Suitability of control materials. General principles and methods of investigation.

Abstract

Abstract We propose methods for characterizing the behavior of quality-control specimens. Candidate quality-control specimens and authentic patients' specimens were analyzed by various methods. Patients' specimens were chosen to be fully representative of those encountered, including subsets from persons who were healthy, had defined disease states, were in therapy, or whose specimens were lipemic, icteric, etc. The analytical methods chosen include those most commonly used as well as reference analytical methods. Procedures for characterizing the behavior of patients' specimens and candidate quality-control specimens are proposed and their applicability is demonstrated. The linear ratio method is a univariate graphical approach in which differences in accuracy among methods for any specimen or group of specimens are each displayed on a linear scale. Correspondence analysis is a descriptive multivariate statistical technique that allows both the specimens and the analytical methods to be characterized. The statistical techniques, in our application, allow the behavior of quality-control specimens to be assessed with respect to authentic patients' specimens without influencing the assessment process. Correspondence analysis provides a graphic representation by projecting both the specimens and the analytical methods on factorial planes. The appropriateness of te behavior of a quality-control specimen may be inferred from its position relative to those of authentic patients' specimens. These statistical techniques also provide some information regarding the specificity of analytical methods.