Abstract
With the ubiquity of wearable technologies as well as the plethora of smartphone applications promoting the quantified-self, quantitative tracking of daily life information (e.g., physical, behavioral and environmental) that previously could only be obtained through self-reporting is now easily accessible to both consumers and healthcare providers. These technologies have offered remarkable opportunities to advance personalized medical care. This dissertation focuses on developing a data-driven personalized treatment support system to improve daily care in a chronic disease which generates ubiquitous data as a part of its regular treatment, type 1 diabetes (T1D).
T1D is an autoimmune disease characterized by destruction of the pancreas’s insulin producing β-cells leading to absolute insulin deficiency and requires life-long insulin replacement. As a result, patients with T1D use devices to self-monitor and -regulate their blood glucose levels in everyday life. This regulation is, in fact, a dynamic optimization problem where patients try to achieve their target blood glucose levels by deciding on the timing and amount of insulin injections while challenged by the varying insulin needs of their body. In the presence of factors that affect glucose metabolism significantly, such as meal intake, physical activity, or stress, this optimization problem becomes even more complicated. Physical activity is one of the major daily life factors that affect glucose metabolism leading to hurdles in blood glucose management. A significant amount of work has been devoted to the development of strategies to help blood glucose management surrounding a structured activity (exercise), though little attention has so far focused on investigating the impact of unstructured (non-exercise) physical activity, which forms most of the daily physical activity.
We present a novel physical activity informed glucose control system originating from data collection through safety and feasibility testing in a clinical trial. By leveraging information from commercially available activity trackers, this work (1) improves the understanding of both exercise and non-exercise physical activity’s impact on glucose control in patients with T1D, (2) quantifies the physical activity induced changes in insulin needs over several hours, and (3) develops personalized treatment support systems for enhancing physical activity related blood glucose management in daily life. Specifically, we first examine the association between daily physical activity and blood glucose control by conducting multi-level statistical analyses on data collected from patients via wearable activity trackers. Using the insights gained from these analyses and available literature, we then develop a physical activity informed insulin dosing system adapted to the patient’s routine physical activity and blood glucose profile. Concepts of system dynamics, computer simulation, and data-driven optimization are used to obtain personalized treatment parameters and pre-clinical trial testing. The safety and feasibility of the proposed approach are demonstrated in a clinical trial. As an extension, this work also proposes a method specific to optimal treatment adjustments following a detected and/or self-announced exercise.
test
