Type 1 diabetes (T1D) is a disease characterized by pancreatic beta destruction with subsequent insulin depletion. The alterations in glucose dynamics are incredibly difficult to manage and are confounded by meals, exercise, menstruation, and stress. Although automated insulin delivery systems are becoming commercially available, the large majority of people with T1D are treated with multiple daily injections of insulin (MDI). Dangerous complications of hypoglycemia and diabetic ketoacidosis can occur from failure to dose insulin correctly, however vigilant adherence to tedious insulin dosing strategies are difficult for MDI users maintain. This difficulty is magnified during exercise, which is critical to ameliorating long-term complications of diabetes; even when guidelines for insulin dosage adjustments are followed, acute hypoglycemia during exercise and night-time hypoglycemia can occur. In our recent survey of 1400 subjects living with T1D, the majority of subjects on MDI therapy were not confident in managing their glucose during exercise and felt they lacked tools to do so. In aggregate, difficult treatment schedules and bolus calculations, associated acute complications from daily activities, and the emotional and psychological toll of this chronic disease can result in treatment non-adherence and poor glycemic outcomes. Therefore, there is a critical need for decision support tools designed for MDI users to improve glycemic control surrounding meals, daily activities and exercise. The goal of this proposal to develop a decision support tool for patients with type 1 diabetes who utilize continuous glucose monitoring systems and multiple daily injection therapy. This tool will be called miTREAT, the multiple injection treatment recommender system for exercise-aware therapies. We hypothesize that use of a novel decision support tool equipped with content-based collaborative filtering methods and dynamic exercise hypoglycemia prediction algorithms will improve overall euglycemia and reduce time spent in hypoglycemia for patients on MDI therapy. In our first aim, we will leverage decades of research in computer science recommender systems and machine learning optimization strategies to develop a novel decision support system that identifies issues in glycemic control and recommends appropriate insulin dose and behavioral modifications. In our second aim, we will develop a new exercise model that reflects both the dynamics of rapid-uptake of glucose through GLUT-4 channels and the longitudinal biphasic insulin sensitivity profile. This new model structure will be used to predict hypoglycemia during and after the exercise period. In our third aim, we will explore the performance of our decision support engine in an in-vivo clinical trial. This clinical trial will assess the usability of a new smart-phone app designed to assist MDI users that we have developed at OHSU. In achieving these goals, we will develop the first decision support system that provides treatment and behavioral recommendations to patients on CGM-augmented MDI therapy. This system will improve overall time in euglycemia, and reduce the occurrence of acute complications surrounding exercise.
Type 1 Diabetes is difficult to manage through multiple daily injections, and there is a paucity of tools available to assist patients with treatment decisions. In this proposal, the fields of computer science, machine learning, constrained optimization and clinical science are united to develop a novel decision support tool for multiple daily injection users. This is accomplished in three parts: 1) we will tune a decision support system to correctly identify issues in patient glycemic control and recommend appropriate treatment modifications; 2) we will develop a new hypoglycemia prediction tool that reflects the longitudinal dynamics of glucose uptake during and after exercise; and 3) we will explore the performance of this decision support system and hypoglycemia prediction tool in a clinical trial.