Abstract:Background and objectives: Recent advances in Automated Insulin Delivery systems have been shown to dramatically improve glycaemic control and reduce the risk of hypoglycemia in people with type 1 diabetes. However, they are complex systems that require specific training and are not affordable for most. Attempts to reduce the gap with closed-loop therapies using advanced dosing advisors have so far failed, mainly because they require too much human intervention. With the advent of smart insulin pens, one of the main constraints (having reliable bolus and meal information) disappears and new strategies can be employed. This is our starting hypothesis, which we have validated in a very demanding simulator. In this paper, we propose an intermittent closed-loop control system specifically intended for multiple daily injection therapy to bring the benefits of artificial pancreas to the application of multiple daily injections. Methods: The proposed control algorithm is based on model predictive control and integrates two patient-driven control actions. Correction insulin boluses are automatically computed and recommended to the patient to minimize the duration of hyperglycemia. Rescue carbohydrates are also triggered to avoid hypoglycemia episodes. The algorithm can adapt to different patient lifestyles with customizable triggering conditions, closing the gap between practicality and performance. The proposed algorithm is compared with conventional open-loop therapy, and its superiority is demonstrated through extensive in silico evaluations using realistic cohorts and scenarios. The evaluations were conducted in a cohort of 47 virtual patients. We also provide detailed explanations of the implementation, imposed constraints, triggering conditions, cost functions, and penalties for the algorithm. Results: The in-silico outcomes combining the proposed closed-loop strategy with slow-acting insulin analog injections at 09:00 h resulted in percentages of time in range (TIR) (70-180 mg/dL) of 69.5%, 70.6%, and 70.4% for glargine-100, glargine-300, and degludec-100, respectively, and injections at 20:00 h resulted in percentages of TIR of 70.5%, 70.3%, and 71.6%, respectively. In all the cases, the percentages of TIR were considerably higher than those obtained from the open-loop strategy, being only 50.7%, 53.9%, and 52.2% for daytime injection and 55.5%, 54.1%, and 56.9% for nighttime injection. Overall, the occurrence of hypoglycemia and hyperglycemia was notably reduced using our approach. Conclusions: Event-triggering model predictive control in the proposed algorithm is feasible and may meet clinical targets for people with type 1 diabetes.

Simulation-driven optimization of insulin therapy profiles in a commercial hybrid closed-loop system

Intermittent Control for Safe Long-Acting Insulin Intensification for Type 2 Diabetes: In-Silico Experiment

Intermittent closed-loop blood glucose control for people with type 1 diabetes on multiple daily injections

Developing the UVA/Padova Type 1 Diabetes Simulator: Modeling, Validation, Refinements, and Utility

Basal Glucose Control in Type 1 Diabetes Using Deep Reinforcement Learning: An In Silico Validation

An Online Stochastic Optimization Approach for Insulin Intensification in Type 2 Diabetes with Attention to Pseudo-Hypoglycemia

Performance of an Automated Insulin Delivery System: Results of Early Phase Feasibility Studies

Run-to-Run Tuning of Model Predictive Control for Type 1 Diabetes Subjects: In Silico Trial

Optimized Glycemic Control of Type 2 Diabetes with Reinforcement Learning: a Proof-of-concept Trial.

Outpatient Randomized Crossover Comparison of Zone Model Predictive Control Automated Insulin Delivery with Weekly Data Driven Adaptation Versus Sensor-Augmented Pump: Results from the International Diabetes Closed-Loop Trial 4

An Insulin Bolus Advisor for Type 1 Diabetes Using Deep Reinforcement Learning

Personalized insulin dosing using reinforcement learning for high-fat meals and aerobic exercises in type 1 diabetes: a proof-of-concept trial

Real-time Artificial Intelligence Assisted Insulin Dosage Titration System for Glucose Control in Type 2 Diabetic Patients: a Proof of Concept Study

Towards the Integration of an Islet-Based Biosensor in Closed-Loop Therapies for Patients With Type 1 Diabetes

Insulin dose optimization using an automated artificial intelligence-based decision support system in youths with type 1 diabetes

Reinforcement Learning-Based Insulin Injection Time and Dosages Optimization.

Refining Insulin on Board with netIOB for Automated Insulin Delivery

Turning the tides: achieving rapid and safe glucose control in adolescents with suboptimally controlled type 1 diabetes using advanced hybrid closed loop systems

Insights into the Feasibility and Acceptability of a Mobile Insulin Titration Application in Clinical Practice and Its Effects on Diabetes Outcomes

Offline Deep Reinforcement Learning and Off-Policy Evaluation for Personalized Basal Insulin Control in Type 1 Diabetes

Improved Glycemic Outcomes With Medtronic MiniMed Advanced Hybrid Closed-Loop Delivery: Results From a Randomized Crossover Trial Comparing Automated Insulin Delivery With Predictive Low Glucose Suspend in People With Type 1 Diabetes