AN AGENT-BASED SYSTEM FOR HEALTHCARE

PROCESS MANAGEMENT

Bian Wu, Minhong Wang

Division of Information and Technology Studies, The University of Hong Kong, Hong Kong, China

Hongmin Yun

Ophthalmological Department, Shengzhen Hospital, Beijing University, China

Keywords: Healthcare, Process Management, Software Agent.

Abstract: An effective approach for healthcare process management is the key to delivery of high-quality services in

healthcare. An agent-based and process-oriented system is presented in this study to facilitate dynamic and

interactive processes in healthcare environment. The system is developed in three layers: the agent layer for

healthcare process management, the database layer for maintenance of medical records and knowledge, and

the interface layer for human-computer interaction. The treatment of primary open angle glaucoma is used

as an example to demonstrate the effectiveness of approach.

1 INTRODUCTION

In healthcare organizations, there are a variety of

processes, such as hospital administration by

managers, registration for treatment by patients, test

report generation by technicians, and diagnosis

decision making by doctors. There is a trend to view

healthcare in a multidisciplinary perspective, as

numerous interactions and cooperation take place

across different functional units in terms of

information sharing, consultation, and combined

treatments. These interactive and collaborative

activities should be well organized and managed,

without which the efficiency of healthcare service

cannot be easily guaranteed. To meet these

requirements, a process-oriented and cooperation-

supported healthcare system is needed, especially

with computer and information technology support.

To facilitate healthcare process management,

workflow technology (van der Aalst and van Hee,

2002) has been applied in a number of studies (Lenz

and Reichert, 2007). In the recent years,

organizational environments have been changing

from stable-and-closed to dynamic-and-open.

Business processes are becoming increasingly

complex and dynamic as they seek to cope with a

wide range of internal and external interactions and

changes. Traditional workflow technologies for

process management are often inadequate for

complex and dynamic situations due to the lack of

flexibility and adaptability (Wang and Wang, 2006).

This change has also happened in the healthcare

sector. Frequent changes of clinical pathways and

increased interactions between different units have

become a big challenge in healthcare process

management. This change has also happened in the

healthcare sector. Due to newly discovered

symptoms of a specific disease or need of further

diagnosis, pre-planned clinical schedules always

request adjustment; ad hoc changes of medical

process are needed during execution. Besides, to

fulfill a simple task such as gonioscopy for

glaucoma diagnosis, cooperation among different

functional units is required. In sum, frequent

changes of clinical pathways and increased

interactions between different units have become a

big challenge in healthcare process management.

To develop a computer-based system that enables

effective process management in healthcare, an

agent-based cognitive approach is presented in this

study. The term “agent” refers to a piece of software

that can perceive its environment through its sensors

and can act upon that environment through the

effectors (Wooldridge and Jennings, 1997).

healthcare management unit has similar

characteristics with software agent, which has a set

Wu B., Wang M. and Yun H. (2009).

AN AGENT-BASED SYSTEM FOR HEALTHCARE PROCESS MANAGEMENT.

In Proceedings of the 11th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

45-50

DOI: 10.5220/0001859000450050

 SciTePress

of goals (e.g., treatment effects and costs),

perceptions of the environment (e.g., symptoms and

records), and actions to take (e.g., recommendations

for test or treatment, and inquiry).The highly

dynamic and unpredictable nature of organizational

processes makes agent-based approaches appealing.

Using agent-based technology, we may decompose a

complex process into a number of loosely coupled

tasks and delegate the tasks to a number of software

agents, each of which works autonomously and

collaboratively in performing tasks and managing

the whole process (Wang and Wang, 2006).

In this study, an agent-based and process-oriented

healthcare system is developed. The system consists

of three layers, the agent layer for healthcare process

management, the database layer for maintenance of

medical records and knowledge, and the interface

layer for human-computer interaction. The emphasis

is placed on dynamic clinical and administrative

process management, and knowledge building as the

foundation for process management. This agent-

based and process-oriented healthcare system

distinguishes from other healthcare systems in terms

of continuous awareness of the healthcare

environment, real-time dynamic decision making of

healthcare processes, and rule-based knowledge

engineering. The treatment of primary open angle

glaucoma (POAG) is used as an example to

demonstrate the effectiveness of this approach.

2 SYSTEM ARCHITECTURE

To meet the requirements arising from dynamic

processes and cooperative functions in healthcare, an

agent-based healthcare process management system

is presented with a three-layer architecture.

2.1 Interface Layer

This layer supports various interactions between

human users (administrators, doctors, nurses, and

medical technicians) and the computer system.

Different type of user may access and interact with

the system through specific interfaces.

2.2 Agent Layer

In the system, a group of software agents are

proposed to perform healthcare tasks. Various

interactions take place between the autonomous

agents or between the agents and human users

including doctors, nurses, technicians, and

administrators. The design of software agents is

process-oriented, i.e., agents are able to execute

tasks according to process rules. What’s more,

process rules can be edited in a graphical form,

which makes it easy to specify and update the rules.

The details of each agent are elaborated as follows.

Clinical Decision Agent works with doctors. It

consists of three components: a knowledge base that

captures glaucoma knowledge into a set of rules for

clinical diagnosis decision making; a reasoning

engine that generates diagnosis decision or treatment

recommendations based on patient records and test

results; and an interpreter that interprets the

decisions or recommendations generated by the

computer.

Patient Management Agent works with clinical

administrators and nurses, managing patient records

and coordinating treatment processes. It also

interacts with other agents for communication of

patient records, test reports, and treatment records

throughout the healthcare process.

Medical Examination Agent works with doctors,

medical technicians, or nurses, as well as

communicates with Patient Management Agent and

Clinical Decision Agent throughout the medical

examination process. The medical examination

process may go through several steps including

patient records access, test scheduling, notification

and reminder, and finally test report generation and

delivery.

Ward Treatment Agent works with doctors and

nurses, as well as communicates with Patient

Management Agent or Clinical Decision Agent

during the inpatient treatment process. The activities

include treatment scheduling, notification and

reminder, information inquiry, and treatment report

generation and delivery. The agent also maintains

the information of treatment activities, progress, and

patient states during the treatment process.

Operation Agent works with doctors and nurses,

as well as communicates with Patient Management

Agent or Clinical Decision Agent to manage the

operation process. The process may go through

several steps including patient records retrieval,

operation scheduling, notification and reminder, and

operation report generation and delivery.

2.3 Database Layer

Medical information is stored in a database to

support the healthcare functions. The information

includes patient records, treatment records,

examination records, diagnostic knowledge, and

treatment knowledge. Considering information

security and maintenance issue, different agents are

ICEIS 2009 - International Conference on Enterprise Information Systems

Figure 1: Hierarchy of clinical algorithm ontology.

associated with different data, with different rights

to access or update the data. For example, Patient

Management Agent has the right to access patient

records; while other agents can acquire patient

records via communication with Patient

Management Agent. Differing from data records,

knowledge for clinical and administrative process

management is more complicated. Relevant

techniques and tools are used to capture and model

the knowledge into a set of rules or guidelines.

3 KNOWLEDGE-BASED

PROCESS MANAGEMENT

To facilitate dynamic and interactive clinical

processes, we need to capture a large amount of

information and knowledge as the foundation for

process management. In addition to clinical

information such as patient records and treatment

data, it is crucial to identify or set up the rules or

guidelines as the knowledge for diagnosis and

treatment decision making and clinical

administration. To capture and represent the

knowledge in the system, ontology-based

technology is adopted in this study. Using ontology,

the narrative paper based guidelines or rules are

encoded into computerized specifications. To build

the ontology, we use Protégé (O’Connor et al., 2007)

as the ontology representation tool. Protégé is a free

and open-source platform that provides a suite of

tools to construct domain models and knowledge-

based applications with ontology. In developing the

proposed system, we use Protégé for knowledge

building by way of class modelling, goal and criteria

building, clinical algorithm construction.

In this study, we choose glaucoma treatment as

an example to demonstrate the proposed approach.

Glaucoma is a leading cause of blindness in most

countries; the irreparable loss of vision by glaucoma

requires lifelong healthcare. Meanwhile, the disease

of glaucoma is well investigated; the process of the

treatment is documented as various kinds of

guidelines that can be used as reference for

knowledge building in this case. There are many

types of glaucoma, from which we choose the

primary open angle glaucoma (POAG) for

illustration.

3.1 Knowledge Modelling

We model the rules or guidelines for clinical

decision-making into clinical algorithms (see Figure

1). Based on the clinical algorithms,

recommendations such as messages, drugs, and

referrals can be generated for diagnosis, treatment,

and other clinical process.

Each algorithm includes three nodes: scenario,

decision, and action. Scenarios are mutually

exclusive entry points for clinical algorithms, which

exhibit unambiguous definition of patient states. A

scenario contains several major properties called

slots: a) the new encounter slot is used to

enable/disable this scenario as a starting point; b) the

precondition slot is to qualify the scenario as a

AN AGENT-BASED SYSTEM FOR HEALTHCARE PROCESS MANAGEMENT

Figure 2: POAG treatment algorithm after topical therapy.

starting point; and c) the followed by slot is a next

step in the algorithm. A decision node contains: a) a

choice step (e.g., severe IOP1 decrease <30%)

which may lead to more than one choice; or b) a

case step which has an evaluated expression to

enable the execution of the next step node. An

action is a node with a rule in/out criteria slot (e.g.,

if risk level = severe and IOP decrease >= 30%) and

a defined action slot (e.g., Filtration surgery) in

forms of message, drug recommendations, or

referrals.

3.2 Knowledge-based Clinical Process

The diagnosis and treatment process of glaucoma

can be divided into several steps:

1) Comprehensive eye examination should be

carried out to differentiate POAG or suspect.

2) Further examinations are taken for establishing a

baseline of intraocular pressure (IOP) and designing

the initial treatment.

3) After establishing a good baseline, it is obliged to

set up a reasonable goal of intraocular pressure

(IOP).

4) Based on the goal established in step 3, an initial

topical therapy is carried out to lower the pressure.

5) If necessary, follow-up treatment is required after

the initial topical therapy. Due to the space

limitation, we only provide details of step 5 as

follows.

In the initial treatment of POAG (step 4), the first

step topical therapy may not reach the initial

intraocular pressure goal. Therefore, treatment

algorithm after topical therapy (Hodapp et al., 1993)

should be followed. Figure 2 shows the modeling of

this specific algorithm in Protégé. Pink circle

indicates a scenario; yellow polygon indicates a

decision; and green square indicates an action

choice.

Figure 3: Class hierarchy in Protégé.

ICEIS 2009 - International Conference on Enterprise Information Systems

They are linked by one direction arrow to show a

follow-up relation with each other and form an

algorithm chart. After implementation, the POAG

treatment algorithm after topical therapy is

represented as an instance of Management

Algorithm Entity shown in Figure 3. The pink

circles, yellow polygons, and green squares in

Figure 2 are mapped to three subclasses (Scenario,

Choice step, and Action Choice) of Management

Algorithm Entity, respectively in Figure 3.

4 SYSTEM IMPLEMENTATION

In our system, five agents are developed, Patient

Management Agent, Medical Examination Agent,

Ward Treatment Agent, Operation Agent, and

Clinical Decision Agent. For implementation of a

multi-agent system, we adopted JADE (Bellifemine

et al., 2007), a widely used open source middleware

developed by TILAB for the development of

distributed multi-agent applications based on the

peer-to-peer communication architecture.

In addition to multi-agent architecture, the

implementation concerns the individual agent. After

specifying clinical guidelines and process rules

using Protégé, we use a plug-in called

beangenerator, implemented by C.J. van Aart, to

create the ontology definition class and the

predicates, agent actions, and concepts classes

(Caire and Cabanillas, 2006) for implementation of

individual agent. Agents can retrieve information

and knowledge to perform tasks under control of

rules or guidelines. Two types of rules are specified,

one for clinical decision and another for

administrative process control.

5 CONCLUSIONS

This paper has presented a three-layer architecture of

an agent-based healthcare process management

system. The agent-based cognitive approach is

applied to facilitate dynamic and interactive

processes in healthcare management. The treatment

of primary open angle glaucoma is used as an

example for demonstration. After the

implementation of the prototype, we will evaluate

the effectiveness of the approach in terms of support

for clinical and administrative process management

in healthcare environment. Relevant clinical and

administrative staff will be invited to use this system

and give feedback and evaluation on the system.

ACKNOWLEDGEMENTS

This research is supported by UGC GRF Grant

(716907) from the Hong Kong SAR Government

and Seeding Funding for Basic Research

(200711159052) from The University of Hong

Kong.

REFERENCES

Bellifemine, F., Caire, G., and Greenwood, D., 2007.

Developing Multi-Agent Systems with JADE,

Chichester, England; Hoboken, NJ: John Wiley.

Caire, G. and Cabanillas, D., 2004. Jade Tutorial

Application-defined content languages and ontologies,

Manual, TILAB.

Hodapp, E.M.D., Parrish, R.G.M.D., and Anderson, D.G.,

1993. Clinical decision in glaucoma, Mosby-Year

Book, Inc.

Lenz, R. and Reichert, M., IT support for healthcare

processes– premises, challenges, perspectives. Data &

Knowledge Engineering, 61 (1), 39-58, 2007.

O’Connor, M., Shankar, R., Tu, S., Nyulas, C., Das, A.,

and Musen, M., 2007. Efficiently Querying Relational

Databases Using OWL and SWRL The First

International Conference on Web Reasoning and Rule

Systems. Innsbruck, Austria, Springer.

van der Aalst, W.M.P. and van Hee K.M., 2002. Workflow

Management: Models, Methods, and Systems, MIT

press, Cambridge, MA.

Wang, M. and Wang, H., 2006. From Process Logic to

Business Logic - A Cognitive Approach to Business

Process Management. Information & Management, 43

(2), 179-193.

Wooldridge, M. and Jennings, N.R., 1999. Software

Engineering with Agents: Pitfalls and Pratfalls, IEEE

Internet Computing, 3(3), 20-27.

AN AGENT-BASED SYSTEM FOR HEALTHCARE PROCESS MANAGEMENT