Recently, glycosylated haemoglobin (HbA1c) has been recommended by the American Diabetes Association (ADA), the World Health Organisation and subsequently by many other professional bodies as a diagnostic tool for diabetes mellitus. However, the cut-off values suggested vary between these groups and uncertainties remain regarding the limitations of this test and its effectiveness as a diagnostic tool. We wished to assess the effect of HbA1c on detection rates for dysglycaemia in a high risk cohort of 200 patients with possible acute coronary syndrome not previously known to have diabetes.
Anthropometric as well as HbA1c, oral glucose tolerance tests (OGTT), random and fasting plasma glucose (RPG and FPG) concentrations, fasting lipids and high sensitivity C-reactive protein data were obtained during admission. We examined each of the recommended cut-off values for HbA1c. Test accuracy was assessed by the degree of misclassification (both under- and over-diagnosis) of patients into normal glycaemic control, impaired glucose tolerance and diabetes mellitus based on OGTT data using WHO criteria. A predictive index (PI) was generated using stepwise ordinal regression models (incorporating FPG, HbA1c, HDL-C, triglycerides, age and gender).
HbA1c alone, using the International Expert Committee cut-off values, had unacceptably high misclassification rates (49.0% under- and 2.5% over-diagnosed). This did not improve when ADA criteria were examined, despite their lower cut-off values for normoglycaemia (44.4% under- and 7.1% over-diagnosed). FPG was marginally better, misclassifying 44.4% (mostly under-diagnosis; 41.4%). The PI had the lowest misclassification rate (35.9%; with 22.7% under- and 13.1% over-diagnosed).
In conclusion, our data suggest that HbA1c alone offers little advantage over FPG in detecting dysglycaemia in this high risk population. Our approach using a predictive index to combine HbA1c with other test data will enhance its performance. Copyright © 2012 John Wiley & Sons.