Interest in artificial pancreas (AP) systems has increased as hybrid-APs have become available for use in daily life. These first-generation systems require manual announcements of meals and exercise, causing management burden to users. Interest in use of additional signals such as heart rate or accelerometer signals to automate and improve AP performance during physical activities is growing. While T1D simulators which accurately model glucose-insulin dynamics exist, they do not provide physiological signals to allow for rapid-prototyping of multivariable AP systems. A multivariable simulator enables rapid advances in multivariable AP research.
The effect of exercise on glucose dynamics is combined with the Hovorka glucose-insulin dynamic model. Additional models compute physiological signals (heart rate, energy expenditure, and accelerometer values) with their dynamics driven by physical activity. Data for these signals was collected in clinical studies of people with T1D and are used to characterize the interpatient variability and determine realistic parameters for the virtual subjects.
The root mean square error for the original Hovorka model (12.86±6.37mg/dL) is reduced (9.85±5.13mg/dL) when validated against 18 adults with T1D (p=6.4x10-7). This simulator (mGIPsim) is freely available for academic research. Users of mGIPsim provide the meal and exercise scenarios and insulin infusions. Model equations are then solved to yield the output variables (physiological biosignals, blood glucose concentration, CGM, and plasma insulin concentration).
A multivariable T1D simulator allows for rapid prototyping of physical activity-oriented AP system that can utilize physiological signals such as heart rate or energy expenditure in their algorithms.