most production animals from fish to cows, that means they are treated or housed in groups. Drugs may be added to water or feed for administration to the pen, cage, or tank and not individual subjects. In this context, there is the possibility of including two random effects in the model to represent the pen-to-pen variability and the animal-to-animal variability with in the pen.
The assumptions made regarding random sites, blocks, and experimental units and their impact on statistical analyses at CVM will be furth