Modeling glucose dynamics during physical activity using a linear model for individuals with Type 1 Diabetes

Abstract

For individuals with type 1 diabetes (T1D), regular exercise is beneficial as it can improve glycemic control, increase insulin sensitivity, improve body composition, and overall increase quality of life. However, physical activity can lead to complications for individuals with T1D, as it causes a fluctuation in glucose levels and increases the risk of hypoglycemia. This thesis focuses on the identification of both an average and patient-tailored linear time- invariant models that predict the glucose levels from the heart rate signal and the injected insulin records during a physical activity session of either aerobic, resistance, or interval exercise. Data from 177 adults with T1D who performed structured sessions of either aerobic, resistance, or high-intensity interval exercise at varying times of day were used. In this thesis, the simulation capabilities of the models were compared to a naive/simplistic model. The performance of the personalized model tested on all subjects in each exercise type for aerobic exercises: RMSE = 4.132 ± 6.480 R2 = 0.776 ± 0.197, for resistance exercises: RMSE = 6.026 ± 24.815 R2 = 0.752 ± 0.239, and HIIT exercises RMSE = 6.723 ± 35.676 R2 = 0.737 ± 0.239. The personalized model had a lower RMSE and higher R2 than both the average model and the naive/simplistic model.