Moving Average for Continuous Quality Control: Time to Move to Implementation in Daily Practice?

To the Editor: Recently, Ng et al. described a new method for optimization of moving average (MA)1 QC procedures (1). The accompanying editorial mentioned that, during the last 50 years, slow but continuous improvements have been made in the understanding and methodology of MA in the move toward continuous analytical quality assurance (2). Although these improvements are being made, general implementation of MA for continuous QC on clinical laboratories has failed and many laboratories are struggling with the implementation and application of MA QC. In this letter, we address several steps that we consider to be important to support a more general implementation of MA as a continuous QC instrument in medical laboratories. Most improvements that really affected the use of MA in clinical laboratories originate from the 1970s and 1980s. For example, the algorithms described by Bull et al. in 1974 are still the basis of the application of MA today in most, if not all, hematology analyzers (3). Interestingly, Bull et al. stated that, because their findings were based on visual inspection of MA patterns, future MA research should focus on developing objective measures of MA performance (3). Recently, 2 methods have been described that allow more objective and realistic insight into MA performance. Ng et al. (1) described an MA optimization method that used the average number of patient samples affected until error detection (ANPed), and we reported the use of bias detection curves and MA validation charts for the selection of optimal MA settings (4). Both methods meet the requirement of a low “manageable” number of false MA alarms. In our opinion, this is an essential MA requirement for the introduction of MA as a continuous QC instrument in daily practice. Optimized MA procedures allow continuous analytical QC with fast detection of systematic errors and detection of temporary assay failures (5). Fig. 1 presents an MA alarm case that illustrates a clinically relevant temporary assay failure detected by MA, which was not detected by regular scheduled internal QC measurements due to the temporary nature of the assay failure. This case illustrates the added value of continuous analytical QC by MA. However, implementation of MA as a continuous analytical QC instrument in clinical laboratories is hampered by the complexity of obtaining optimal MA procedures for each specific laboratory. A major obstacle is that no commercial tools are available to generate optimal MA settings. By applying new MA optimization methods, both Ng et al. and our group were able to implement MA for continuous analytical QC in daily practice (1, 5). Therefore, it is time to make these new MA optimization methods available to clinical laboratories, in an understandable and easy-to-use format. MA is plotted in an accuracy plot. The green area represents the reference range of the MA and is bounded by the upper 1 and lower −1 MA control limits [van Rossum and Kemperman (5)]. One y-axis unit represents half of the MA reference range and is referred to as the dimensionless quantity MA reference index. Blue dots represent generated MA values and red dots with blue circles represent MA values outside the control limits. GLIMS (MIPS) LIS software is used for MA management and graphical presentation. MA sodium alarm case: the first MA sodium alarm (September 12, 2016) was followed up according to our standard procedure. The alarm was due to a sodium ISE failure that seemed to be resolved without intervention; internal QC results were within limits and reproducibility (n = 20) of 1 sample revealed no outliers, and the CV was appropriate. Reanalyzing the patient samples (as part of the MA alarm workup) resulted in 30 sodium results that were considered to have a clinically significant error. Subsequently, these results were corrected and the physicians were informed. Despite extensive troubleshooting, the cause of the ISE failure was not identified and the MA alarm was thought to have resulted from temporary contamination of a mixing cup, or sample probe. Two days after the first alarm, a second sodium MA alarm occurred. The manufacturer was contacted again and (after about 4 h of troubleshooting) it was established that a valve in the ISE unit did not always close properly. Assay failure was not detected by routine internal QC because of the temporary nature of the assay failure. After obtaining optimal MA procedures, the next step is to implement these in the clinical laboratory. Most laboratories have an option to use either analyzer software, middleware, laboratory information system (LIS), or other dedicated software to perform MA. We recently performed an MA trial using our LIS for MA management, and experienced some MA management issues that are not unique to the LIS software used (5). One reason we chose our LIS for MA management was that our LIS allowed MA calculation after each new assay result and MA calculation frequency did not need to depend on the MA calculation algorithm. Using our optimized MA settings for sodium we always were able to detect a ≥6% bias within 10 patient results. Waiting for more than 10 patient results would result in an unnecessary MA alarm delay (4, 5). Furthermore, before starting the MA trial some additional features were incorporated in the LIS to allow practical management of MA and MA alarms. These included the option to exclude specific patients or samples from the MA calculation (e.g., nonpatient materials and specific patients with confirmed extreme results) and graphical presentation of MA values in an accuracy graph as presented in Fig. 1(5). Also, if MA protocols are properly designed, they should yield a low number of false MA alarms and as such all MA alarms must be investigated promptly. We identified some practical consequences during our MA trial: MA alarms should be generated real time and immediately detected by laboratory staff; a protocol should be available that clearly describes how to determine the relevance of the MA alarm (also during evening, night, and weekend shifts); after an MA alarm workup, an MA reset is helpful to prevent subsequent false MA alarms (5). The MA management issues encountered during our trial and the additional software features we needed to incorporate in the LIS for MA management purposes emphasize the need for improvements in the available MA management software. Addressing these issues is the necessary final step to allow a more general implementation of continuous QC by MA in clinical laboratories. moving average laboratory information system.