Generalised stochastic model for characterisation of subcutaneous glucose time series

A generalised stochastic model with second order differential equations is proposed to describe the response of blood glucose concentration to meals in groups of nondiabetic people and two types of diabetic patients. A variational Bayesian approach is applied in order to infer parameters of the models, and the best model was selected based on the computed log-evidence for each prandial event. The model with a linear structure represents most of the events, while the nonlinear terms need to be included more frequently for Type II diabetic patients. This indicates different physiological mechanisms of glucose absorption for different groups. The deterministic parameters and intensities of stochastic components are compared by groups using the ANOVA test, and the results show significant differences between the groups. This model can potentially be used for long term prediction of the glucose concentration response to external stimuli.