Objective: To test the validity of a published clinical model for predicting incident diabetes in Japanese Americans.
Research design and methods: A total of 465 nondiabetic Japanese Americans (243 men, 222 women), aged 34-75 years, were studied at baseline and at 5-6 years. A total of 412 subjects were studied at 10 years. The clinical model included age, sex, ethnicity, BMI, systolic blood pressure, fasting plasma glucose (FPG), HDL cholesterol, and family history of diabetes at baseline. Diabetes status at 5-6 and 10 years was determined by 75-g oral glucose tolerance test. The clinical model, 2-h glucose, and FPG were compared using receiver-operating characteristic (ROC) curves. RESULTS-The diabetes risk associated with BMI, sex, and HDL cholesterol differed by age (P <or= 0.011). At 5-6 years, the clinical model ROC curve area (0.896) was higher than that for FPG (0.776, P = 0.008), but not for 2-h glucose (0.851, P = 0.341), for subjects aged <or=55 years. For older subjects, the clinical model ROC curve area (0.599) was lower than that for 2-h glucose (0.792, P <or= 0.001), but not for FPG (0.627, P = 0.467). At 10 years, there were no significant differences between the clinical model, FPG, and 2-h glucose ROC curve areas in either age group.
Conclusions: In Japanese Americans aged <or=55 years, a clinical model was better than FPG for predicting diabetes after 5-6 years but not after 10 years. The model was not useful in older Japanese Americans, whereas 2-h glucose was useful for predicting diabetes risk regardless of age.