Meta-analysis of technologies for diabetes treatment: glycemic control, prediction, meal and physical activity detection

is a widespread chronic disease with steadily growing prevalence and associated comorbidities. Current treatment of diabetes can be quite cumbersome for the patients, leading to global efforts to develop a fully-automated artificial pancreas. Such a device will need to employ some form of blood glucose prediction, as well as algorithms to detect meal intake and various physical activities. Many methods were already developed for these tasks, enabling the meta-analysis of the current state of the art. First, an overview of glycemic control strategies and sensors is provided. Then, the relevant studies are introduced and described prior to the meta-analysis. The resulting meta-analysis quantifies the accuracy of prediction models for the various prediction horizons (15, 30, 45, 60, and 120 min) and the performance of meal and physical activity detection models using sensitivity and metrics related to false-positivity. Following the observed patterns across the prediction horizons, a novel approach to evaluating the physiological plausibility of prediction methods is proposed. The baseline state-of-the-art model performance for said tasks is estimated. Finally, a discussion about the current issues in the research of diabetic technologies and their potential solutions is conducted.