State-Space Modelling of Continuous Glucose Monitors

Continuous glucose monitors (CGM) frequently measure glucose levels in interstitial fluid (ISF), subject to multiple sources of uncertainty. This study reconstructs a model of blood glucose using CGM data through a state-space model with a Kalman filter, a Rauch-Tung-Striebel (RTS) smoother, and an expectation-maximization-based parameter estimation. An offline setting is chosen to focus on understanding glucose transport and sensor dynamics rather than real-time prediction. True blood glucose and CGM data were generated from a physiological diabetes simulator (Simglucose[6]). True blood glucose is used as a driving signal in the simulation. Most CGM research prioritizes online application and primarily relies on forward processing due to the lack of future data in real-time settings. This setting provides an explanatory, offline view of model behaviour and highlights how simplified linear models can be used to explore CGM sensor dynamics and physiological mechanisms.