2014 IEEE 11th International Conference on e-Business Engineering

A Patient-centered Self-care Support System for Diabetics

Chien-Ho Wu
Department of Statistics and Information Science
Fu Jen Catholic University
New Taipei City, 24205 Taiwan
[email protected]

Abstract—According to WHO statistics there are about 347 problem and develop a diabetes care support system called
million people worldwide having diabetes. Unfortunately cures for ScasDia. It is developed using open-source ware with the
both type 1 and type 2 diabetes have proved elusive to medical ability to analyse daily measures and build ARIMA models
science. The causes of diabetes are a complex, but are in large on selected diabetes indicators for forecasting purposes. In
part due to rapid increases in overweight, obesity and physical the end ScasDia is able to help improve the quality of self-
inactivity. In general to help prevent diabetes and its complications, care and quality of life for diabetics.
people should achieve and maintain healthy body weight, be
physically active, eat a healthy diet, and avoid tobacco use. II. BRIEF SURVEY
Achieving this goal requires a patient-centered approach on the
part of healthcare. In this research we pull together IT in public A. Diabetes
domain and propose a system, called ScasDia, with features of
reminders, plotting and advance-warnings that can benefit Diabetes is a metabolic disorder of multiple etiology. It is
diabetics by improving the quality of self-care. ScasDia, a noncommunicable chronic disease that occurs when the
implemented in Java, facilitates the recording of data related to pancreas does not produce enough insulin, or when the body
diet control, physical exercises and diabetes indicators. cannot effectively use the insulin it produces.
Nonetheless it builds ARIMA models with R on selected indicators Hyperglycaemia is a common effect of uncontrolled diabetes
so that predictions for the selected indicators can serve as advance and over time leads to serious damage, dysfunction and
warnings on the status of diabetes care. ScasDia can help failure of various organs (WHO 1999). Complications
reduce risk factors that contribute to the development of related to diabetes include stroke, kidney failure,
complications. amputations, cardiovascular diseases and premature death et
cetera. Report has also shown that 50% of people with
Keywords- Diabetes; ARIMA; Java; R diabetes die of cardiovascular disease, primarily heart
disease and stroke (Morrish 2001). Good metabolic control
I. INTRODUCTION prevents or delays these complications. In fact the overall
risk of dying among people with diabetes is at least double
Diabetes is a chronic progressive disease. It can be
the risk of their peers without diabetes (Roglic et al. 2005).
prevented but is not yet reversible if developed. According to
There are two major forms of diabetes. Type 1 diabetes
statistics from WHO there are about 347 million people
(T1D) is characterized by a lack of insulin production. T1D
worldwide having diabetes. It is predicted to become the
usually develops in childhood and adolescence. Patients of
seventh leading cause of death in the world by the year 2030
T1D require lifelong insulin injections for survival. Type 2
(WHO 2013).
diabetes (T2D) results from the body's ineffective use of
Diabetes care is a lifelong responsibility and takes round-
insulin. T2D usually develops in adulthood and is related to
the-clock commitment. Careful diabetes care can reduce risk
obesity, lack of physical activity, and unhealthy diets. Other
of serious, even life-threatening, complications. Preventing
categories of diabetes include gestational diabetes and other
diabetes and its complications requires a patient-centered
rarer causes. T2D accounts for around 90% of all diabetes
approach on the part of healthcare.
worldwide (WHO 2013).
A diabetes treatment plan usually dictates a patient to
Intermediate states of hyperglycemia (impaired fasting
achieve and maintain healthy body weight, be physically
glucose, IFG or impaired glucose tolerance, IGT) have also
active, eat a healthy diet, and avoid tobacco use. Also
been defined. These states can progress to diabetes but with
tracking clinical indicators has been shown to improve the
the possibility of being prevented or delayed (WHO 1999).
quality of diabetes care, ultimately improving the outcomes
In people who do not have diabetes, the normal range of
for people with diabetes (NIH 2013, Rossi et al. 2011).
blood /plasma glucose concentration level is about 70 to 120
Despite the recognized importance of tracking clinic
mg/dl. Blood glucose goes up after eating but 1 or 2 hours
indicators and criticality of strict commitment to the
later returns to the normal range. Blood glucose targets for
treatment plan, there is no proper tool that is cost-effective
most people with diabetes are 70 to 130 mg/dl before meals
available to the patients to help achieve those tasks. To help
and below 180 mg/dl 1 to 2 hours after the start of a meal.
patients of type 2 diabetes track daily measures and achieve
Another test called HbA1c can be done to approximate
treatment plan, we adopt a systems approach to address the

978-1-4799-6563-2/14 $31.00 © 2014 IEEE 298

DOI 10.1109/ICEBE.2014.58
metabolic control over previous 2-3 months and to guide can act as advanced warnings on the possible adverse change
treatment decisions. An HbA1c of 5.7% or less is normal. of conditions of people with diabetes.
For many people with diabetes the goal is to keep the level at
or below 6.5-7% (WHO 2011). C. Arima Model
Diagnosis is made by the presence of classic symptoms Any finite or infinite sequence of uni-variate
of hyperglycaemia and an abnormal blood test. Signs of observations {Xt : tT, where T is an ordered set of time} is
diabetes include polyuria, polydipsia, feeling very hungry or called time series. If the time index t can take all the values
tired, having sores that heal slowly, having blurry eyesight, in an interval of real numbers, the series is continuous. A
or having dry, itchy skin et cetera (NIH 2013). series is discrete when the set of the possible values t is a
The overall aim of treatment for both T1D and T2D is to discrete set (Dufour 2003). We generally assume that time is
bring the elevated blood sugars down to a normal range, so a discrete variable. A time series can be viewed as one
as to improve symptoms of diabetes as well as to prevent or possible realization of its underlying stochastic process (or
delay diabetic complications. In general daily care of type 2 data generating process, DGP) defined on a common
diabetes shall focus on the following actions (NIH 2013). probability space (Kunst 2009).
x Develop and follow a diabetes meal plan. Don't skip In order to create a statistical framework and to
meals. formalize the problem of prediction, the underlying
x Be physically active. Physical activity helps stay stochastic process of a time series is assumed to be stationary.
healthy. Try to be active almost every day for a total The stationarity for the corresponding moments of the
of about 30 minutes. stochastic process {Xt} are defined as follows (Kirchgassner
x Check blood glucose as recommended. The closer to & Wolters 2007).
normal the blood glucose is, the lower the chances x Mean stationarity: A process is mean stationary if
are of developing serious health problems. Levels of E(Xt) = Pt = P is constant for all t.
ketones in blood or urine may be checked to prevent x Variance Stationarity: A process is variance
the risk of having ketoacidosis which can cause stationary if V[Xt] = E[(Xt - Pt)2] = V2 is constant
death. and finite for all t.
x Take diabetes medicines every day if prescribed. x Covariance Stationarity: A process is covariance
x Keep daily records. By keeping daily records there is stationary if Cov[Xt, Xs] = E[(Xt - Pt)( Xs - Ps)] =
a better chance of telling how well the diabetes care J(|s–t|) is only a function of the time distance
plan is. between the two random variables and does not
Monitoring and controlling the levels of diabetes depend on the actual point in time t.
indicators, including HbA1c, fasting plasma glucose (FPG), A stochastic process is weakly stationary when it is of
postprandial plasma glucose (PPG), body mass index (BMI), mean and covariance stationarity. For most results and
blood pressure and lipids, is an integral component of techniques of time-series analysis, this concept suffices.
diabetes care (Calvert et al. 2009, Rossi et al. 2011). One of the general objectives of time series analysis is to
develop models for describing the behavior of time series.
B. Systems for Diabetes Care
ARIMA models are, in theory, the most general class of
Since diabetes is one of many types of chronic diseases, models for forecasting a time series. A seasonal ARIMA
self-monitored care plays an important role in improving and model is classified as an ARIMA(p, d, q)x(P, D, Q) model,
maintaining the quality of life for diabetic patients. There is where P=number of SAR terms, Q=number of SMA terms,
no doubt that proper self-monitored care can not only reduce D=number of seasonal differences. Very often we rely on
the risk of developing complications but also extend ACF and PACF plots to decide on the possible values of (p,
accordingly the life expectance of people with diabetes. d, q)x(P, D, Q).
Information technology is a powerful tool that has The Box-Jenkins approach to building an ARIMA model
profound impacts on all aspects of human society. In consists of the following three steps (Box et al. 2008).
particular, internet-based technology has been used to 1) Identification: Determining the values of
improve quality and efficiency of care for patients of type 2 (p, d, q)x(P, D, Q).
diabetes. Studies have shown that internet-based health care
2) Estimation of ARIMA parameters.
service can achieve better glycemic control and may provide
effective and safe diabetes management for diabetic patients 3) Diagnostic checking. We should check that the
(Kwon et al. 2004, Cho et al. 2006, Lim et al. 2011, Kang et residuals of the resulting model follow the
al., 2012). assumptions given below.
Although systems approach to supporting self-monitored a) Normality: .
care of diabetes has been shown to be effective and
beneficial to diabetics, current systems alike have not yet b) Constant variance: .
included functions for building predicative models on control c) Independence: .
indicators such as HbA1c, SBP and DBP etc. It is deemed The base, TSA and forecast packages of R provide most
that predicative models on control indicators have the of the functions needed for building and forecasting with
potential of providing information on future directions and ARIMA models.
trends of variations of control indicators. This information

299
 III. SYSTEM DESIGN
Following the NIH guide to diabetes (NIH 2013) and the
idea of advance warnings for prevention we pull together
related open-source ware in public domain and develop a
tentative system, i.e. ScasDia, that can help not only track
daily diabetes records but also build forecast models on
selected diabetes indicators for forecasting purposes.
A. System Functions
Fig. 1 is the use case diagram of ScasDia. At this moment
ScasDia is designed for users that assume roles as either
record keeper or care professional. A record keeper can use
the system to track and maintain daily records about meals,
physical exercises and diabetes indicators such as blood
pressures and concentration levels of blood glucose. Besides
record keeping, users of ScasDia can receive preliminary Figure 2. System block diagram.
recommendations on physical fitness and calories taken from
meals. Furthermore users can plot line charts and time series
charts of selected diabetes indicators. A care professional can IV. IMPLEMENTATION
also exploit the model building function to predict values of
A. Deployment Requirements
diabetes indicators in the immediate future.
ScasDia is implemented in Java with NetBeans IDE as an
application in the Windows environment. Any platform
conforming to the following requirements shall be able to
run the system.
x Java runtime environment (SE. 1.7v or later)
x R statistical software (rJava/jri, RODBC incl.)
x ODBC manager
x MySQL(5.5v or later) with JDBC connector and
MySQL ODBC driver
System environment variables that must be set for the
sake of running the system include R_HOME,
JAVA_HOME, CLASSPATH and PATH.
B. Selected Screen Shots
Selected screen shots of ScasDia are shown from Fig. 3
to Fig. 5. The zigzagged first line in the upper left of Fig. 3 is
the line chart of recorded blood sugar tests 2HR after meal in
the morning and the wiggly second line is the line chart of
Figure 1. Use case diagram of ScasDia
recorded blood sugar tests before meal in the morning. The
line charts on the upper right of Fig. 3 are charts of blood
B. System Architecture sugar tests recorded in the afternoon.
The system contains five modules as shown in Fig. 2.
The GUI model is the interface that interacts with users of
the system. The data collection module is in charge of the
recording of daily details on food taken, physical activities,
and diabetes indicators. Data collected are stored in a
MySQL database. The data maintenance module provides
the house keeping functions for data stored in the database.
By referring to the data stored in the database, the suggestion
module provides recommendations about the suggested
intake of daily calories and preliminary diagnosis on the
frequency and strength of physical exercises. The forecasting
module builds, either automatically or manually, ARIMA
models by invoking model building functions in R through
Figure 3. Line charts of blood sugar and blood pressures.
the jri of rJava package (Venables 2013, Wu et al. 2012, Wu
2013, Wu et al. 2013).

300
 Figure 4. ARIMA model summary for sample data.

and keep daily records. There is no argument that diabetes

care is a lifelong responsibility requiring 24-hour
commitment.
To help reduce the burden of carrying out a care plan and
improve the quality of diabetes care, we have developed a
support system with functions to remind daily activities,
track daily records, plot recorded data, and build forecast
models on selected diabetes indicators with predictions for
advance warnings of possible adverse outcomes. The
tentative system is developed in Java and invokes R for
plotting and statistical functions. No commercially
proprietary code is employed in the implementation. In the
future, this research can be extended in the following
directions.
x Transform ScasDia into a web-based application
(Wu 2014), so that records related to the monitoring
Figure 5. Time series chart with forecasts for advance warning. and controlling of the levels of diabetes indicators
can be shared with patient’s care group on a real
Fig. 4 is the model summaries built for the sample data time basis. As a result of online sharing of daily
on diastolic blood pressures, and Fig. 5 is the time series records, risk factors that contribute to the
chart for the sample data with predictions for advance development of complications can be better
warning purposes. monitored and controlled.
Forecast models for diabetes indicators in ScasDia are x In light of the rapid growth of mobile devices,
built by invoking R functions. For the same set of data, the exploring the possibility of porting the GUI module
details of a model in ScasDia may be different from those of ScasDia to a mobile platform would prove
built by other software such as SPSS or SAS. This beneficial in the long run.
difference may due to the fact that different algorithms are x Collaborate with dietitians and doctors to provide
adopted in different software to compute the estimates of the functions adapted to their needs so that care for
model. patients can be more complete.
ScasDia is cost-effective and has the potential to
V. FINAL REMARKS AND FUTURE WORK
improve the quality of diabetes care, ultimately leading to an
Diabetes is a chronic disease. The overall risk of dying improvement of life over time.
among people with diabetes is at least double the risk of
their peers without diabetes. According to NIH guide, care ACKNOWLEDGEMENT
for type 2 diabetes requires patients to be physically active, Thanks to Jei-Wun Jheng and Yi-Ting Miao for
eat a healthy diet, achieve and maintain healthy body weight, participating in the programming of the system. We are also
check blood glucose, take diabetes medicines if prescribed,

301
obliged to Huei-Ci Jiang, Chia-Hsin Wu and Wan-Jhen Liu [11] WHO, Definition, diagnosis and classification of diabetes mellitus
for helping out with clerical errands. and its complications. Part 1: Diagnosis and classification of diabetes
mellitus (WHO/NCD/NCS/99.2), Geneva, World Health
Organization, 1999.
REFERENCES
[12] WHO, Use of Glycated Haemoglobin (HbA1c) in the Diagnosis of
[1] G. E. P. Box, G. M. Jenkins and G. C. Reinsel, Time Series Analysis, Diabetes Mellitus: Abbreviated Report of a WHO Consultation
Forecasting and Control, 4th ed. NJ: Wiley, 2008. (WHO/NMH/CHP/CPM/11.1), Geneva, World Health Organization,
[2] M. Calvert, A. Shankar, R.J. McManus, H. Lester & N. Freemantle. 2011.
"Effect of the quality and outcomes framework on diabetes care in [13] WHO, 10 Facts About Diabetes, Fact sheet N°312, 2013.
the United Kingdom: retrospective cohort study," BMJ, 338(2), 2009,
[14] C. H. Wu, J.B. Li, and T.Y. Chang, "Implementing Statistical Agents
pp. b1870-b1870.
on JADE Platform," Applied Mathematics & Information Sciences,
[3] Jean-Marie Dufour, Introduction to time series analysis, University 6( 2S), 2012, pp. 379S–385S.
of Montreal, Canada, 2003.
[15] C. H. Wu, "Arima Models are Clicks Away," Applied Mechanics and
(http://www2.cirano.qc.ca/~dufourj/Web_Site/ResE/Dufour_1998_C
Materials, Vol. 411-414, 2013, pp. 1129-1133.
_TS_IntroductionTS.pdf)
[16] C. H. Wu, "A Web-based Multiple Regressor for Data Analysis using
[4] G. Kirchgassner and J. Wolters, Introduction to Modern Time Series
R," International Journal of Computer and Communication
Analysis, Berlin: Springer, 2007.
Engineering, 3(3), 2014, pp. 216-220.
[5] R. M. Kunst, Applied Time Series Analysis-Part I, University of
[17] C. H. Wu , J.B. Li and T.Y. Chang, SlinRA2S: "A Simple Linear
Vienna and Institute for Advanced Studies Vienna. (2009)
Regression Analysis Assistor," Proceedings of ICEBE, 2013, pp.
(http://homepage.univie.ac.at/robert.kunst/ts1.pdf)
219-223.
[6] N. J. Morrish, S. L. Wang, L. K. Stevens, J. H. Fuller and H. Keen,
[18] H. S. Kwon, J. H. Cho and H. S. Kim et. al., "Establishement of
"Mortality and causes of death in the WHO Multinational Study of
Blood Glucose Monitoring System Using the Internet, " Diabetes
Vascular Disease in Diabetes," Diabetologia, 44 Suppl 2, 2001, pp.
Care, 27(2), 2004, pp. 478-483.
S14–S21.
[19] J. H. Cho, S. A. Chang and H. S. Kwon et al., "Long-Term Effect of
[7] NIH, "Your Guide to Diabetes:Type 1 and Type 2," NIH Publication
the Internet-based Clucose Monitoring System on HbA1c Reduction
No. 13–4016, Maryland, National Institutes of Health, 2013.
and Glucose Stability," Diabetes Care, 29(12), 2006, pp. 2625-2631.
[8] G. Roglic, N. Unwin, P.H. Bennett, C. Mathers, J. Tuomilehto, S.
[20] S. Lim, S. M. Kang and H. Shin et al., "Improved Glycemic Control
Nag et al., "The burden of mortality attributable to diabetes: realistic
Without Hypoglycemia in Elderly Diabetic Patients Using the
estimates for the year 2000," Diabetes Care, 28(9), 2005, pp. 2130–
Ubiquitous Healthcare Service, a New Medical Information System,"
2135.
Diabetes Care, Vol. 34, 2011, pp. 308-313.
[9] M.C.E. Rossi, G. Lucisano, M. Comaschi, C. Coscelli, D. Cucinotta,
[21] S. M. Kang, M. J. Kim and H. Y. Ahn et al., "Ubiquitous Healthcare
P. Di Blasi, G. Bader et al., "Quality of diabetes care predicts the
Service Has the Persistent Benefit on Glycemic Control and Body
development of cardiovascular events: results of the AMD-QUASAR
Weight in Older Adults With Diabetes," Diabetes Care, Vol. 35,
study," Diabetes Care, 34(2), 2011, pp. 347-52.
2012, p.e19.
[10] W. N. Venables, D. M. Smith, and the R Core Team, An Introduction
to R, The R Project, 2013.

302