Prediction of Undiagnosed Diabetes in Africans is Optimized by Using Fasting Plasma Glucose at a Threshold of 100 mg/dL: The Africans in America Study

Introduction and Objectives: Detecting undiagnosed diabetes is the first step towards addressing the diabetes epidemic in Africa. Although there is little experience with diabetes prediction equations in Africans, the A1C-modified Atherosclerosis Risk in Communities (ARIC) diabetes prediction equation was optimized in African Americans. Therefore, we evaluated (a) the ability of the A1C-modified ARIC prediction equation to detect undiagnosed diabetes in African-born blacks living in the United States and (b) the contribution of each biochemical variable to the effectiveness of the equation. Methods: Participants were 400 self-identified healthy African-born blacks enrolled in the Africans in America study (age 38 ±10 (mean ±SD) years, BMI 27.5 ±4.5, range 18.2-42.4 kg/m 2 ). Glucose tolerance status was diagnosed by glucose criteria for the OGTT. The prediction equation had 9 variables; 5 clinical (age, parent history of diabetes, height, waist circumference, and systolic BP); and 4 biochemical (A1C, fasting plasma glucose (FPG), high density lipoprotein (HDL) and triglycerides (TG)). Area under the receiver operating characteristic curve (AUC-ROC) predicted discrimination and Youden Index identified optimal cut-points. Four models were tested. Model 1 determined the predictive value of the full equation. Model 2 included the clinical variables and 3 of the 4 biochemical variables at a time. Model 3 included the clinical variables and 1 biochemical variable at a time. Model 4 evaluated the independent prediction by each biochemical variable. Results: Diabetes, prediabetes and normal glucose tolerance were detected in 7% (26 of 400), 27% (108 of 400) and 66% (266 of 400), respectively. Model 1 had an AUC-ROC of 0.8. Model 2 with FPG excluded had an AUC-ROC of 0.7. When A1C, HDL or TG were the excluded AUC-ROC increased to >0.8. Model 3 with clinical variables and FPG the AUC-ROC was 0.9. However, when A1C, HDL or TG were included AUC-ROC declined to ≤0.7. In Model 4, FPG as a single predictor had an AUC-ROC of 0.9. In contrast, A1C, HDL or TG as single predictors, all had AUC-ROC ≤0.7. For the prediction of diabetes, the optimal cut-point for FPG was 99 mg/dL (sensitivity 85%, specificity 87%). Conclusions: As the prediction of prevalent diabetes by the entire equation and FPG alone were similar, FPG ≥100 mg/dL in Africans may be a sufficient and cost-effective way forward.