A globally applicable screening model for detecting individuals with undiagnosed diabetes

Abstract

Aims: Current risk scores for undiagnosed diabetes are additive in structure. We sought to derive a globally applicable screening model based on established non-invasive risk factors for diabetes but with a more flexible structure. Methods: Data from the DETECT-2 study were used, including 102,058 participants from 38 studies covering 8 geographical regions worldwide. A global screening model for undiagnosed diabetes was identified through tree-structured regression analysis. The performance of the global screening model was evaluated in each of the geographical regions by receiver operating characteristic (ROC) analysis. Results: The global screening model included age, height, body mass index, waist circumference and systolic- and diastolic blood pressure. Area under the ROC curve ranged between 0.64 in North America and 0.76 in Australia and New Zealand. Overall, to identify 75% of the undiagnosed diabetes cases, 49% required further diagnostic testing. Conclusions: We identified a globally applicable screening model to detect individuals at high risk of undiagnosed diabetes. The model performed well in most geographical regions, is simple and requires no calculations. This global screening model may be particularly helpful in developing countries with no population based data with which to develop own screening models.