Comparison of routine analytical methods in the Netherlands for seven serum constituents using pattern recognition.

Routine analytical methods for seven serum analytes (calcium, chloride, cholesterol, glucose, inorganic phosphate, urate, and urea) are assessed using data from the Netherlands coupled external/internal quality control program. From the results of a trial each method can be described by four features: measures of bias, between-day precision, tendency to give erroneous results, interlaboratory variance. These four features of each trial determine a vectorpoint in the four-dimensional space for a particular method. From 12 trials a maximum of 12 vectorpoints per analytical method was obtained. Pattern recognition techniques allowed the detection of clusters of vectorpoints. Analytical methods having vectorpoints classified in different clusters perform differently. The mean feature values of the vectorpoints forming a cluster determine the quality of that cluster. A weighting procedure reveals the importance of the respective features for discriminating the clusters. For all of the seven analytes, clusters of vectorpoints were found. Different features appeared to contain discriminatory power for different analytes. For six analytes (calcium, chloride, cholesterol, glucose, inorganic phosphate, and urea) an analytical method was found to classify predominantly in the qualitative best cluster. One analytical method for the determination of chloride and one for glucose, inorganic phosphate, and urea did not cluster at all.