Statistical quality control: further assessment of more-complex QC charting

Control charts are ubiquitous in modern medical laboratories. We already touched on them in the previous article in the series. There we focused on the graphs most familiar to us, the Levey-Jennings (LJ) chart and range charts. This time we go a step further and start looking at ways we can increase our sensitivity to detect process errors. In particular we are interested in detecting small deviations – around one standard deviation (1SD). Large deviations are comfortably dealt with by LJ charts, but changes in target mean of around 1SD may not be picked up by these charts. To mitigate the risk of this, and to help us identify process changes as early as possible, we can incorporate other charts such as smooth average, exponential weighted moving average (EWMA) and cumulative sum (CUSUM) to help us get stronger understanding of how the assay is performing.

Limitations of using simple charts

One small problem with traditional control charts is that they can over-interpret data and suggest out of control situations that they are not set up to detect. Charts will represent both random and systematic error (as described in Article 7; June, page 15). Can we reliably identify a trend as it develops – before it fails rules? This changes our process from monitoring to forecasting. Control chart limits are constructed using baseline random error. Biological variation may or may not be accounted for in that same composition. We only want to detect deviations, either random or systematic, that exceed these expectations.