Prediction of Patient's Individual Blood Glucose Levels from Home Monitored Readings of Type I Diabetics

Health Information Management and Technology Department, College of Applied Medical Sciences, University of Dammam, Dammam, KSA

^*Corresponding author:

A.Karim Jabali, Associate Professor

Health Information Management and Technology Department

College of Applied Medical Sciences

University of Dammam

Dammam 31441

P.O.Box (1982)

Eastern Province, KSA

Email: aajabali@ud.edu.sa

Abstract

In this paper, a comparison of two different approaches that can be used in developing time series mathematical models (MM) of diabetes mellitus was carried out. The trade-off should be considered between the complexity of the model and its accuracy to predict future glucose concentration. This work is a continuation of the author’s work and results obtained previously and showed the potential and superiority of using autoregressive with exogenous terms (ARX) model in describing the dynamics of diabetes. Moreover, it is shown that despite the models are of general form but they are different depending on individuals' regimen of diabetes management. The last fact was demonstrated by using six diabetic patients' records, with rich information about their life style and treatment program, to derive models. In addition to that an answer is given to two main questions: how many future samples of glucose levels can be predicted with acceptable accuracy and what is the acceptable order of the model –complexity-, if the prediction horizon is specified. Both types of models were developed, tested and compared. This work emphasizes the fact that diabetes management plan should be formulated as an individualized therapeutic to achieve the desired level of diabetes control. This can be of help in improving the metabolic control of type-1diabetes patients by implementing these characteristics and models in both computerized controlled decision support system and simulation systems for educating and training of healthcare professional staff. Additionally, these MM of glucose-insulin interaction are expected to aid in reaching a generalized model.

Keywords: diabetes, time series, modeling, prediction.

Download the full article (PDF)