A Priori Classification of Type 2 Diabetes Patients to Determine the Needed Therapy Option

Introduction: T2D patients represent the majority of diabetes patients, and a large and increasing proportion is on insulin, with a huge impact on health care costs. For some a certain degree of automation of insulin dosing and/or regimen adjustment is required. The aim here is to define a priori the automation level required and beneficial for the individual patient. Methods: Data from an outpatient study of 14 T2D patients currently unsuccessfully using MDI therapy were used as a baseline and different optimized insulin therapy options were tested in simulation. Among them 9 were able to reach their therapeutic goals (defined here as HbA1c < 7% and time in hypoglycemia < 2%) with CSII therapy, while 5 needed an Artificial Pancreas (AP) system. In a next step it was attempted to correlate the simulation outcomes with patient features to check whether it is possible to define the preferred therapy option before implementation by classifying those patients into AP or CSII group based on features that could be available beforehand. Results: Coefficient of variation (CV) computed from 14 days of CGM data at baseline MDI therapy and HbA1c have been found to be suitable features for an a priori classification via Naïve Bayes or Classification Tree (CT) algorithm which results in a perfect separation of classes. Conclusion: It was found that it seems possible to predict which subgroup of T2D patients has an additional benefit of using an AP.