Effect of uncertainty components such as recalibration on the performance of quality control charts

Abstract

Background. Uncertainty components (recalibration, new reagent lots, etc.) may be the source of random changes in the level of quality control (QC) values, thus causing false alarms. We propose a method for reducing false alarms. Material and methods. Daily QC measurements of 22 biochemical quantities were recorded over 5 months, while methods and analyser showed no signs of malfunctioning. Each time series of QC values was subdivided according to reagent lot, electrolyte diluent or disposable electrode used. Analysis of variance (ANOVA) was used to examine whether mean levels changed significantly between subseries. For each quantity, the entire time series and each subseries were examined for autocorrelation. An X chart and an exponentially weighted moving average (EWMA) chart (or EWMAST [EWMA for stationary processes] in autocorrelated series) were calculated from the first 50 values of each series or subseries and applied to the whole series. Values falling outside the three standard deviation control limits were noted. Results. In 22 out of 24 time series, the mean level differed significantly between subseries. These changes caused spurious autocorrelation. However, in 51.3 % of all subseries a significant autocorrelation could also be demonstrated. In total, 5.6 % and 29.1 % of all time series values fell outside the control limits of the X charts and the EWMA or EWMAST charts, respectively. These percentages were significantly reduced to 0.44 and 0.70 when the subseries‐specific control charts were used. Conclusions. The mean level may change because of recalibrations or change of electrolyte diluent lot with subsequent false alarms. False alarms may be significantly reduced by revising QC charts when significant changes occur.

• Autocorrelation
• control charts
• EWMA charts
• EWMAST charts
• quality control
• statistics
• uncertainty components

Acknowledgements

We express our thanks to Hanne Brun, chief technologist in the Clinical Biochemical Laboratory at Storstrømmens sygehus in Nykøbing Falster for her assistance with data acquisition. We also thank Christian Gluud at the Copenhagen Trial Unit, Center for Clinical Research Center, H:S Rigshospitalet and Will Guthrie at the NIST for their comments and suggestions.