Objectives. To compare the ability of fasting glucose, glycosylated haemoglobin (HbA1c), the American Diabetic Association (ADA) score and measures of the metabolic syndrome (waist circumference, high-density lipoprotein (HDL), and triglycerides) in predicting an initial diagnosis of diabetes mellitus or abnormal glucose tolerance based on the World Health Organization (WHO) criteria.
Design. A cross-sectional, diagnostic study was undertaken of 120 patients admitted to the cardiology wards of Pretoria Academic Hospital for elective coronary angiographic studies. Main outcome measures. All subjects underwent a modified glucose tolerance test whereby fasting and 2-hour post glucose (75 g) plasma glucose levels were measured. Using the revised WHO criteria, the overall incidence of diabetes was found to be 11.7% (95% confidence interval (CI): 6.5 - 19) and the overall incidence of abnormal glucose tolerance was 46% (CI: 37 - 55). Results. In univariate analysis for the diagnosis of diabetes, HbA1c (p < 0.05) yielded the largest area (0.76) under the receiver operating characteristic (ROC) curve, with a sensitivity of 21%, specificity of 99%, positive predictive value (PPV) of 75% and negative predictive value (NPV) of 91%. For the prediction of abnormal glucose tolerance, applying multivariate analysis using a logistic regression model, the combination of age, ethnic group, serum HDL, serum triglycerides and HbA1c yielded an area under the ROC curve of 0.79, sensitivity of 66%, specificity of 80%, PPV of 76% and a NPV of 71%.
Conclusion. Most subjects with diabetes mellitus (9 of 14) would not have been detected if a 2-hour oral glucose tolerance test had not been done. Addition of either HbA1c or lipid parameters to the model of age and ethnic group were similar in predicting abnormal glucose tolerance.