Predicting the 20-year diabetes incidence rate

BACKGROUND: The long-range prediction from clinical variables of the onset of diabetes is important to patients and clinicians. Our objective was to evaluate the efficacy of various glucose-related clinical measurements in predicting the 20-year risk of developing type 2 diabetes (T2DM) in an elderly population. METHODS: In a prospective study, 672 men and women aged 59-92 years, who were not diabetic in 1980 and were part of a nationwide longitudinal randomized study, were followed-up in 2000-2003. Fasting glucose, 1- and 2-h post-oral glucose tolerance and insulin were measured in 1980 and 2000-2003. RESULTS: A group of 174 (25.9%) survivors had progressed to diabetes during the 20-year follow-up. Fasting glucose values were a good predictor for diabetes. With the 100 mg/dL cut-off of impaired fasting glucose (IFG), a 2-4-times higher predictive sensitivity followed the dramatic increase in IFG prevalence compared to the 110 mg/dL cut-off, but at a cost of reduced specificity and positive predictive value (PPV). By receiver operating curve (ROC) analysis, a 1-h post-load glucose was similar to 2 h and fasting glucose in prediction of the 20-year incidence of diabetes, and classifying correctly the 77, 74 and 73% of the group, respectively. In adjusted logistic regressions, 2.28, 1.78 and 1.69-folds increased the 20-year risk, and were associated with each SD increment of the respective glucose values (p < 0.001). CONCLUSIONS: Although the best population-based strategy for the diagnosis of T2DM would be the combination of fasting glucose followed by post-load glucose, for the purposes of long-term prediction of T2DM risk, fasting glucose is sufficient.