Knowledge Modeling in the Health Care Domain to Support Software
Development & Maintenance
Thomas Reichherzer, John Coffey, Bilal Gonen and Irad Gillett
Dept. of Computer Science, University of West Florida, 11000 University Pkw., Pensacola, Florida, 32503 U.S.A.
Keywords: Knowledge Modeling, Concept Mapping, Software Engineering, Health Information Systems.
Abstract: This article contains a description of a knowledge elicitation effort and representation pertaining to the
modeling of conceptual knowledge in the health care field. The project has the goal of building a conceptual
model of data in the Military Health System Data Repository, a large DoD/VA aggregation of databases that
can be used in the implementation of software. The goal is to create a just-in-time conceptual model of the
data to facilitate software development and foster software developer understanding of the domain.
1 INTRODUCTION
Healthcare information systems play a critical role in
providing information about patients, insurers, and
service providers. Modern systems have become
huge, complex, and difficult-to-understand with no
real consistency in the use or meanings of the
vocabularies that describe services or service
providers. Better integration among health care
systems is critical to the goals of cost containment
and improved health care delivery. With vast
amount of health care data becoming accessible it is
important that the transmitted data are well
understood by end users and machines in order to
support decision making and automatic processing
respectively.
Concept mapping (Novak & Gowin, 1984) is a
proven technology that helps people express and
visualize their knowledge (Briggs et al., 2004;
Coffey & Eskridge, 2008; Coffey et al., 2006).
Concept maps are graphical representations of
propositional knowledge based upon concepts and
phrases that make explicit the relationships between
concept pairs. This projects aims to develop a
knowledge model using concept mapping (Novak &
Gowin, 1984; Novak & Musonda, 1991; Turns et al.,
2000) as a means to capture domain knowledge
pertaining to the concept of "health care provider" as
a proof-of-concept to show that:
(1) a concept map-based knowledge model can
be an effective tool to develop a domain
vocabulary that describes data streams and
services generated by health care systems,
(2) an improved understanding on data models
can assist software developers to build and
maintain integrated systems that can deliver
desired services to end users.
The remainder of this paper contains a brief
review of literature pertaining to concept mapping
and knowledge modeling, and a description of the
current work. Included in the description of the
current work is motivation for the work, the
activities and results that have been achieved to date,
and a scheme to assess the utility of the work to the
sponsors of the project. The paper closes with a
description of anticipated future work.
2 CONCEPT MAPPING &
KNOWLEDGE MODELING
Concept maps represent knowledge as a two-
dimensional graph capturing concepts and their
relationships on a specific topic. In this graph,
concepts are visualized as labeled nodes and
relationships as labeled links. A concept can be a
word or phrase that describes a perceived regularity
in events or objects, or records of events or objects
(Novak & Gowin, 1984). In contrast, a labelled link
includes any word or phrase so that the two concepts
connected by a link give rise to a meaningful
statement about the concepts. Such a statement is
also known as a proposition or a semantic unit that
captures human knowledge on a specific topic. Fully
developed maps include a cohesive set of concepts
470
Reichherzer T., Coffey J., Gonen B. and Gillett I..
Knowledge Modeling in the Health Care Domain to Support Software Development & Maintenance.
DOI: 10.5220/0005239004700476
In Proceedings of the 3rd International Conference on Model-Driven Engineering and Software Development (MODELSWARD-2015), pages 470-476
ISBN: 978-989-758-083-3
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
and their relationships to address a specific question
that focuses the discussion in the map. Furthermore,
a single concept at the top of the map known as the
root concept serves as a starting point to explore the
captured knowledge of the map.
Concept maps elucidate an individual’s
conceptualization on a domain, which can be useful
to track the individual’s learning progress. Novak
and Gowin (1984) originally conceived of Concept
Maps as a means to externalize conceptual
knowledge of science students in service of
uncovering what they knew, as well as
misapprehensions they held. Subsequently, several
groups have embraced Concept Mapping for
knowledge elicitation (KE). McNeese et al. (1993;
1995) used Concept Maps to externalize expert
knowledge over a variety of knowledge domains.
These included the expertise of pilots regarding their
decision-making strategies and to create conceptual
designs of cockpits, and that of managers, for
internal information management systems and
procedures. Novak (1998) described knowledge
elicitation with Concept Maps as a means of idea
generation in groups of people. Beyond the use of
concept maps as a learning and teaching tool,
successful applications of concept mapping include
institutional memory preservation (Coffey &
Eskridge, 2008; Coffey et al., 2006), sharing of
domain expertise for public outreach (Briggs et al.
2004), visualization of artifacts in Service-Oriented
Architecture (SOA) composite applications (Coffey
et al., 2012), and representation of software security
assurance cases (Snider et al., 2014).
Electronic concept mapping offers the prospect
of connecting concept maps via navigational links
and annotating concepts with multimedia resources
such as video clips, images, and links to Web sites to
provide additional information and examples on the
concepts in a map. A set of closely-related and
interconnected concept maps and their multi-media
resources are organized into a knowledge model
(Ford et al., 1991; Ford et al., 1993; Ford et al.,
1995), a hypermedia representation of domain
knowledge. CmapTools (Cañas et al., 2004) is a
publicly available software tool for creating
knowledge models and publishing them on the Web
for knowledge sharing purposes. It offers special
drawing and labelling operations for constructing
maps as well as a suite of tools for annotating nodes
in a map with multimedia resources or links to other
concept maps in the same knowledge model.
CmapTools is extensively used for research
involving concept maps and, generally, as a learning
and teaching tool by teachers and students alike
(Walker and King, 2002).
Snider et al (2014) describe a study in which
concept maps were elicited as a means of efficiently
developing security assurance cases. They describe a
two-phase process based upon concept map-based
knowledge elicitation and modeling. In the first
phase, the security expert performs the preliminary
work with a KE to identify the critical security
concerns, creating a "skeleton" concept map
containing nodes for each critical concern. In the
second phase, the security expert and KE interview
the software developer who demonstrates how he or
she has addressed those security concerns. In the
process, the skeleton concept map from the first
phase is elaborated to document program files, other
documentation, and other application responsibilities
(such as the server software being configured to
produce logs) that address the concerns. The concept
map of the software assurance case is augmented
with links to the actual files that address the security
concerns.
3 THE CURRENT PROJECT
Creating knowledge models typically involves a
domain expert and knowledge engineers
interviewing the expert and eliciting his/her
knowledge and representing it as concept maps (a
concise, unambiguous format). The goal of the
current project is to conduct a case study that
answers questions regarding how knowledge
elicitation of domain knowledge via concept
mapping might be useful to assist software
developers in creating data definitions for programs
in an efficient manner. For the project, a new
knowledge model was built to help understand the
semantics of a complex data model of an actual
Department of Defense (DOD) medical information
system. The model was then used to develop data
type definitions in the form of Enterprise Java
Beans. Software developers and domain expert
evaluated the process and artifacts that were
produced for this case study for usability and
completeness.
Motivation for the project comes from the fact
that bottlenecks in communication between the
domain expert and software developers threaten the
timely development of software. The
communication problem comes from the differences
in expertise between the software developer and
domain expert and the inherent complexity and
ambiguity of the medical domain in general.
Medical terms that label data in healthcare systems
KnowledgeModelingintheHealthCareDomaintoSupportSoftwareDevelopment&Maintenance
471
lack standardization, requiring interpretation for
proper use. The domain expert has deep knowledge
about the data and their origin and application in the
medical field, while the software engineer
understands how to organize data into logical units
for efficient access and processing. There is a
tremendous amount of information that has to be
exchanged to capture and understand data used in
healthcare information systems. The lack of
knowledge of the software engineer about the
domain makes the interpretation of data models very
difficult. A concept map-based knowledge model
can be an effective tool to develop a domain
vocabulary that describes data stored in complex
data models. We attempt to show that concept maps
can be an effective vehicle for communication
between the domain expert and software engineer,
giving the engineers relevant information to interpret
the data from the data models, alleviating the
knowledge gap between software engineer and
domain expert. In addition, the possible reuse of
built maps may help in breaking down bottlenecks in
communication.
4 RESEARCH METHODS
Researchers collaborated with a domain expert who
understood the large system of databases named the
Military Health System Data Repository (MDR).
The MDR is a "centralized data repository that
captures, archives, validates, integrates and
distributes Defense Health Agency (DHA) corporate
health care data worldwide." (Health.Mil, n.d.) The
MDR is comprised of more than 95 database tables,
each with 50 to 300 attributes, keeping healthcare
records to support 9.6 million military and related
beneficiaries world-wide.
The current case study involves information
retrieval pertaining to providers. The provider topic
is of particular interest to this study because of the
ambiguities in the representation of provider
information. Two simple but realistic scenarios were
created around the general problem of matching
patients to providers (e.g. finding a family
practitioner or finding a surgeon and facility). From
this initial effort, two use cases for software
development were created:
Use Case 1: Determine the simple provider
workload given the patient history.
Background: The user gives the system a
provider name and the system returns the workload
of that provider. The system answers the question:
What work load does a given provider have with his
current enrollment panels?
Use Case 2: Given the workloads of the members
of a group of providers, to which provider should a
patient be assigned?
Background: The user gives the system patient
name and pertinent other information (geographic
location, healthcare needs, personal preferences,
etc.) and the system recommends a physician in a
way that balances physician workload. Patients
might need to be assigned to a new provider because
they have not yet seen a provider in the area due to
relocation, the previous provider leaving the
network, etc.
One-sentence summaries of the use cases were
used as focus questions (a question intended to keep
the concept map creation effort focused on one
particular topic so as to avoid inclusion of
extraneous information). A concept map-based
knowledge elicitation effort aimed at building
conceptual models of the data needed in the use
cases was completed. The goal of this activity was to
identify the actual data from the MDR needed to
implement the use case. The output of this phase
was annotated concept maps describing the data
needed for the implementation of the use case. The
annotations were links into the MDR database
schema identifying relevant data for the use cases.
After the knowledge model was developed, the
software developers set about creating Java beans to
implement the use cases. An initial goal was to link
the Java beans to the knowledge models and the use
cases. This was to be achieved by including URLs
and a list of relevant concepts at the top of each bean
referring back to the Web-exported concept map.
The software developers presented questions about
the information in the knowledge model to the
domain expert and the KE for clarification. A
transcript was used to capture ongoing conversations
among these participants.
5 DATA ANALYSIS
A goal of the initial modeling attempt was to
determine what information was needed in the
concept maps in order to: (1) identify and link to
data elements in the MDR that were needed to
develop the software that implemented the use cases,
(2) provide conceptual information to help
developers understand the application domain, (3)
identify base elements that might be in a skeleton
concept map for such data modeling activity.
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
472
Figure 1: A concept map with domain knowledge and links into the MDA.
The initial knowledge modeling effort produced two
concept maps for Use Case 1 and one concept map
for Use Case 2 to answer the focus question
associated with each use case. The concept map
developed for Use Case 2 links back to one of the
concepts maps produced for Use Case 1 taking
advantage of prior elicited knowledge. Concept
maps for each use case were built over two 1-hour
modeling sessions that produced progressively more
refined maps. The resulting concept maps include
for Use Case 1 on average a total of 16 concepts and
17 connections and for Use Case 2 a total of 15
concepts and 16 connections. The initial concept
maps were further revised in a follow-up knowledge
modeling session after deficiencies in the elicited
knowledge were identified during the creation of
Java Beans by the software engineer. The revised
maps increased in size by an average of 10 concepts
and 8 connections for Use Case 1 and 4 concepts
and 7 connections for Use Case 2. Figure 1 contains
a portion of a concept map from the knowledge
model covering information about Appointments.
The specific map is part of a larger map created for
Use Case 1.
In the course of the knowledge modeling effort
several conventions were developed. The first was to
indicate when nodes in the concept map pertain to
databases. Such a node can be seen at the top of
Figure 1, "Appointments (Appointments Database)."
The icon at the bottom of that node links to a
database schema that describes all elements of that
table, the element's data type and format, and some
explanatory information.
When specific elements within the database are
identified as necessary to the data definitions in the
use case, their line numbers are indicated in the
concept map. For instance, "Adjusted RVUs" is an
important field pertaining to the amount of credit the
physician gets for handling an appointment. The
"Adjusted RVUs" element is found on line 30 within
the Appointments database. The remainder of the
concept map contains conceptual knowledge
regarding RVUs, specifically, that they are assigned
on the basis of management and procedure codes
that describe the type and duration of a visit.
As previously mentioned, questions asked by the
software developers and the domain expert's answers
were captured. Analysis reveals that two general
categories of questions were asked and answered:
low-level technical issues
higher level conceptual issues.
An example of a low-level question is "What
kind of data value does Num(5,3) describe?" An
example of a high-level question was "What is the
difference between Adjusted RVUs and RVUs in the
concept map?" A total of 11 questions and answers
KnowledgeModelingintheHealthCareDomaintoSupportSoftwareDevelopment&Maintenance
473
were exchanged for the two use cases.
The software developer produced a total of three
classes to address the two use cases: Provider,
Appointment, and Patient. The Provider
class had 11 attributes, the Appointment class
had 4 attributes, and the Patient class had 11. The
domain expert (who also has knowledge of software
development) was asked to judge the quality of the
data definitions according to the following criteria:
1. All necessary data is defined to address the use
case or
2. Some data is missing and:
A. the elements were missing from the
conceptual model.
B. the elements were in the conceptual model
but not made explicit enough for the
software developer to notice them.
C. the elements were included in the
conceptual model, made explicit, but
simply missed by the software developer.
The domain expert judged the data definitions to
be sufficient for Use Case 1. However, for Use Case
2, the software developer had created two attributes
that were not part of the conceptual model, due to a
somewhat idiosyncratic approach to the
implementation of containers for the Patient and
Provider classes. However, all needed (and
specified) attributes were present both in the
conceptual model and in the data definitions
produced by the software developer.
6 RELATED WORK
The work described in this paper proposes a new
process to support software development and
maintenance. It applies concept mapping to capture
semantic information of selected healthcare concepts
in a just-in-time knowledge modelling effort to
address questions related to new use cases. The
captured domain knowledge provides context for
interpreting information in large-scale, complex
healthcare data models. This work differs from a
number of other approaches discussed in the
literature that use conceptual modeling techniques
for system development. Widely used conceptual
modeling techniques are the Extended Entity
Relationship (EER), the Unified Modeling Language
(UML), and variations of it (Combi and Oliboni,
2006). They produce visual data models directly
from the narrative of a business use case as
abstractions of real-world phenomena of interest. In
all cases a specific language must be used to
translate the mental representation of a problem
domain into a physical diagram. However, as several
authors have pointed out in their research (Aguirre-
Urreta and Marakas, 2008; Castro, Baião &
Guizzardi, 2011), experience in building models,
domain expertise, and expressiveness of the
modeling language can have a significant effect on
the modelling outcome and the usefulness of the
resulting model for system development. In contrast,
concept mapping requires little training experience
for building models and interpreting them. It allows
a domain experts to participate in and shape the
modeling effort eliminating the need for the software
developer as the model builder to have knowledge of
the modeling domain or perform interpretation of a
narrative that could introduce ambiguity or errors
into the model.
In more closely related work, concept mapping
has been used to efficiently develop security
assurance cases in which pre-built maps were used
to facilitate and structure ongoing conversations
between software developer and security experts and
build comprehensive knowledge models on security
assurance cases (Snider et al. 2014). The resulting
maps provide evidence as a set of statements and
annotations that the software product meets the
necessary security requirements prior to its release.
However, in the current work concept mapping was
used to capture relevant semantic information about
healthcare data from a domain expert that is needed
by the software developer to create new data
definitions and algorithms for processing healthcare
data and develop new services.
7 CONCLUSIONS
Several conclusions can be drawn from the current
work. First and most important, it was possible to
develop the data definitions from the models. An
evaluation of the data definitions produced by the
software developer revealed that all attributes
needed to implement the Use Cases were present.
This result was achieved with a minimal amount of
information (11 questions and answers) exchanged
among the domain expert, KEs and software
developer outside of the knowledge contained in the
conceptual model.
It was not possible to calibrate accurately the
time efficiency of the knowledge modeling effort
because quite a bit of initial time was consumed in
figuring out what elements should be in a baseline
map and how to make the data items in the MDR
explicit in the concept maps. Now that these
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
474
elements are known, future work can include an
evaluation of efficiency.
Some conclusions can be drawn regarding the
effectiveness of the representation. As evidenced by
the questions exchanged among the domain expert,
KEs and software developer, the initial knowledge
models failed to convey some information both of a
high-level, conceptual nature and low-level technical
nature. No significant effort was made to address the
latter problem. However, KE and domain expert
conducted a follow-up knowledge modeling session
to address the issue of missing conceptual, high-
level information in the initial model, which enabled
the software engineer to complete the data
definitions for the two use cases. While it seems
unavoidable that some conceptual knowledge of
interest to the developers might be missed, given the
limited amount of time devoted to knowledge
elicitation in this study, it would likely have
occurred in any representation. The representation
used here is well known to be more concise and less
ambiguous than textual descriptions. Further, on the
basis of the exchanges between the domain expert
and the developer it can be determined that, while
gaps in the model's conceptual knowledge
representations existed, the developers at least knew
the right questions to ask.
ACKNOWLEDGEMENTS
Partial funding for this project was provided by
Blue-Cross Blue-Shield through the Security and
Software Engineering Research Center
(http://www.serc.net).
REFERENCES
Aguirre-Urreta, & M., Marakas, G. M., 2008. Comparing
conceptual modeling techniques: a critical review of
the EER vs OO empirical literature. The DATA BASE
for Advances in Information Systems, 39(2), 9-32.
Briggs, G., Shamma, D., Cañas, A. J., Scargle, J., and
Novak, J. D., 2004. Concept maps applied to Mars
exploration public outreach. In Cañas, A. J., Novak, J.
D., and González, F., editors, Concept Maps: Theory,
Methodology, Technology. Proceedings of the First
International Conference on Concept Mapping, pages
125-133, Pamplona, Spain.
Cañas, A. J., Hill, G., Carff, R., Suri, N., Lott, J., Eskridge,
T., Gómez, G., Arroyo, M., & Carvajal, R., 2004.
CmapTools: A knowledge modeling and sharing
environment. In Cañas, A. J., Novak, J. D., and
González, F., editors, Concept Maps: Theory,
Methodology, Technology. Proceedings of the First
International Conference on Concept Mapping,
Pamplona, Spain.
Castro, L., Baião, F., & Guizzardi, G., 2011. A semantic-
oriented method for conceptual data modeling in
ontoUML based on linguistic concepts. Conceptual
Modeling-ER, Springer-Verlag, pp. 486-494.
Combi, C. & Oliboni, B., 2006. Conceptual modeling of
XML data, In Proceedings of the 2006 ACM
Symposium on Applied Computing, Dijon, France.
Coffey, J. W., Moreman, D. & Dyer, J., 1999. Institutional
memory preservation at NASA Glenn Research
Center. Technical Report, NASA Glenn Research
Center, Cleveland, OH.
Coffey, J. W., & Hoffman, R. R., 2003. A knowledge
modeling approach to institutional memory
preservation. The Journal of Knowledge Management,
7(3), 38-49.
Coffey, J.W. & Eskridge, T., 2008. Case Studies of
Knowledge Modeling for Knowledge Preservation and
Sharing in the U.S. Nuclear Power Industry. Journal
of Information and Knowledge Management. 7(3). pp
173-185.
Coffey, J.W., Hoffman, R.R., & Cañas, A. J., 2006.
Concept Map-based Knowledge Modeling:
Perspectives from Information and Knowledge
Visualization. Information Visualization, 5. pp. 192-
201.
Coffey, J., Reichherzer, T., Wilde, & N., Owsnicki-Klewe,
B., 2012. Automated Concept-Map Generation from
Service-Oriented Architecture Artifacts. Proceedings
of the 5th International Conference on Concept
Mapping. Valetta, Malta, September.
Ford, K. M., Cañas, A. J., Jones, J., Stahl, H., Novak, J., &
Adams-Webber, J., 1991. ICONKAT: An integrated
constructivist knowledge acquisition tool. Knowledge
Acquisition, 3, 215-236.
Ford, K. M., Cañas, A. J., & Coffey, J. W., 1993.
Participatory Explanation. Proceedings of FLAIRS'93,
the Sixth Florida Artificial Intelligence Research
Symposium, Ft. Lauderdale, FL, pp111-115.
Ford, K. M., & Bradshaw, J.M., 1995, “Beyond the
Repertory Grid: New approaches to Constructivist
knowledge acquisition tool development”,
International Journal of Intelligent Systems, 8, 287-
333.
Health.Mil, The Official website of the Military Health
System and Defense Health Agency. Online, Available
from: http://www.health.mil/Military-Health-
Topics/Technology/Clinical-Support/Military-Health-
System-Data-Repository [14 August 2014].
McNeese, M., Zaff, B., Brown, C., Citera, M. & Selvaraj,
J., 1993. Understanding the context of
multidisciplinary design: Establishing ecological
validity in the study of design problem solving,
Proceedings of the 37th Annual Meeting of the Human
Factors Society, Santa Monica, CA.
McNeese, M., Zaff, B. S., Citera, M., Brown, C. E., &
Whitaker, R., 1995. AKADAM: Eliciting user
knowledge to support participatory ergonomics.
KnowledgeModelingintheHealthCareDomaintoSupportSoftwareDevelopment&Maintenance
475
International Journal of Industrial Ergonomics, 15,
345-363.
Novak, J. & Gowin, D., 1984. Learning How to Learn.
Cambridge. University Press, New York, NY.
Novak, J. D. & Musonda, D., 1991. A twelve-year
longitudinal study of science concept learning.
American Educational Research Journal, 28(1):117–
153.
Turns, J., Atman, C. J., & Adams, R., 2000. Concept maps
for engineering education: A cognitive motivated tool
supporting varied assessment functions. IEEE
Transactions on Education, 43(2):164–173.
Walker, J. M. T. & King, P. H., 2002. Concept mapping as
a form of student assessment and instruction. In
Proceedings of the 2002 ASEEE Annual Conference
and Expo.
Snider, D., Coffey, J., Reichherzer, T., Wilde, N., Terry,
C., Vandeville, J., Heinen, A., & Pramanik, S., 2014.
Using Concept Maps to Introduce Software Security
Assurance Cases. CrossTalk: The Journal of Defense
Software Engineering, vol. 27, no. 5.
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
476