A time series analysis : exploring the link between human activity and blood glucose fluctuation
Eric A. Sadowski · Scholars Commons (Wilfrid Laurier University) · 2010
In this thesis, time series models are developed to explore the correlates of blood glucose (BG) fluctuation of diabetic patients. In particular, it is investigated whether certain human activities and lifestyle events (e.g. food and medication consumption, physical activity, travel and social interaction) influence BG, and if so, how. A unique dataset is utilized consisting of 40 diabetic patients who participated in a 3-day study involving continuous monitoring of blood glucose (BG) at five minute intervals, combined with measures for sugar; carbohydrate; calorie and insulin intake; physical activity; distance from home; time spent traveling via public transit and private automobile; and time spent with other people, dining and shopping. Using a dynamic regression model fitted with autoregressive integrated moving average (ARIMA) components, the influence of independent predictive variables on BG levels is quantified, while at the same time the impact of unknown factors is defined by an error term. Models were developed for individuals with overall findings demonstrating the potential for continuous monitoring of diabetic (DM) patients who are trying to control their BG. Model results produced significant BG predicting variables that include food consumption, exogenous insulin administration and physical activity.