1. Non-Invasive Glucose Modeling With Limited Data For Insulin
Dependent Diabetic Subjects
Marlie J. Quintero1
, Tricia Salinas2
and Derrick K. Rollins Sr.1
1
Department of Chemical and Biological Engineering
2
Department of Mechanical Engineering
Iowa State University, Ames, Iowa 50011
There are 25.8 million children and adults in the United States suffering from diabetes;
that is 8.3% of our country’s population.1
Type I diabetes, otherwise known as juvenile diabetes,
causes extreme shifts in blood glucose concentration (BGC). The inability to control these
extreme variations can cause hypoglycemia, hyperglycemia, and organ degeneration. The current
methods of controlling glucose levels provide a limited ability to control within healthy range of
BGC; they are unable to consistently estimate a proper dose of insulin for the patient. As a result,
there is a great need for an automatic device that administers the correct dose of insulin to the
body; the device would work best if it is closed-loop, that is, the individual is not required to
make insulin dosage calculation because the process is completely automated. To create a closed-
loop system, we must first accurately model the blood glucose concentration (BGC) to predict
glucose levels due to food consumption, various activities or changes in the body. The focus of
the Rollins’ research group at Iowa State University is to create a subject specific glucose
modeling method which takes into account the subject’s food, activity, and stress levels.
Type 1 diabetes cases are modeled in which carbohydrates, bolus insulin and basal insulin
are the input variables or disturbances. The data obtained from these three variables was
collected for 3 to 5 days modeled using a modified version of the Weiner block-oriented method
of Rollins el al.3
This method is modified to give a better physiological behavior of glucose. The
objective of this work is to determine if about 3 days is adequate to develop an accurate glucose
model. Results are given for several subjects for input only models, input and output models and
output only models. Predictive modeling results are also given for 5, 30, 60 and 90 minutes into
the further. While some cases gave very good results, in general it appears that more than three
days of data are needed to have high certainty that enough data has been collected to develop an
accurate model.
1. Subject Specific Multiple Input Block-Oriented Glucose Modeling for Type 1 Diabetic Subjects
2. American Diabetes Association
3. D.K. Rollins, N. Bhandari, J. Kleindler, K. Kotz, A. Strohbehn, L. Boland, M. Murphy, D. Andre, N. Vyas, G. Wlk, and W.E. Franke. Free-
living inferential modeling of blood glucose level using only noninvasive inputs. Journal of Process Control, 20(1):95-107, 2010.