Analysis of assay errors in drug measurements from the Heathcontrol interlaboratory quality assessment schemes

Measurements of drug concentrations from the Heathcontrol quality assessment schemes were analysed to detect the presence of intralaboratory, interlaboratory, or intermethod errors. We developed weighted least-squares regression procedures as significance tests for evaluating intralaboratory noise, curvature vs linearity, proportional errors and additive errors. The latter intercept test for additive errors was unsatisfactory and an effective alternative based on the difference between measurements and estimates of the true drug concentrations was developed. Significance of residuals was tested against the population noise, which was defined by smooth mathematical functions fitted to the standard deviation (SD) data for the drug samples. We evaluated these tests for 1,647 sets of data. Only small amounts of curvature were present, validating the linear-regression approach. Both random and proportional errors were demonstrated. The most frequent errors were additive in nature, components of which were demonstrated to be the result of intermethod differences.