Advanced Closed-Loop Control System Improves Postprandial Glycemic Control Compared With a Hybrid Closed-Loop System Following Unannounced Meal

Abstract

<b>Objective:</b> Meals are a major hurdle to glycemic control in type 1 diabetes (T1D). Our objective was to test a fully-automated closed-loop control (CLC) system in the absence of announcement of carbohydrate ingestion among adolescents with T1D, who are known to commonly omit meal announcement. <p><b>Research Design and Methods: </b>Eighteen adolescents with T1D (age 15.6±1.7 years; HbA1c 7.4%±1.5; 9F/9M) participated in a randomized crossover clinical trial comparing our legacy hybrid CLC system (USS-Virginia) with a novel fully-automated CLC system (RocketAP), during two 46h supervised admissions (each with one announced and one unannounced dinner), following 2 weeks of data collection. Primary outcome was the percent time-in-range 70-180mg/dL (TIR) following the unannounced meal, with secondary outcomes related to additional CGM-based metrics.</p> <p><b>Results: </b> Both TIR and time-in-tight-range 70-140mg/dL (TTR) were significantly higher using RocketAP than using USS-Virginia during the 6h following the unannounced meal (83% [64-93] vs. 53% [40-71]; p=0.004 and 49% [41-59] vs. 27% [22-36]; p=0.002, respectively), primarily driven by reduced time-above-range (TAR >180mg/dL 17% [1.3-34] vs. 47% [28-60]), with no increase in time-below-range (TBR <70mg/dL 0% median for both). RocketAP also improved control following the announced meal (mean difference TBR: -0.7%, TIR: +7%, TTR: +6%), overall (TIR: +5%, TAR: -5%, TTR: +8%), and overnight (TIR: +7%, TTR: +19%, TAR: -5%). RocketAP delivered less insulin overall (78±23U vs. 85±20U, p=0.01)</p> <p><b>Conclusions:</b> A new fully-automated CLC system with automatic prandial dosing was proven to be safe and feasible and outperformed our legacy USS-Virginia in an adolescent population with and without meal announcement.</p>

New Contributions Towards Meal and Exercise Announcement-Free Artificial Pancreas Systems

“…Other methods use information from behavioral meal patterns to confirm a meal occurrence (Cameron et al 2012;Villeneuve et al 2020). Lastly, classification algorithms have also been used to discern the meal events, such as logistic regression (Garcia-Tirado et al 2021c;Garcia-Tirado et al 2021b;Corbett et al 2022), linear discriminant analysis (Kölle et al 2017;Kölle et al 2020), extended isolation forest (Zheng et al 2020), fuzzy logic (Samadi et al 2017), or recursive neural networks (Askari et al 2022).…”

“…Some targets announcement simplification, requiring only the mealtime (Tsoukas et al 2021a; or a qualitative approximation of the carbohydrates (Gingras et al 2016b). Others completely removed the meal announcement; most meal-announcement-free systems rely on meal detection (or, at least, some detection of persistent hyperglycemia, like in Colmegna et al 2021a;Garcia-Tirado et al 2021b;Majdpour et al 2021) to trigger a set of feedforward actions playing the role of pre-meal boluses, that is, increasing the aggressiveness of the insulin delivery to reduce postprandial hyperglycemia. The three most frequent actions triggered at detection time are the following: 1) delivering a single insulin bolus (Mahmoudi et al 2019;Samadi et al 2017;Harvey et al 2014b); 2) delivering a train of insulin boluses calculated through estimations of the glucose derivative or rate of glucose appearance (Garcia-Tirado et al 2021b;Turksoy et al 2015;Hyunjin et al 2009); and 3) modifying the controller structure or tuning (Hajizadeh et al 2020;Fushimi et al 2019).…”