Context: Better procedures are needed whereby national proficiency testing survey providers can assess and improve the accuracy of laboratory measurements in clinical chemistry.
Setting: The 1994 College of American Pathologists Comprehensive Chemistry Survey.
Design: This study of matrix effects and the accuracy of laboratory measurements for 11 analytes linked the logistics of the Survey to definitive methods at the National Institutes of Standards and Technology, reference methods at the Centers for Disease Control and Prevention, proficiency testing materials, and a fresh frozen serum sample. The data were analyzed with a statistical model of laboratory measurements.
Results: (1) Matrix biases affected the results reported from 69% of the 644 peer group/survey specimen pairs evaluated. (2) Because of matrix biases, the reference value was the correct target value only 32% of the time; thus, the traceability established by definitive method and reference method value assignments on Chemistry Survey specimens did not assure accuracy on patient samples. (3) In contrast to matrix biases, the error caused by random matrix effects with proficiency testing samples was about the same as that caused by random specimen effects with fresh frozen serum, and both were less than within-run random analytic error. (4) Calibration biases occurred in 73% of the 180 peer groups evaluated, and, after matrix biases were removed, the total variance of interlaboratory measurements was due to peer group calibration bias (48%), within-peer-group random calibration error (31 %), within-run random error (14%), and random specimen effects (7%).
Conclusions: An opportunity exists to improve method calibration accuracy in clinical chemistry. With improved design, national proficiency testing surveys can monitor and help reduce method calibration error by converting reported survey results to a true accuracy base that predicts accuracy on patient samples. For medical purposes, the correct target values on artificial (matrix-modified) chemistry materials are reference values adjusted for the matrix bias of each peer group. Matrix biases estimated by the use of fresh frozen serum can be used as factors to transfer the accuracy of definitive methods from artificial reference materials to patient samples.