2012-LB: AI-Based Meal Detection Enables Fully-Automated Pramlintide and Insulin Closed-Loop System to Improve Postmeal Glucose in Type 1 Diabetes (T1D)

Le résumé fourni par la source

Introduction and Objective: Commercial hybrid closed-loop systems require users to announce meals. Amylin, co-secreted with insulin by β-cells, suppresses glucagon, slows gastric emptying, and reduces post-meal glucose. Pramlintide is an FDA-approved amylin analog, used as an adjunctive therapy in T1D. We developed an AI meal detection algorithm and integrated with automated insulin and pramlintide delivery. Objective: To evaluate the meal detection algorithm in a fully closed-loop system with insulin and pramlintide (I-Pram) compared to insulin-only (I-Only) with no premeal announcements. Methods: This randomized crossover trial enrolled 31 adults with T1D (18 F, mean age 35.4, mean A1c 7.3%). Two standardized meals (~67g carbs) at two visits in-clinic were consumed 6 hours apart. The system comprised the iPancreas model predictive controller, two Insulet pods, one filled with insulin and the other with pramlintide, and the Dexcom G6 CGM. Pramlintide and insulin were delivered at a fixed 6 mcg/u ratio. The meal detection neural network uses 32 features from a two-hour history of CGM/insulin data and dosed insulin and pramlintide in response to meal probability and estimated carbs. Insulin was dosed with meal detection at 50% of the estimated carbs within 30 minutes of start-of-meal. If no meal detection occurred by 30 mins after start-of-meal, then a manual bolus was given. We previously reported outcomes showing statistically significant improvement in time in range (70-180 mg/dL: TIRmeal1 44.8% vs. 49.2% and TIRmeal2 from 49.9% vs. 74.4% for I-Only vs I-Pram). Results: 55% of meals were detected in I-Only and 56% in I-Pram within 30 minutes of start-of-meal. A post-hoc analysis showed the meal detection would have detected 70% of meals in I-Only and 68% of meals in I-Pram within 60 minutes of start-of-meal. Conclusion: An AI-enabled meal detection algorithm performed well in automated meal insulin and pramlintide dosing, contributing to achieving a TIR of 74% in the 6 hours after a large meal. Disclosure L.M. Wilson: None. C.M. Mosquera-Lopez: None. D.K. Aby-Daniel: None. S. Biswas: None. J.H. Eom: None. M. Howard: None. H. Ling: None. D. Branigan: None. J.A. Leitschuh: None. W. Hilts: None. K. Ramsey: None. R. Dodier: None. A. Ahmann: Advisory Panel; Medtronic. M.B. Gillingham: Speaker's Bureau; Abbott Nutrition. Research Support; Nestlé Health Science, National Institutes of Health. R. Zahr: None. J. Pinsonault: None. J. Lloyd: None. M.C. Riddle: None. P.G. Jacobs: Research Support; Dexcom, Inc. Speaker's Bureau; Dexcom, Inc. Advisory Panel; Eli Lilly and Company. Stock/Shareholder; Pacific Diabetes Technologies. Research Support; SFC Fluidics, Eli Lilly and Company. Funding NIH/NIDDK (R01DK129382); Dexcom

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.