Use and misuse of meta-analysis in Animal Science

In animal sciences the number of published meta-analyses increases with a rate of 15% per year highlighting an actual success. This current review focuses on the good practices and traps in the conduct of meta-analyses in animal sciences, nutrition in particular. The implementation of a meta-analysis is done in several phases after the definition of the study objectives. Clearly described rules and principles of traceability should be applied as soon as the publications are collected and selected with a target of meta-analysis. Then, the coding phase is essential because it determines the quality of the graphical and statistical interpretations of the database. Following this step, the study of the levels of orthogonality of factors and of the degree of data balance of the meta-design represents an essential phase to ensure the validity of statistical processing. The issue of the choice between fixed or random effect to study and to control heterogeneity is also discussed. It appears on the basis of several examples that this choice does not generally have any influence on the conclusions of a meta-analysis when the number of experiments is sufficient. Finally, reflections are presented on the potential interest of meta-analyses in the context of systemic approaches as well as to improve mechanistic modelling work.