Internal quality control system for non-stationary, non-ergodic analytical processes based upon exponentially weighted estimation of process means and process standard deviation

• Published in: Clinical chemistry and laboratory medicine Vol. 40; no. 6; pp. 616 - 624
• Main Authors: JANSEN, Rob T. P, LAEVEN, Mark, KARDOL, Wim
• Format: Journal Article
• Language: English
• Published: Berlin Walter de Gruyter 01.06.2002; New York, NY
• Subjects: Biological and medical sciences; Biometry - methods; Chemistry, Clinical - standards; Clinical Laboratory Techniques - standards; Humans; Investigative techniques, diagnostic techniques (general aspects); Laboratories - standards; Medical sciences; Miscellaneous. Technology; Pathology. Cytology. Biochemistry. Spectrometry. Miscellaneous investigative techniques; Quality Control; Human; Statistical analysis; Estimation; Clinical biology; Quality control; Internal; Technique; Deviation
• ISSN: 1434-6621; 1437-4331
• DOI: 10.1515/CCLM.2002.107

The analytical processes in clinical laboratories should be considered to be non-stationary, non-ergodic and probably non-stochastic processes. Both the process mean and the process standard deviation vary. The variation can be different at different levels of concentration. This behavior is shown in five examples of different analytical systems: alkaline phosphatase on the Hitachi 911 analyzer (Roche), vitamin B12 on the Access analyzer (Beckman), prothrombin time and activated partial thromboplastin time on the STA Compact analyzer (Roche) and PO2 on the ABL 520 analyzer (Radiometer). A model is proposed to assess the status of a process. An exponentially weighted moving average and standard deviation was used to estimate process mean and standard deviation. Process means were estimated overall and for each control level. The process standard deviation was estimated in terms of within-run standard deviation. Limits were defined in accordance with state of the art- or biological variance-derived cut-offs. The examples given are real, not simulated, data. Individual control sample results were normalized to a target value and target standard deviation. The normalized values were used in the exponentially weighted algorithm. The weighting factor was based on a process time constant, which was estimated from the period between two calibration or maintenance procedures. The proposed system was compared with Westgard rules. The Westgard rules perform well, despite the underlying presumption of ergodicity. This is mainly caused by the introduction of the starting rule of 12s, which proves essential to prevent a large number of rule violations. The probability of reporting a test result with an analytical error that exceeds the total allowable error was calculated for the proposed system as well as for the Westgard rules. The proposed method performed better. The proposed algorithm was implemented in a computer program running on computers to which the analyzers were linked on-line. Each result was evaluated on-line, and a limit violation was immediately reported. The system has performed satisfactorily in our laboratory for ten analyzers for over 1 year.