An evaluation of statistical process control techniques applied to blood component quality monitoring with particular reference to CUSUM

Abstract

Statistical process control (SPC) is used to monitor the performance of blood component collection and production processes in the UK and elsewhere. The sensitivity of the applied technique(s) needs to be matched to the clinical importance of the parameter being monitored such that significant deviations in the process mean and/or variability of critical parameters (e.g. the leucocyte content of leucodepleted components) are detected and investigated immediately. This study assessed the sensitivity and specificity of a range of techniques for variable and attribute (proportion non-conforming) data. Comparison was based on a range of simulated and 'live' blood component quality monitoring data including X/R, cumulative sum (CUSUM) procedures, the scan statistic and np charts. X/R and CUSUM could detect shifts of two standard deviations in the process mean within 5 days. Current leucocyte count data (substantially skewed even after log transformation) was found to be better suited to attribute analysis. CUSUM alone was able to detect shifts on the same day when based on 20 or more samples and achieved acceptable specificity. CUSUM procedures for proportion non-conforming can usefully augment existing X/R techniques for leucodepletion monitoring, provide valid control limits and the required sensitivity. The scan statistic and 'np' charts offered no obvious advantages.