A Nonlinear Data-Driven Model of Glucose Dynamics Accounting for Physical Activity for Type 1 Diabetes: An In Silico Study

Physical activity is an important physiological information which should be taken into account by artificial pancreas to achieve optimal control of blood glucose in Type 1 Diabetes patients. An accurate glucose dynamic model with physical activity as an additional input is highly desirable for the next generation artificial pancreas. In this paper, we present a nonlinear data-driven model that captures both the insulin-independent and -dependent effect of physical activity, especially the prolonged effect of physical activity on insulin sensitivity that can last 24–48 hours post exercise. The model was identified and validated using data sets generated by a physiological glucose-exercise model under a clinical training protocol. Compared to modeling the effect of physical activity as a linear additive term only in a glucose dynamic equation, the proposed nonlinear model showed significant improvement of prediction accuracy in all three metrics, particularly in large prediction horizons (P < 0.05). Further investigation in time-series data indicates that the improvement mainly resulted from the better prediction of glucose around the first meal time after exercise (6 to 8 hours after the meal was taken).