[Statistical aspects of the definition of reference values]

The criteria adopted in the study of reference values have a great influence on the statistical analysis of the data. To avoid a poor statistical treatment of the data, the scope of the study must be well defined: a study to establish reference values for a chemical in the body fluids of a population is not a survey to prove relations among variables and the chemical. In this latter case, criteria to select subjects and methods are different. Confusion between the two types of surveys can lead to poor results. Reference subjects must be selected with precision, using reproducible criteria. It can be suggested to pick-up subjects from a variety of sub-groups of the population. It must be used selection criteria strictly defined to exclude from the study groups of subjects having specified characteristics (patients, resident in contaminated-zone...). The use of partition criteria is useful to know the distribution of some characteristics in the population, or to describe reference values in particular subgroups. On the contrary, partition can not be used to study relationships. Particular attention must be payed in choosing number of strata and variable of partition. When a population is analysed for reference values, the choice of subjects and relevant variables must be made after consulting a statistician.