Computational Intelligence, Volume 25, Number 3, 2009
TOWARD CASE-BASED REASONING FOR DIABETES MANAGEMENT:
A PRELIMINARY CLINICAL STUDY AND DECISION SUPPORT
SYSTEM PROTOTYPE
CINDY MARLING,
1
JAY SHUBROOK,
2
AND FRANK SCHWARTZ
2
1
School of Electrical Engineering and Computer Science,
Russ College of Engineering and Technology, Ohio University, Athens
2
Appalachian Rural Health Institute, Diabetes and Endocrine Center,
College of Osteopathic Medicine, Ohio University, Athens
This paper presents a case-based decision support system prototype to assist patients with Type 1 diabetes on
insulin pump therapy. These patients must vigilantly maintain blood glucose levels within prescribed target ranges
to prevent serious disease complications, including blindness, neuropathy, and heart failure. Case-based reasoning
(CBR) was selected for this domain because (a) existing guidelines for managing diabetes are general and must be
tailored to individual patient needs; (b) physical and lifestyle factors combine to influence blood glucose levels; and
(c) CBR has been successfully applied to the management of other long-term medical conditions. An institutional
review board (IRB) approved preliminary clinical study, involving 20 patients, was conducted to assess the feasibility
of providing case-based decision support for these patients. Fifty cases were compiled in a case library, situation
assessment routines were encoded to detect common problems in blood glucose control, and retrieval metrics were
developed to find the most relevant past cases for solving current problems. Preliminary results encourage continued
research and work toward development of a practical tool for patients.
Key words: case-based reasoning, diabetes management, intelligent decision support.
1. INTRODUCTION
People with Type 1 diabetes are unable to produce their own insulin, and thus they must
depend on exogenous insulin to survive. Maintaining good blood glucose control is a difficult
task for patients with Type 1 diabetes, who must keep their blood glucose levels as close to
normal as possible to maintain their health and avoid serious complications (The Diabetes
Control and Complications Trial Research Group 1993). Too little insulin results in elevated
blood glucose levels, called hyperglycemia, which can lead to numerous complications over
time, including blindness, neuropathy, and heart failure. Too much insulin results in depressed
blood glucose levels, or hypoglycemia, during which patients may experience weakness,
confusion, dizziness, sweating, shaking, and if not treated promptly, loss of consciousness
or seizure.
A person with Type 1 diabetes tries to maintain blood glucose levels between assigned
high and low targets and monitors actual levels through finger stick testing from four to
six times a day. A patient on insulin pump therapy is continually infused with basal insulin
at varying rates throughout the day and can instruct the pump to deliver additional doses,
called boluses, before meals and to correct for hyperglycemia. The amount of bolus insulin
depends on the amount of carbohydrate to be consumed, the current blood glucose level,
the anticipated level of physical activity and the patient’s own historical response to insulin.
Insulin pump therapy differs from conventional insulin therapy, in which patients use syringes
to inject themselves with a predetermined amount of insulin three or four times a day. With
This is an update and expansion of the paper “Toward Case-Based Reasoning for Diabetes Management,” which was
presented at the Workshop on Case-Based Reasoning in the Health Sciences at the International Conference on Case-Based
Reasoning (ICCBR-07).
Address correspondence to Cindy Marling, Russ College of Engineering and Technology, Ohio University, Athens, OH
45701; e-mail: marling@ohio.edu
C©
2009 The Authors. Journal Compilation
C©
2009 Wiley Periodicals, Inc.

166 COMPUTATIONAL INTELLIGENCE
conventional insulin therapy, there is less opportunity to fine-tune control or to account for
variations in daily routine.
In our experience, physical and lifestyle factors frequently combine in complex ways
to impact blood glucose levels in people with Type 1 diabetes. Consider, for example, a
patient of the third author who is on his high school wrestling team. Exercise, including
participation in competitive sports, is known to be beneficial in controlling blood glucose
levels. However, this patient’s blood glucose could be too high, too low, or within range
when he wrestled, depending on (a) whether he was at practice or at a competitive meet and
(b) whether he won or lost his match. Therefore, to help each patient maintain good blood
glucose control, despite individual variations and complex physical/social interactions, we
are exploring case-based decision support. While we are not the first to employ case-based
reasoning (CBR) in the diabetes domain (Bellazzi et al. 2002), we are the first to use it
as the primary reasoning modality for supporting patients with Type 1 diabetes on insulin
pump therapy. Our goal is to build a CBR therapy adviser that will help patients improve and
maintain their control while reducing the data analysis workload on physicians.
New continuous glucose monitors (CGM) that record glucose values every 5 min,
coupled with insulin pump technology, make it possible for patients to electronically record
and send their data to physicians on a daily or weekly basis. Because data is automatically
collected, but not automatically analyzed, this puts a tremendous burden on the physician to
review blood glucose records and then make appropriate adjustments in insulin dosages. The
third author, for example, has over 700 patients on insulin pump therapy and may make over
20 therapeutic adjustments per day. To reduce this data overload, we began investigating ways
to automatically analyze these records and provide appropriate therapeutic recommendations.
We envision that such recommendations would initially be provided to physicians for review,
but might eventually, once proven safe and effective, be provided directly to patients.
CBR, which has been successfully applied to managing other long-term medical con-
ditions (Bichindaritz 1996; Bichindaritz, Kansu, and Sullivan 1998; Montani et al. 2003b;
Marling and Whitehouse 2001), seems especially appropriate for this domain. Type 1 di-
abetes is marked by a wide variability among patients in terms of sensitivity to insulin,
response to environmental and lifestyle factors, propensity for complications, compliance
with physician’s orders, and response to treatment. Guidelines for managing the disease are
well established (American Diabetes Association 2003), but they are general in nature and
must be tailored to the needs of each patient. The opportunity for cases to complement
guidelines in such situations was presented in (Bichindaritz and Marling 2006). We have
built a case-based decision support system prototype to assist patients with Type 1 diabetes
on insulin pump therapy. Preliminary results are encouraging, and work continues toward a
practical tool for patients.
2. REPRESENTING DIABETES MANAGEMENT KNOWLEDGE IN CASES
2.1. Case Design
A case represents knowledge by storing (a) the description of a specific problem that has
occurred; (b) the solution that was applied to that particular problem; and (c) the outcome
of applying the solution to that problem (Kolodner 1993). According to (Kolodner 1993),
in representing a problem, it is necessary to include all information that is typically used to
describe such a problem as well as all information that is explicitly taken into account by a
human problem solver in solving the problem. Typical information used to describe diabetes
management problems includes blood glucose target levels, actual blood glucose levels,