ADAPTIVE MOTIF-BASED CONTROL (AMBC): A FUNDAMENTALLY NEW APPROACH TO AUTOMATED TREATMENT OPTIMIZATION FOR TYPE 1 DIABETES

点击翻译按钮获取中文摘要

PROJECT SUMMARY Adaptive Motif-Based Control (AMBC): A Fundamentally New Approach to Automated Treatment Optimization for Type 1 Diabetes Automated Insulin Delivery (AID) has transitioned to the clinical practice and one of the most advanced systems to date –Control-IQ (Tandem, Inc.)– is based on the UVA-AID, developed and tested by our team. Following two pivotal trial in adults and children with type 1 diabetes (T1D), both published in the New England J of Medicine, the system was cleared by the FDA and European regulatory agencies, and is now in use worldwide. With this clinical translation now accomplished, our academic objective is to design and test next-generation AID solutions. The key innovative concept behind the new Adaptive Motif-Based Control (AMBC) class of AID algorithms proposed here, is the ability to learn from a user’s past glycemic-control patterns and from population patterns, and optimize this user’s treatment in real time; thus, unlike any other AID system, the AMBC will utilize both a person’s own history and the history of others, to forecast glycemic changes and adapt AID action accordingly. To achieve its goals, the AMBC will employ: (i) a newly discovered fundamental structure underlying the multitude of possible daily continuous glucose monitoring (CGM) profiles in diabetes, which allows classification of these profiles into a finite number of basic “motifs”, and (ii) a new Adaptation-to-Profile treatment optimization process. To test the AMBC, we propose a pilot study, followed by a randomized cross-over trial enrolling 90 participants with T1D and Control-IQ experience, to compare the UVA-AID (as built in Control-IQ) to 3 treatment modalities: AMBC with meal and exercise announcements; AMBC-A without meal or exercise announcements, i.e. a “full closed-loop,” and an intermediate AMBC-EA which will have meal but no exercise announcement. Participants will be randomized to two groups following different sequences of treatment modalities: UVA-AID-->AMBC- A-->AMBC-EA-->AMBC and AMBC-->AMBC-EA-->AMBC-A-->UVA-AID. Each treatment modality will continue for 5 weeks. This time-tested design enables four crossover comparisons, which will test the following hypotheses: (1) AMBC with meal/exercise announcements will be superior to UVA-AID in terms of time in the range 70- 180mg/dl and reduced incidence of hypoglycemia (measured by CGM), and technology acceptance; (2) AMBC-A without meal/exercise announcements will be non-inferior to UVA-AID in terms of time >180mg/dL during the day, incidence of hypoglycemia during and after exercise, and postprandial glucose variability; (3) Deescalating AMBC-->AMBC-EA-->AMBC-A vs escalating AMBC-A-->AMBC-EA-->AMBC deployment of meal and exercise announcements will have no influence on the outcome within each treatment modality. Overall, we affirm that reliable technology has been developed and sufficient data accumulated to warrant the development of a new class of AID algorithms – AMBC – which is expected to become the first AID system learning from a user’s own historical metabolic patterns and from the patterns of others structured in a new relational data library. We view the AMBC as a paradigm for transition from hybrid to full (no announcements) AID, but do not claim that the AMBC is a full closed-loop system.