[Evaluation of the diagnostic efficacy of a radioimmunologic assay: the example of thyroid tests]

Optimization of the utilization of a new radioimmunoassay (RIA) requires the identification of explicit criteria regarding when this new RIA should be used. The first step consists in the definition of the normal range on a well defined learning sample of patients. The authors describe the methods which can be used; mean and standard deviation, ROC analysis, Bayes theorem. These limits must be therefore validated on a control sample. The second step of this diagnostic evaluation must estimate the impact of the new RIA on the medical decision procedure. The historical account of thyroid function testing illustrates these different points.