Analytical performance, reference values and decision limits. A need to differentiate between reference intervals and decision limits and to define analytical quality specifications

作者：Petersen Per Hyltoft^*; Jensen Esther A; Brandslund Ivan

来源：Clinical Chemistry and Laboratory Medicine, 2012, 50(5): 819-831.

DOI：10.1515/cclm-2011-0844

• action limits
• analytical bias
• analytical imprecision
• analytical quality specifications
• assessment of analytical quality
• choice of reference individuals
• creation of analytical quality
• use of patient median in control

With the increasing use of decision limits (action limits, cut-off points) specified for a number of analytical components in diagnosis and for action in critical situations, formulated in national or international recommendations, the traditional interpretation of reference intervals has been uncertain, and sometimes the two concepts are being mixed up by incorporating risk calculations in the reference intervals. There is, therefore, a need to clarify the two concepts and to keep them definitely separated. Reference intervals are the 95% limits for the descriptions of the distributions of the values of analytical components measured on reference samples from reference individuals. Decision limits are based on guidelines from national and international expert groups defining specific concentrations of certain components as limits for decision about diagnosis or well-defined specific actions. Analytical quality specifications for reference intervals have been defined for bias since the 1990s, but in the recommendations specified in the clinical guidelines analytical quality specifications are only scarcely defined. The demands for negligible biases are, however, even more essential for decision limits, as the choice is no longer left to the clinician, but emerge directly from the concentration. Even a small bias will change the number of diseased individuals, so the demands for negligible biases are obvious. A view over the analytical quality as published gives a variable picture of bias for many components, but with many examples of considerable bias which must be critical - yet no specifications have been stipulated until now.

• 出版日期2012-5