Statistical quality control: an overview of monitoring process with rules

In the previous article in this series (June, page 21) we associated required performance with the complexity of control required. Complexity is determined by the application of rules. Every rule is triggered by a particular pattern of results and thereby detects error we need to control – either random or systematic, but not both. Here, we compare single and multiple rule approaches.

Rules are often described as warning, rejection or investigate further. Rejection rules are simple to understand. If they are breached, the process must be stopped. In the past, compliance to the 1_2S rule was satisfactory. Further investigation was not required. Nowadays, further investigation can be instigated before rules are breached. This is particularly true for trend and systematic shift rules that use multiple sequential runs and allows the scientist to identify whether there is an issue prior to the actual alarm.

            The simplest and most familiar of all rules is the 1_2S rule. An internal quality control (IQC) result exceeding ±2 standard deviations (SD) from the target mean will trigger the alarm (Table 1). This is a measure of increased imprecision in the system and therefore random error. Taken alone, there is a 5% probability of a false-positive alarm. This non-specificity can be mitigated by increasing the limit to 3_S. By reducing the false-positive rate to below 1%, this significantly improves its specificity for a genuine error, but this in itself introduces concerns of clinically relevant errors being undetected. Equally, a 2_2S rule will increase the specificity of detecting a genuine shift in the process – this time suggesting a systematic change. This sacrifices the detection of increased imprecision.