Estimation and SVM classification of glucose-insulin model parameters from OGTT data: a comparison with the ADA criteria

Abstract

Background: The oral glucose tolerance test (OGTT) is used for the diagnosis of diabetic conditions. The American Diabetes Association (ADA) provides criteria for definition, namely, impaired fasting glucose (IFG), impaired glucose tolerance (IGT), type 2 diabetes mellitus (T2DM), and normal glucose tolerance (NGT). Purpose of the study: To examine the application of parameters estimated in models of the glucose-insulin regulatory system during the OGTT, as a classification tool of diabetic conditions. Methods: Given a set of OGTT data, parameters for each subject are estimated from and ODE model using a Bayesian approach. Point clouds are constructed with parameter pairs and inspected for classification, using a support vector machine (SVM) learning technique. The classical train-test split is used for validation. The training set is comprised of 80 non-related, female volunteers recruited at the Mexico General Hospital. Results: The parameters peak glucose concentration and average of glucose removal rates are suitable for classification. For the training set, the classification was successful for at least 85% of subjects. Noteworthy, a linear separation suffices. The classification is tested on an independent population of OGTT data for 24 males and 33 females. Classification is successful for 91% of males and 87% of females. Ill patients are correctly classified. Conclusion: Peak glucose concentration and average of glucose removal rates are proposed as potential patient’s indices for diabetic condition. As a graphical tool, clinicians may interpret the SVM classification diagrams. These show a transition from healthy to diabetic. The gray area might suggest pre-diabetic subjects.