Detection of calibration errors in interlaboratory experiments

A procedure is proposed for detecting systematic errors using data of an in terlaboratory experiment by comparing the coefficients of correlation betwe en the coefficients of regression equations for different laboratories with the calculated value. Two samples with significantly different concentrati ons of analytes are used in this procedure. The results of a cycle of an ex ternal quality assessment of Russian clinical laboratories were examined in terms of the analysis of covariance. Concentrations of 14 substances (tota l protein, glucose, several lipids, nitrogen-containing compounds, and some elements) were determined in two lyophilized samples of human blood serum with significantly different concentrations of the analytes. The results we re grouped in accordance with analytical methods. Linear regression equatio ns that relate the results obtained in particular laboratories to the certi fied values were different. In the majority of cases, both the absolute ter ms and the slope ratios of linear regression equations were reliably differ ent. The coefficients of correlation between the absolute terms and the slo pe ratios were compared with theoretically calculated values. In many cases , the empirical coefficients of correlation were reliably higher than the t heoretical values, which indicates that the data contain a systematic error . The cause of this phenomenon is supposedly that a simplified procedure fo r plotting calibration functions was used in the analytical methods under c onsideration.