Calibration Error, a Neglected Error Source in the Clinical Laboratory Quality Control

"Calibration" conveys the connotative meaning of "correction." Therefore, calibration is frequently viewed as "perfect," but it is a measurement, and no measurement is error-free. This study aims to uncover the sources of calibration errors, to estimate their size, and assess their consequences in quality control. The analytical bias is the difference between the working (determined) graph and the ideal graph (how the reagents behave). The source of the calibration random component is the random error committed in the calibration. The primary source of the systematic component is the reference material value error, which cannot be reduced to the nominal value error. Even if the avoidable human errors are neglected, the reconstitution errors, including two volume measurements, are inherent. The random component was estimated by making five calibrations in repeatability conditions and calculating the coefficient of variation of the slope factors. The total calibration error was estimated by comparing the slope factors of new calibrations using the same reagent and calibrator lots (one-year data). The results confirmed the presumptions: the calibration error is bigger than the coefficient of variation measured in repeatability conditions. Smaller biases are incorrigible by calibration, and quality control rules must be designed to prevent them from being detected. Using the σ parameter in the QC graphs would result in too frequent alarms. Westgard proportionally increased the decision limits by overestimating σ with the standard deviation measured in reproducibility within laboratory conditions. A more accurate solution is to increase all decision limits to account for the incorrigible bias and design the QC graphs with the standard deviation measured in repeatability.