Patient-Based Real-Time Quality Control Techniques in Clinical Laboratories Quiz

Analytical Performance of Moving Averages and Moving Medians for Skewed Distributions

• Moving averages and moving medians were used for skewed distributions, as well as for symmetrical distributions.
• Normal QQ plots were used to assess the distribution of patient results.
• Truncation limits were set based on deviation from the linear portion of the Q-Q plot to minimize the influence of diseased population on the moving average.
• Bias detection curves for symmetrical distributions and skewed distributions were examined for different PBRTQC approaches.
• Error detection varied depending on the analyte, PBRTQC protocol used, and whether the bias was positive or negative.
• Detection of bias was difficult for certain analytes, requiring a large number of samples to detect shifts.
• The addition of wide truncation limits had little benefit over a simple moving average with no truncation, except for magnesium.
• Error detection for right skewed analytes was generally poorer compared to symmetrically distributed analytes.
• Box-Cox transformation of the data markedly improved the performance of simple moving averages for several analytes.
• The performance of the moving median was also improved by the Box-Cox transformation of the data.
• The study assessed the retrospective detection of a real instrument failure affecting creatinine and urea.
• Different algorithms were applied to patient results retrospectively, and error detection time of the algorithms was summarized in Table 2.

Analytical Performance of Patient-Based Retrospective TQC Protocols

• The study compares different patient-based retrospective TQC protocols for skewed and symmetrical patient data distributions.
• The protocols aim to minimize the influence of diseased patients on the moving average (MA) and detect bias in patient results.
• For skewed distributions, truncation limits are used to censor patient data at the inflection point of the frequency distribution curve.
• The detection of bias varies based on the analyte, protocol used, and the nature of the bias (positive or negative).
• Negative bias in albumin and positive bias in bicarbonate are difficult to detect, requiring more than 300 samples for each protocol.
• Error detection is better for negative shifts in bicarbonate, with an average number of patient samples until error detected (ANPed) of less than 50 samples.
• For symmetrical distributions, adding wide truncation limits offers little benefit over a simple MA with no truncation, except for magnesium.
• Excluding abnormal results with narrow truncation limits paradoxically increases ANPed for larger shifts.
• For right skewed distributions, error detection is generally poorer than for symmetrically distributed analytes, and differences between protocols are more marked.
• Box-Cox transformation of the data improves the performance of the MA and moving median for detecting negative shifts in certain analytes.
• The study retrospectively assessed the detection of a real instrument failure affecting creatinine and urea, with the MA and moving median showing improved detection.
• Table 2 summarizes the error detection time of the different algorithms for the real instrument failure affecting creatinine and urea.