A modified statistical approach for the detection of outlying values in external quality control: Comparison with other techniques

Laboratory results submitted to External Quality Assessment Schemes (EQAS) are evaluated against the arithmetic mean ( m) and standard deviation (sd) of the results from other participants assuming a normal distribution of individual results. The sample statistics m and sd can only be reliable estimates of the theoretical mean (μ) and standard deviation (σ), if they are not distorted by outlying values. Outliers are commonly detected by defining a confidence interval m ± k sd, typically with k-values fixed at 3.0 or 3.2. The objective of our study was to prove the deficiency of the use of these fixed limits in large-scale EQAS, because it results in an unacceptable increase of outlier rate and hence in a serious distortion of the estimate of σ. We have proposed a modified approach using a variable k-value which is dependent upon sample size and keeping the significance level constant at 5%. The influence of different decision limits on the number of outliers and on the sample statistics m and so was compared using data from the Belgian National EQAS obtained in 6 consecutive surveys.