Curt L Rohlfing1, Curtis A Parvin2, David B Sacks3, Randie R Little4. 1. Department of Pathology & Anatomical Sciences, University of Missouri, Columbia School of Medicine, 1 Hospital Drive, Columbia, MO 65212, United States. Electronic address: RohlfingC@missouri.edu. 2. Quality Systems Division, Bio-Rad Laboratories, Plano, TX 75074, United States. 3. Department of Laboratory Medicine, National Institutes of Health, Bethesda, MD 20892, United States. 4. Department of Pathology & Anatomical Sciences, University of Missouri, Columbia School of Medicine, 1 Hospital Drive, Columbia, MO 65212, United States; Department of Child Health, University of Missouri, Columbia School of Medicine, 1 Hospital Drive, Columbia, MO 65212, United States.
Abstract
BACKGROUND: Direct comparison of analytical performance criteria that utilize different statistical approaches can be problematic. We describe a mathematical approach to compare performance criteria for hemoglobin A1c (HbA1c) analysis used by the NGSP standardization program and the College of American Pathologists (CAP) to enhance consistency between the schemes. METHODS: The imprecision (CV) and bias combinations required to pass each criterion at probabilities of 0.95, 0.99 and 0.999 were calculated and used to construct contour plots to compare them. The CV/bias requirements were calculated mathematically for the 2011-2012 CAP (3/3 results within ±7% of the target) and different proposed NGSP (33/40 to 40/40 results within ±7% of the target) criteria, and using computer simulations for the existing NGSP criterion (95% confidence interval of the differences between the method and NGSP within ±0.75% HbA1c). RESULTS: Requiring 37 of 40 results to be within ±7% of the NGSP target best matched the CAP criterion at zero bias (95% chance of passing). CONCLUSIONS: The NGSP Steering Committee recommended a certification criterion of 37 of 40 results within ±7% of the NGSP (reduced to ±6% in 2014). The described evaluation approach may be useful in other situations where comparison of different performance criteria is desired.
BACKGROUND: Direct comparison of analytical performance criteria that utilize different statistical approaches can be problematic. We describe a mathematical approach to compare performance criteria for hemoglobin A1c (HbA1c) analysis used by the NGSP standardization program and the College of American Pathologists (CAP) to enhance consistency between the schemes. METHODS: The imprecision (CV) and bias combinations required to pass each criterion at probabilities of 0.95, 0.99 and 0.999 were calculated and used to construct contour plots to compare them. The CV/bias requirements were calculated mathematically for the 2011-2012 CAP (3/3 results within ±7% of the target) and different proposed NGSP (33/40 to 40/40 results within ±7% of the target) criteria, and using computer simulations for the existing NGSP criterion (95% confidence interval of the differences between the method and NGSP within ±0.75% HbA1c). RESULTS: Requiring 37 of 40 results to be within ±7% of the NGSP target best matched the CAP criterion at zero bias (95% chance of passing). CONCLUSIONS: The NGSP Steering Committee recommended a certification criterion of 37 of 40 results within ±7% of the NGSP (reduced to ±6% in 2014). The described evaluation approach may be useful in other situations where comparison of different performance criteria is desired.
Authors: Zhen Zhao; Jeffrey Basilio; Steven Hanson; Randie R Little; Anne E Sumner; David B Sacks Journal: Clin Chim Acta Date: 2015-04-08 Impact factor: 3.786