Leveraging Support Vector Machines for Enhanced Diagnosis of Diabetes and Prediabetes

The widespread prevalence of diabetes poses a serious threat to global health, with millions of individuals affected and thousands of deaths annually. Early diagnosis and effective management of diabetes and prediabetes are essential to prevent severe complications. The oral glucose tolerance test (OGTT) remains a widely used diagnostic tool; however, analyzing individual blood sugar readings may not provide a comprehensive assessment of glucose metabolism. This study explores the potential of Support Vector Machines (SVMs), a machine learning technique, to analyze comprehensive glucose response curves derived from the OGTT. Area Under the Curve of Glucose (AUCG) and Insulin (AUCI) were calculated based on OGTT measurements, excluding fasting and 120-minute values typically used for standalone diagnosis. Patient data (n=188) with confirmed diabetic or prediabetic status was used to train and validate the SVM models. Performance was evaluated using sensitivity, specificity, accuracy, F1 score, and predictive values. AUCG demonstrated significantly superior predictive power compared to AUCI for both diabetes and prediabetes. All performance metrics for AUCG exceeded 93.7% for diabetes diagnosis and 65.9% for prediabetes. AUCI metrics were respectable but lower, reaching 60.6% for diabetes and 43.6% for prediabetes. These findings suggest that AUCG, capturing the dynamic glucose response during the OGTT, offers a more robust tool for diagnosing diabetes and prediabetes compared to analyzing individual blood sugar readings. This advantage might be attributed to a better assessment of insulin resistance, a hallmark of type 2 diabetes. This study demonstrates the potential of SVMs combined with AUCG derived from OGTT data to enhance the accuracy of diagnosing diabetes and prediabetes. Further research is warranted to confirm these findings in larger and more diverse populations.