Glucose Dynamics: How to Predict Them Efficiently in People with Type 1 Diabetes

Physical activity, meals, and individual insulin sensitivity all influence glucose dynamics in people with type 1 diabetes. However, these factors affect the metabolism differently from person to person, making it difficult to develop a model that accurately reflects each individual. Matteo Ragni, Chiara Toffanin, Paolo A. Mongini, and Lalo Magni took on the challenge of developing a dynamic model that accounts for individual differences and daily fluctuations experienced by people with diabetes.

In March, the researchers published the paper “Periodic Predictor of Glucose Dynamics in Type 1 Diabetes Patients”, introducing an algorithmic approach that combines three separately trained models. Each model is associated with one of the main daily periods (breakfast, lunch and dinner) where changes in glucose-insulin dynamics and insulin sensitivity are particularly relevant. Transitions between these models occur smoothly, ensuring that glucose estimation remains accurate throughout the day and physiologically realistic.

Comparing two Estimation Methods

One of the main goals of the study was to determine which estimation technique performs best for periodic glucose prediction and how these approaches compare with invariant models. To investigate this, the researchers evaluated the predictors on 100 virtual adult patients using the UVA/Padova metabolic simulator. The first technique, the Kalman filter, is an algorithm that continuously combines the most recent prediction with new measurements. This enables the system to immediately correct its estimates and improve prediction accuracy in real time. The second technique, the Moving Horizon Estimator, works differently. Instead of relying only on the latest state, it considers a time window containing several past measurements. This allows the system to incorporate a broader history of glucose dynamics into its predictions. Both estimation techniques were then compared with invariant predictors, which do not account for fluctuations over the course of the day caused by factors such as physical activity or dietary changes.

Artificial Pancreas – Improving Glucose Prediction for Future Automated Insulin Delivery

One of the key findings of the study is that the periodic predictor improves prediction performance with both estimation methods. However, each technique offers different advantages, making them suitable for different clinical settings. While the Kalman filter enables computationally efficient real-time adaptation, the Moving Horizon Estimator provides greater flexibility and can incorporate physiological constraints, which may improve robustness and safety. Despite these differences, the authors demonstrate that periodic predictors, which combine multiple time-dependent models and account for glucose fluctuations throughout the day, outperform invariant approaches.

The findings could further improve automated insulin delivery systems and warning systems for dangerous glucose levels. In the long term, such approaches may help prevent sudden episodes of hypoglycemia and hyperglycemia and reduce the daily burden of diabetes management for people living with type 1 diabetes.

Read the full article here: Matteo Ragni, Paolo Alberto Mongini, Lalo Magni, Chiara Toffanin. 2026. Periodic predictor of glucose dynamics in type 1 diabetes patients. IFAC Journal of Systems and Control 35. https://doi.org/10.1016/j.ifacsc.2026.100401