Thursday, April 28, 2005

Use of patient data to derive long term random error in laboratory assays: Application to glycohemoglobin testing

George Cembrowski, David V. Tran, Tammy L Hofer, and Terrence Lee.

The measurement of glycohemoglobin is the best measure of mean glucose within a three month range. As it is used for patient education, counseling, feedback control and ultimately for patient motivation, its measurement should be optimally accurate and precise. Estimates of imprecision are usually based on the repeated analysis of reference samples. These estimates are dependent on the reference sample's characteristics and where it enters the analytical stream.

We describe a novel, independent approach for deriving total imprecision of glycohemoglobin assays in which intra-individual glycohemoglobin variations are plotted against the time between sampling. Extrapolation to zero time will yield the total random error.

Glycohemoglobin measurements of pairs of outpatient blood samples drawn between 0 and 30 days were made on the Bio-Rad Variant II's cation exchange high performance liquid chromatography (HPLC) assay (Hercules, CA), and the Beckman LX-20 turbidimetric immunoinhibition system (Beckman Coulter, Fullerton, CA). The average variation of grouped duplicates was calculated and graphed against corresponding time intervals. Regression to the y-intercept (0 day separation between readings) was used to determine the analytic variation.

2789 and 790 pairs of HPLC and immunochemical glycohemoglobin values were obtained with the time between sampling varying from 0 to 30 days. After outlier removal, analytic coefficients of variation (CVs) for the HPLC and immunoassay were determined as 2.8% and 4.4%, respectively.

The immunochemical assay's random error, at 4.4%, significantly exceeds the maximum limits for random error established by biologic variation (2 to 3 %) as well as the limits of the National Glycohemoglobin Standardization Program (3% for laboratories involved in clinical trials and 4% for reagent / instrument manufacturers and other laboratories). In contrast, the random error of the HPLC method, at 2.8%, appears to be acceptable. This approach to deriving total imprecision should be extended to all available glycohemoglobin analyzers.

See more of Poster Session
See more of The 2005 Institute for Quality in Laboratory Medicine Conference