Keywords: Physiological modeling, Other technological contributions to diabetes management
Abstract: Type 1 Diabetes (T1D) management is a complex task due to many variability factors. Artificial Pancreas (AP) systems have alleviated patient burden by automating insulin delivery through advanced control algorithms. However, the effectiveness of these systems depends on accurate modeling of glucose-insulin dynamics, which traditional mathematical models often fail to capture due to their inability to adapt to patient-specific variations. This study introduces a Biological-Informed Recurrent Neural Network (BI-RNN) framework to address these limitations. The BI-RNN leverages a Gated Recurrent Units (GRU) architecture augmented with physics-informed loss functions that embed physiological constraints, ensuring a balance between predictive accuracy and consistency with biological principles. The framework is validated using the commercial UVA/Padova simulator, outperforming traditional linear models in glucose prediction accuracy and reconstruction of unmeasured states, even under circadian variations in insulin sensitivity. The results demonstrate the potential of BI-RNN for personalized glucose regulation and future adaptive control strategies in AP systems.

Keywords: Physiological modeling
Abstract: The development of new automatic insulin delivery strategies is based on the use of simulators of virtual patients with type 1 diabetes. These simulators are often limited due to a lack of description of realistic scenarios and disturbances. One of the main disturbances ever present in the patients' lives is meals, but the glucose response to meals has great variability among meal compositions and patients. Most widely used meal models in simulators have a single carbohydrate input, limiting the description of the meal compositions. This paper aims to analyze the performance of the most widely used meal models for type 1 diabetes in different meal composition scenarios while exploring the feasibility of a systematic identification procedure.

Keywords: Physiological modeling
Abstract: Meals with a high content of fat (F) and protein (P) still represent a significant challenge for optimal glycemic control in individuals with type 1 diabetes (T1D). Current strategies dealing with such meals require patients to undergo a series of trial-and-error visits until satisfactory glycemic control is achieved. In this study, we updated the UVA/Padova T1D simulator by integrating F and P effects on model parameters to allow a realistic simulation of meals with varying content. Glycemic profiles of one-hundred virtual subjects were simulated after a nominal meal (75 g of carbohydrates, 20 g of F, and 24 g of P), a high-F meal (HF, +20 g of F to the nominal meal), a high-P meal (HP, +20 g of P), and a HF/HP meal (+20 g of F and +20 g of P) using the standard (carbohydrate-only based) meal insulin bolus. Simulations were evaluated in terms of average blood glucose (BG), time in range (TIR), and time below range (TBR). The simulated glucose time courses were compared against continuous glucose monitor profiles collected on real individuals who ate similar meals, showing comparable median profiles and variability. Then, the simulator was employed to test the efficacy of a dual-wave bolus (+30% of the dose, given 50% immediately and 50% over 2 h) over the standard bolus on the HF/HP meal challenge. The dual-wave bolus produced a significantly lower BG (p<0.0001) and higher TIR compared to the standard therapy (p<0.01), with TBR similar to that of the nominal meal (p=0.29), demonstrating its ability to restore normal glucose levels, without excessively increasing the time in hypoglycemia. However, the updated simulator still lacks proper validation against data that are free from any confounding factor. Therefore, the next step will involve the collection of data in a trial specifically designed to detect the effect of different macronutrients in the meal. Once validated, the proposed platform will be suitable for testing therapies involving meals with different macronutrient content, reducing the burden for clinical trials. Future work will also focus on exploring other dietary factors and integrating time-varying effects of F and P to improve precision and applicability of the simulator.

Keywords: Physiological modeling
Abstract: Type 1 diabetes mellitus is a chronic condition that requires insulin delivery to maintain blood glucose levels within a desired range. The artificial pancreas (AP), which integrates a continuous glucose sensor, an insulin pump, and a control algorithm, is a promising solution for automating insulin delivery. Designing optimal controllers for the AP is crucial to its effectiveness. Existing approaches often rely on advanced controllers based on models of the insulin-glucose system. However, this system is highly complex, nonlinear, and subject to time-varying dynamics and inter-patient variability, which pose significant challenges for model accuracy and control design. Hence, data-driven and machine learning-based models are emerging as powerful alternatives.

This paper presents a novel data-driven modeling approach that combines two components: a linear model and a machine learning-based model. This latter is computed with the CHoKI learning method, to capture the nonlinear deviations of the actual system from the linear model, enabling the combined model to better represent the insulin-glucose system. This hybrid modeling approach offers improved prediction accuracy compared to previously proposed models in the literature. The improved model accuracy can lead to better controllers for the AP. The proposed approach is validated using the virtual patients of the FDA-accepted UVA/Padova simulator. The results outperform state-of-the-art models in prediction errors, demonstrating its potential as a step forward in AP control system design.

Keywords: Physiological modeling, Reducing the computational footprint for embedded Automatic Insulin Delivery systems, New control system algorithms for Automatic Insulin Delivery
Abstract: The primary objective of this paper is to demonstrate a systematic approach to deriving multiple polytopic representations of the Quasi-Linear Parameter Varying (QLPV) state-space model of the Identifiable Virtual Patient (IVP) model. The paper illustrates how to apply a polytopic model transition method to generate polytopic models with both tight and loose convex hulls of the vertices, as well as intermediate configurations between these extremes. The paper then investigates how convex hull manipulation affects the components of the polytopic controller derived using the Linear Matrix Inequality-based Parallel Distributed Compensation design framework. The paper highlights several critical considerations, such as the phenomenon of convex hull explosion in both the polytopic model and the controller. Furthermore, it affirms the design heuristic that a tighter convex hull of the polytopic model generally results in tighter convex hull in the polytopic representations of the controller.

Keywords: Physiological modeling, Other technological contributions to diabetes management
Abstract: Glucose prediction models are essential for managing type 1 diabetes through digital therapeutics. Achieving reliable predictions requires precise calibration of key model parameters, which is non-trivial. This paper introduces a novel methodology for ranking Hovorka’s glucose-insulin model parameters based on their influence on model output variance. The methodology shows how the fed-fast cycle alters the parameter importance, emphasizing the need for real-time model adaptation. Calibrating the selected parameters using this methodology results in a 90% and 32% reduction in Root Mean Square Error (RMSE) in simulated and real data, respectively, demonstrating the efficacy of the approach.

Keywords: Other technological contributions to diabetes management, Diabetological education
Abstract: Automated insulin delivery (AID) systems have shown significant potential in managing type 1 diabetes (T1D), yet personalizing therapy parameters remains challenging. This study advances the optimization of AID therapy profiles through an updated decision support system (DSS) leveraging the University of Virginia Replay Simulator (UVA-RS). The DSS employs a personalized glucose-insulin dynamics model to simulate glucose response to therapy adjustments and an optimization algorithm to determine therapy parameters that improves overall glycemic control. We evaluated the system’s performance through three in-silico scenarios, focusing on recommendation reliability, constraint impact, robustness to metabolic and behavioral variability, and performance over five-month simulated use. Results indicate improved therapy personalization and glycemic control, supporting the potential for DSS to enhance AID system efficacy.

Keywords: Automated algorithms for multiple day injection insulin therapy, New control system algorithms for Automatic Insulin Delivery, Other technological contributions to diabetes management
Abstract: Type 1 diabetes management requires compensating carbohydrate intake with bolus insulin matched to the meal size. Recent clinical studies revealed diurnal variations in insulin sensitivity (SI) in patients with type 1 diabetes, where the insulin resistance varies over the day. Diurnal variations in insulin sensitivity requires different bolus insulin dose for the same meal size depending on the time of the meal. Standard bolus calculators that use patient-specific parameters such as insulin-carb-ratio (CR) and correction factors (CF), however do not account for such diurnal variations. To address this gap, this paper proposes a fully data-driven safe and personalized bolus calculator that explicitly accounts for the diurnal variations. The proposed algorithm safely learns the optimum bolus needs tailored to each patient without the need for any patient-specific parameters such as carb-ratio, correction factor, insulin sensitivity etc., nor any historical clinical data. The proposed algorithm is tested and verified on the 10-adult cohort of the FDA-accepted UVA/Padova T1DM simulator.

Keywords: Diabetological education, Other technological contributions to diabetes management, Physiological modeling
Abstract: This paper presents the design and implementation of an advanced Continuous Glucose Monitoring system based on the ESP32 microcontroller. By integrating sensors, wireless communication, and real-time data processing capabilities, the system aims to provide accurate, low-latency glucose readings suitable for diabetes management. The use of the ESP32 enables seamless connectivity with mobile devices for remote monitoring and data analysis, facilitating a personalized approach to glucose control. The performance of the system is evaluated for accuracy, reliability, and user-friendliness, offering a cost-effective and scalable solution for continuous health monitoring.

Keywords: Diabetological education, New control system algorithms for Automatic Insulin Delivery
Abstract: We provide an open-source implementation of the NMPC algorithm for closed-loop insulin therapy presented by Hovorka et. al. in the seminal paper from 2004. An extended Kalman filter is added as observer, since specifics on state estimation have been omitted in the original paper. The controller is now part of the TypeScript-based closed-loop diabetes simulator LT1, which also comprises a library of models describing CGM sensors, infusion pumps, and patient physiology. The controller can thus be extensively tested in various closed-loop configurations. Also, the graphical user interface enables its integration in educational tools.

Keywords: Diabetological education
Abstract: This paper presents a preliminary test on the satisfaction of users and the feasibility of a workshop using robotics and gamified simulation performed in a hospital setting where children with diabetes and their parents participated. In order to do so, three different workshops were performed in a hospital, and surveys were conducted to test the feasibility of the approach and the users’ satisfaction. In these workshops, the children learned about Type 1 Diabetes and its management. From the gathered data and the analysis performed, it can be supposed that workshops using the new materials are feasible and that the users are satisfied.

Keywords: Other technological contributions to diabetes management, Physiological modeling
Abstract: Accurate blood glucose prediction can enable novel interventions for type 1 diabetes treatment including personalized insulin and dietary adjustments. Although recent advances in transformer-based architectures have demonstrated the power of attention mechanisms in complex multivariate time series prediction, their potential for blood glucose (BG) prediction remains underexplored. We present a comparative analysis of transformer models for multihorizon BG prediction, examining forecasts up to 4 hours and input history up to 1 week. The publicly available DCLP3 dataset (n=112) was split (80%-10%-10%) for training, validation, and testing, and the OhioT1DM dataset (n=12) served as an external test set. We trained networks with point-wise, patch-wise, series-wise and hybrid embeddings, using CGM, insulin, and meal data. For short-term blood glucose prediction, Crossformer, a patch-wise transformer architecture, achieved a superior 30-minute prediction of RMSE (15.6 mg / dL on OhioT1DM). For longer-term predictions (1h, 2h, and 4h) PatchTST, another path-wise transformer, prevailed with the lowest RMSE (24.6 mg/dL, 36.1 mg/dL, and 46.5 mg/dL on OhioT1DM). In general, models that used tokenization through patches demonstrated improved accuracy with larger input sizes, with the best results obtained with a one-week history. These findings highlight the promise of transformer-based architectures for BG prediction by capturing and leveraging seasonal patterns in multivariate time-series data to improve accuracy.

Keywords: Physiological modeling
Abstract: Preterm birth is the leading cause of death in young children, and is often associated with long-term neurodevelopmental impairment in children that reach school age. Parenteral and enteral nutrition play a crucial role in the development of the infant admitted to the neonatal intensive care unit (NICU), as it is the only mean of nutrition for the neonates. However, standard nutrition protocols, not individualized to the neonate’s specific needs, are followed. A personalization of parenteral and enteral nutrition could help in the development of the infant brain, and be associated to better clinical outcomes. In particular, hypoglycaemic and hyperglycaemic episodes have to be avoided, as they are associated with cognitive impairments and increased mortality rates. Therefore, the availability of a nutritional clinical advisor, able to suggest the optimal amount of glucose to feed the hospitalized neonates, would be of great help in the NICU. Needless to say, the preterm neonates are very fragile and glycaemic control algorithms have to be carefully tested before being applied in this population. Therefore, the first step toward the design of these tools, is to build up a model able to simulate reliable glucose traces of neonates, to safely and effectively test such advisory systems. The aim of this study is thus to build a Neonate Glucose Simulator, able to generate reliable glucose time courses and usable to optimize the control strategy and avoid hyperglycaemic and hypoglycaemic episodes. The simulator must include a model of glucose-insulin-C-peptide interaction, possibly accounting for the metabolic processes known to be altered in infants born preterm, a model of both parenteral and enteral nutrition and a set of virtual neonates representative of real ones. The proposed Neonate Glucose Simulator was equipped with a population of 100 virtual neonates and the glucose time courses obtained for the virtual subjects were compared with real data collected on a population of infants born preterm, monitored during their stay in the NICU, demonstrating the reliability of the newly built tool.

Keywords: Other technological contributions to diabetes management, Diabetological education, Physiological modeling
Abstract: Despite advances in automatic insulin delivery systems, therapy optimization in type 1 diabetes remains constrained by one-size-fits-all strategies that don't account for individual patient preferences. This work presents a novel approach that leverages the University of Virginia Replay Simulator (UVA-RS) within a multi-objective optimization framework to address this limitation. This method enables the systematic exploration of trade-offs between competing objectives in diabetes management, such as reducing time in hypoglycemia range versus reducing time in hyperglycemia. By generating patient-specific digital twins and visualizing the Pareto front of possible outcomes, the proposed approach allows patients and healthcare providers to identify therapy configurations that best suit individual preferences and needs. Initial results demonstrate the feasibility of visualizing and selecting among different treatment configurations based on individual risk tolerance and management priorities.