A Technique for the Objective Assessment of Routine Analytical Methods in Clinical Laboratories Using Pattern Recognition

Abstract

A technique is presented to assess objectively the reliability of analytical methods used routinely in clinical laboratories. From the Netherlands National Coupled External/Internal Quality Control Programme information can be gathered about the performance of routine analytical methods. The performance of a method in a trial is described by four ‘features’: the accuracy of a method; its day to-day precision; its susceptibility to give erroneous results (eg, extreme bias, drift, extreme week-to-week variations); and its susceptibility to give systematic errors for different laboratories. These four features obtained in a trial for a given analytical method determine the position of a ‘pattern’ in the four-dimensional space. The results of six trials discussed in this paper provided six ‘patterns' per analytical method. Using pattern recognition techniques, clusters of patterns were detected in the four-dimensional space. A weighting procedure revealed the relative importance of the various features for discrimination between the detected clusters. For various blood components, different features are of importance for this discrimination. Patterns belonging to the same clusters appeared to be patterns of the same (or comparable) analytical methods; thus analytical methods could be distinguished from each other. The means of the feature values of the patterns in a cluster determine the quality of that cluster. Thus the quality of an analytical method can be objectively assessed. Some tentative conclusions on the validity of analytical methods are given.