predictive values of one test and importance of analytical bias and imprecision
BACKGROUND: In Denmark, the use of HbA1c in the diagnosis of diabetes was adopted from March 2012. We evaluated the change in the number of diabetes cases diagnosed by haemoglobin A1c (HbA1c) versus fasting venous plasma glucose (FPG), and estimated the influence of analytical variation and bias on the HbA1c-based prevalence of diabetes. METHODS: The study population constituted 4239 individuals not known to have diabetes randomly selected from all inhabitants aged 25-75 years in the former County of Vejle, Denmark. The number of undiagnosed patients with diabetes in the study population using FPG or HbA1c as the diagnostic criterion was estimated. Furthermore, changes in the analytical bias and coefficient of variation (CV) for HbA1c analysis were simulated and the effect on the number of diabetes cases was observed. RESULTS: Changing the diagnostic test from FPG to HbA1c reduced the number of patients with diabetes by approximately 46% based on one measurement. The predictive value of one test of HbA1c was 91% versus only 66% for one test of FPG. Analytical variation had a much greater impact on the number of patients with diabetes than bias. At a bias of 0%, an increase of CVanalytical from 2.7% to 3.7% increased the number of diabetes cases by 90%. CONCLUSIONS: In the study population, the percentage of undiagnosed patients with diabetes aged 25-75 years was reduced from 3.6% (95% CI 3.0%-4.2%) based on one FPG measurement (FPG ≥7.0 mmol/L) to only 1.9% (95% CI 1.5%-2.3%) if the diagnosis of diabetes was based on the criterion of HbA1c ≥48 mmol/mol (6.5% DCCT).
Clinical Chemistry and Laboratory Medicine, 2014, Vol 52, Issue 7, p. 1069-77