AN AUTONOMIC COMPUTING FRAMEWORK FOR

SELF-MANAGED EMERGENCY DEPARTMENTS

Serene Almomen and Daniel Menascé

Volgenau School of Information Technology and Engineering, George Mason University

4400 University Drive, Fairfax, VA 22030, U.S.A.

Keywords: Autonomic Computing, Healthcare, Emergency Department, Quality of Service, Self-Managed System.

Abstract: The delivery of cost-effective and quality Emergency Department (ED) services remains an important and

ongoing challenge for the healthcare industry. ED overcrowding has become a common problem in

hospitals around the world, threatening the safety of patients who rely on timely emergency treatment.

Despite numerous advances in medical procedures and technologies, EDs continue to experience

overcrowding problems. The combination of increased demand and diminished resources makes optimizing

emergency departments a difficult problem for healthcare decision makers. We examine this problem by

applying an autonomic computing framework for self-managed emergency departments to maintain optimal

Quality of Service (QoS) during its operation. Our work has potential implications in guiding a hospital’s

effort to optimize their emergency department system.

1 INTRODUCTION

In hospitals all over the country, healthcare

emergency departments (ED) are severely

overcrowded resulting in delays in care, difficulty in

providing quality care, patient discomfort and

dissatisfaction, and higher service cost. In addition,

overcrowding also leads to staff burnouts and

inefficient utilization of resources. Many ED nurses

leave for other departments or units as a result of

getting overwhelmed with the ED workload (ACEP,

2010).

The challenges of the ED, including

overcrowding and boarding, have been a subject of a

great deal of discussion. Several meetings, reports,

and research studies have been conducted to

understand the causes, implications and possible

solutions to ED overcrowding and boarding issues.

The ED is one of the most critical units in any

healthcare organization. Consequently, improving

performance of this unit is vital to the success of the

healthcare organization.

Due to the dramatic increase in the cost of

healthcare over the past few decades, researchers

and healthcare professionals examined new ways to

improve efficiency and at the same time reduce

healthcare costs. Simulation tools have assisted

healthcare decision-makers in this endeavour

(Hashimoto & Bell, 2007). Another attempt to

improve the ED system relies in capturing ED

workflow patterns and analyzing these patterns to

create an automated and enhanced ED system design

(Moss & Xiao, 2004). Another approach discussed

the use of workflow technologies and web services

to automate emergency healthcare processes

(Poulymenopoulou, Malamateniou, &

Vassilacopoulos, 2008). That work discussed the

need to provide an appropriate technological

infrastructure for automating and managing the

emergency healthcare processes in both intra- and

inter-organizational services. The implementation of

this approach involves capturing process logic

requirements for healthcare workflow systems with

a view to design a system that is easily adjustable to

process changes and to evolving organizational

structures. Some tools have also been developed for

hospital capacity planning simulation to conduct

both process flow analysis and capacity forecasting

(Mengwasser & Berger, 2009).

The dynamic nature of the ED adds to the

complexity of the problem. Sudden changes to the

workload due to emergencies such as fire, natural

disasters, and terrorist attacks are difficult, if not

impossible, to predict.

Furthermore, the ED environment is complex in

nature. ED systems are composed of a collection of

Almomen S. and Menascé D..

AN AUTONOMIC COMPUTING FRAMEWORK FOR SELF-MANAGED EMERGENCY DEPARTMENTS .

DOI: 10.5220/0003138200520060

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2011), pages 52-60

ISBN: 978-989-8425-34-8

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

resources including both humans (e.g., doctors,

nurses, and technicians) and equipments (e.g., X-ray

machines and CT-Scan). ED systems also involve

human processes and decision making where

humans in the loop determine how the application

evolves based on their awareness of the situation and

infrastructure. Consequently, there is a need for self-

managing EDs. We examine this problem by

applying an autonomic computing framework for

self-managed EDs to maintain optimal operational

Quality of Service (QoS). Our work has potential

implications in guiding a hospital’s effort to

optimize their emergency department systems.

The rest of the paper is organized as follows.

Section 2 discusses background information on the

proposed approach. Section 3 discusses the

environment of EDs. The next section describes the

autonomic computing framework for EDs. Section 5

discusses different implementation techniques for

the autonomic framework. Finally, Section 6

presents some concluding remarks.

2 BACKGROUND

Computing systems have reached a level of

complexity where traditional IT support that

involves human effort to maintain the systems and

keep them operational is becoming increasingly

challenging. A similar problem was experienced in

the 1920s in telephony before automatic branch

exchanges were introduced to eliminate human

intervention (Mainsah, 2002).

Autonomic computing seeks to enhance the

performance or QoS and at the same time minimize

human intervention. The autonomic computing

paradigm has been inspired by the autonomic

function of the human central nervous system

(Kephart & Chess, 2003). It is the body’s master

controller that monitors changes inside and outside

the body, integrates sensory inputs, and effects

appropriate response (Ashby, 1960). Autonomic

controls in the human body use motor neurons to

send indirect messages to organs at a sub-conscious

level. These messages regulate temperature,

breathing, and heart rate without conscious thought

(Ashby, 1960). The implications for computing are

immediately evident; a network of organized, smart

computing components that give us what we need,

when we need it, without a conscious mental or even

physical effort (IBM, 2010).

Autonomic computing attempts to intervene in

computing systems in a similar fashion to its

biological counterpart. There has been significant

research to create autonomic systems. An example

of such effort is IBM’s MAPE-K (Monitor, Analyze,

Plan, Execute, Knowledge) reference model

(Huescher & McCann, 2008). A similar model is

proposed by Russel and Norvig (2003) in which an

intelligent agent monitors its environment through

sensors and uses the collected data to determine

actions to be performed in the environment (Russel

& Norvig, 2003).

In either model, an autonomic system has a

managed element (such as software or hardware

resources), the organ in the human body, that is

given an autonomic behaviour and an autonomic

manager, the nervous system in the human body,

that monitors the managed element and specifies

actions to be executed by the managed element

(Huescher & McCann, 2008) (Russel & Norvig,

2003). There have been several different

implementations of the MAPE-K model including

autonomic toolkit, ABLE, Kinesthetics eXtreme

(KX), and self-management tightly coupled with

application. The latter implementation is closely

related to the proposed approach discussed in this

paper. Such implementation involves either using an

autonomic middleware framework that offers self-

management properties to applications built on top

of this middleware or through encapsulating tasks in

components and defining self-management and

adaptation in terms of these components (Huescher

& McCann, 2008).

IBM also portrayed four fundamental properties

of self-management: self-configuration, self-

optimization, self-healing, and self-protection.

Briefly, these properties mean that an autonomic

computing system configures itself according to

high-level goals, optimizes its use of resources,

detects and diagnose problems, and protects itself

against malicious attacks and end users who

inadvertently make changes to the system

components such as its software (Huescher &

McCann, 2008). Consequently, any implementation

of autonomic computing systems should realize

these properties.

In addition, studies in the autonomic systems

field describe approaches to plan the changes or

actions to be effected on the managed element of an

autonomic system. Some propose policy-based

adapting planning, architectural models, or process-

coordination approach (Sloman, 1994) (Wise, Cass,

Lerner, Call, Osterweil, & Jr., 2000) (Huescher &

McCann, 2008).

With recent advances in embedded computing,

networking, and related information technologies, it

is now feasible to deploy a variety of sensing

AN AUTONOMIC COMPUTING FRAMEWORK FOR SELF-MANAGED EMERGENCY DEPARTMENTS

devices, communication networks and IT services in

the real world. These physical spaces include a

variety of sensors such as optical sensors, RFIDs, as

well as specialized sensors such as people-counters

and load-cells that enable monitoring the state of the

physical world and its activities. These sensors are

connected to communication networks such as

Ethernet, cellular, Bluetooth, and WiFi. (Kim, et al.,

2008). These sensors provide a mechanism to

monitor the different resources of a system. Such

technology makes it even easier to implement

autonomic computing systems in real world

environments.

Another concept that can facilitate the

implementation of autonomic systems in real world

environments is utility functions. Utility functions

express the usefulness of a system to one or more

stakeholders as a function of the attributes of a

system. The concept of utility is one of the methods

used to represent Knowledge in autonomic systems.

A utility function is written as follows:

U = f (x

, x

, ..., x

) (1)

where x

, …, x

are attributes and the function f

combines these attributes in way that expresses the

usefulness of a system as a function of these

attributes. In general, utility functions are

normalized in the [0,1] range with zero representing

the lowest utility and one representing the highest

utility. It is generally easier to specify a utility

function as a function of several utility functions,

one for each attribute. An example would be where

(2)

the global utility function is a weighted sum of all

the individual utility functions.

Autonomic computing systems use utility

functions as the goal to be optimized. The attributes

in this case are several Quality of Service (QoS)

metric of interest such as response time, throughput,

and availability of the computing resources

(Menasce, Bennani, & Ruan, 2005). As failures and

performance degradations occur, the autonomic

computing system automatically changes its

configuration parameters in a way that maximizes

the utility function for the system. Consequently, the

utility function of an autonomic system can be

written as follows:

U = f (QoS

, QoS

, ...QoS

) (3)

where both the utility function and QoS metrics are

defined by domain experts.

3 EMERGENCY DEPARTMENT

ENVIRONMENT

We interviewed a Director of Emergency Services

and a staff nurse at a Pediatric Emergency

Department to gain a better understanding of the ED

environment. Our findings are summarized in what

follows.

The ultimate goal of an ED is patient

satisfaction, which is normally measured by the

length of stay at the ED. ED length of stay is the

patient time in the ED as follows (Welch, Augustine,

Camargo, & Reese, 2006):

 For admitted patient: arrival time to

conversion time

 For discharged patients: arrival time to

discharge time

 For transferred patients: arrival time to

transfer conversion time

Many QoS metrics are collected and analyzed at

EDs to determine areas of improvement that are

necessary to meet this goal. These QoS metrics can

be time measures or proportion measures (Welch,

Augustine, Camargo, & Reese, 2006). The time

measures include arrival time to first seen by a

doctor, doctor to discharge time, doctor to decision

to admit time, arrival time to rooming, disposition to

discharge, and many others. The proportion

measures include number of patients who left before

they were supposed to, complaints, hospital

diversion, and ED patient flow to name a few. These

measures are commonly referred to as Core

Measures and are often specified by a healthcare

governing body serving local or nationwide

hospitals. These Core Measures data are saved in an

advanced analytical tool where comparative reports

taking into account national averages can then be

generated allowing hospitals to proactively assess

performance and identify opportunities for quality

improvement including potentially preventable

readmissions and complications. These Core

Measures also depend on several Census and

utilization metrics including pediatric patients per

day, high-acuity patients per day, number of self-

paid patients, medication doses administered per 100

patients seen, and service hours per day of

physicians.

There are many critical issues in EDs that

contribute to an unsatisfactory level of these QoS

metrics, many of which are demography-dependent.

Inadequate patient beds always lead to long arrival

to discharge time for example. The lack of an

ultrasound machine may also be critical in an ED

U = w

i=1

∑

)

HEALTHINF 2011 - International Conference on Health Informatics

that expects many pregnancy-related emergencies.

This is an example of a demography-related issue.

Other issues include lack of outpatient psychiatric

service and lack of outpatient programs for referral.

However, we found that lack or resources,

specifically nurse shortage, is a major contributor to

not meeting the overall goal of quick patient

turnaround.

There are compelling reasons to collect and

control ED QoS metrics. Regulatory burdens, ED

operations management, and ED body of knowledge

expansion are some (Welch, Augustine, Camargo, &

Reese, 2006). The Joint Commission on

Accreditation of Healthcare Organizations

(JCAHO), for example,

is pursuing clinical quality

improvement (QI) data in the form of Core

Measures. Any facility that does not have in place

the infrastructure to track these data risks its

accreditation.

In addition, to determine whether ED

process innovations are effective, quality measures

will be required. To date, much QI work goes

unpublished, and therefore ED QI workers are

failing to build a body of research that is pertinent to

operational efficiency.

In an effort to control QoS metrics within an

acceptable range, hospitals in collaboration with

nurse and the physician groups document their ED

standard processes, treatment protocols, and

regulations and orient new staff on them. The

processes take into account national averages of ED

patients’ length of stay and try to stay within or

below that range. In addition, these processes are

evaluated often to accommodate technology

changes, equipment increase, and changes in

practices. Pilots and time audits are sometimes used

to determine the compliance to those processes and

protocols.

To maximize efficiency at the ED, the

Emergency Severity Index (ESI) to classify patients

coming into the ED is used. ESI is a five-level

emergency department (ED) triage algorithm that

provides clinically relevant stratification of patients

into five groups from 1 (most urgent) to 5 (least

urgent) on the basis of acuity and resource needs.

The Agency for Healthcare Research and Quality

(AHRQ) funded initial work on the ESI (AHRQ,

2010).

A triage nurse is responsible for ESI level

assignment to ED patients. The ESI level determines

the waiting time of patients. ESI level 1 patients

have no waiting time for example. In addition, EDs

are divided into care areas or zones based on the

severity of the case treated. Consequently, the ESI

level also determines the zone the patient will

occupy. Level 4-5 patients are often assigned to the

‘fast track’ zone since they usually do not require

many resources before they can be discharged. The

triage nurse is also responsible for zone assignments.

A charge nurse uses the ESI level and zone

assignment for each patient in the ED to determine

the most efficient workflow of the ED. The charge

nurse’s role, consequently, is to run the ED as

efficiently as possible to help minimize patients’

length of stay. In the EDs we visited, the charge

nurse uses a computer application that collects the

QoS metrics as well as patient status to assist in

making decisions on the most efficient workflow at

the ED at any given time. These decisions may

include changing a nurse’s assignment to balance the

workload of ED nurses, task a nurse to dispense

medication if a doctor’s order is ready, start the

hospital admission process for a patient after

doctor’s diagnosis is complete, and request an on-

call nurse to come to the ED if the workload is high.

The staff nurses are assigned patients and take care

of patients once they are in a room at the ED.

It is also important to note that there is no cost

constraint that limits the operation of an ED. In other

words, no patient will ever be turned away because

of insufficient resources. EDs normally have

working agreements with other hospitals in their

area where they can quickly transfer patients to due

to heavy workload or resource shortage. EDs also

use what they call a ‘float pool’ of nurses who can

be used to staff EDs if needed. Nevertheless,

maximizing the efficiency of an ED does reduce the

cost of service.

4 AN AUTONOMIC COMPUTING

FRAMEWORK FOR ED

Our framework involves developing an autonomic

computing system or a self-managed ED system that

can regulate and maintain itself without human

intervention. This is ideal in an ED environment

since the goal is to create a system that will be able

to adapt to a constantly changing environment (such

as patient flow, workload, and resource availability)

in a way that preserves given operational goals (such

as performance goals or QoS goals).

To achieve that, the proposed approach attempts

to implement the MAPE-K (Kephart & Chess, 2003)

model in an ED environment. Within the ED

context, autonomic managers define a control loop

(MAPE-K loop), as shown in Figure 1. Changes are

made through action operations. Sensors look at the

AN AUTONOMIC COMPUTING FRAMEWORK FOR SELF-MANAGED EMERGENCY DEPARTMENTS

state of the managed ED resources, and action

operations can change the current state. The entire

ED environment is a set of managed resources.

Autonomic managers, just like a charge nurse,

continuously monitor the system and handle events

that need action to be taken. They monitor the ED

environment using inputs from the sensors installed

in the environment and analyze what is found. Based

on the defined utility function of the ED, the

autonomic manager then plans and executes any

specific actions needed to maximize the utility

function. The steps of monitoring, analyzing,

planning, and executing may be executed

concurrently. For example, if the X-ray machine in

an ED is experiencing high utilization, the system

could decide to provision an additional machine.

The system can then return to monitoring, and if

utilization drops, the X-ray machine can be

deprovisioned and made available to other

departments in the hospital.

Figure 1: Autonomic Computing Model for ED.

The ED system depicted in Figure 1 can have a

variety of architectures. It can consist of integrated

applications or components within the ED such as

the pharmacy dispensing application, the scheduling

application, and the resource allocation application.

We also consider a cyber-physical ED system as

including smart devices such as patient wrest

sensors, nurse PDAs, bed sensors, and other devices.

Our approach to self-managed ED systems does

not address the various ED system architectures.

Rather, it assumes a cutting edge system such as a

cyber-physical environment that collects data or

metrics through different sensors and devices to send

to the autonomic manager. Contrary to common

systems in the autonomic computing environments,

the ED system in the diagram consists of human

resources in addition to the hardware and software

resources. This means that doctors and nurses are

considered resources of the system as well as CT-

Scan, X-ray machines, bed sensors, and pharmacy

dispensing application for example.

The ED autonomic manager attempts to optimize

pertinent QoS metrics. One of the metrics, called

wait time ratio and defined as W/T, combines the

average length of stay (T) with the average time (W)

spent by the patient in the waiting area. Other

relevant metrics are patient throughput (X

), and

resource utilization (U

) for resource i.

The values of these QoS metrics depend on

several of the census and utilization parameters

discussed in section 3 which can be categorized as

workload intensity parameters (e.g., the arrival rate

of patients of a given group) and the service

demands parameters of each group at each resource

(e.g., the average time spent by a patient using the

CT-Scan).

The goal of the ED autonomic system is to find

settings of the managed resources that optimize a

given utility function provided by the domain expert,

which depends on the values of several QoS metrics.

This is important in order to realize the self-

optimization property of an autonomic system. The

autonomic manager uses the provided utility

function to plan appropriate changes or actions to be

effected on the managed resources of the ED. As

mentioned before, utility function may be obtained

by combining utility functions for the different QoS

metrics, such as:

 Utility function for the throughput of the ED:

)

 Utility function for the average length of stay,

T, in the ED: U

(T)

 Utility function for the average time spent by

a patient waiting in the ED: U

(W)

 Utility function for the wait time ratio W/T:

W/T

(W/T)

As an example, Figure 2 shows two utility

functions: one for U

W/T

(W/T) in Figure 2(a) and the

other for U

) in Figure 2(b). A global utility

function U

is a function of the individual utility

functions U

), U

(T), U

(W), and U

W/T

(W/T):

= f (U

), U

(T), U

(W),

W/T

(W/T)).

(4)

HEALTHINF 2011 - International Conference on Health Informatics

Figure 2: Example of Utility Functions.

5 IMPLEMENTATION

TECHNIQUES

There are two main approaches to implementing

autonomic computing systems; model-driven based

on system performance models and data-driven

based on reinforcement learning. In what follows,

we describe the two approaches to autonomic

computing implementation and illustrate how one

model-driven technique, the queuing-theoretic

model, can be applied in the context of the ED.

5.1 Model-driven Approach

The model-driven approach relies on being able to

build models that can be used to predict the values

of a system’s QoS metrics as a function of its

configuration parameters and resource sharing

policies. Parameter and policy optimization

techniques need to be defined in this approach. The

parameter and policy optimization techniques map

system states to action operations, hence, are used to

plan the changes to the autonomic system to

maximize its utility function (Gracanin, Bohner, &

Hinchey, 2004).

Model-driven approaches focus on algorithms

that make use of explicit system performance

models such as queuing-theoretic or control-

theoretic models.

Using a model-driven approach to autonomic

computing makes it possible to generate run-time

models that reflect the current state of the system

without the unnecessary dependency on the system

platform (Rohr, Boskovic, Giesecke, & Hasselbring,

2006). This capability makes this approach possible

to adapt in implementing dynamic and complex

autonomic systems such as the ED system.

However, the design and implementation of accurate

performance models of complex computing systems

can be highly knowledge-intensive and labour-

intensive and may require original research

(Tesauro, Jong, Das, & Bennani, 2006).

5.2 Data-driven Approach

The data-driven approach focuses on knowledge-

free trial-and-error methodology in which a learner

tries various actions in numerous system states, and

learns from the consequences of each action

(Tesauro, Jong, Das, & Bennani, 2006). This

approach is also referred to as Reinforcement

Learning (RL). RL has successful applications in

Markov Decision Process (MDP) in which RL can

potentially learn decision-theoretic optimal policy in

dynamic environments where the effects of actions

are stationary and history-independent. RL has

successful implementation in real-world problems

such as helicopter control and financial markets

trading (Tesauro, Jong, Das, & Bennani, 2006).

Contrary to the model-driven approach, RL does

not require an explicit model of the computing

system. In addition, RL has the capability to

properly react to dynamical phenomena in an

environment due to its roots in sequential decision

theory. Other methods tend to treat dynamical

effects only approximately or ignore them all

together, or deal with the decision making problem

as a series of unrelated instantaneous optimization

(Tesauro, Jong, Das, & Bennani, 2006). Thus, RL is

a possible implementation approach in autonomic

computing.

The use of RL in real-world applications such as

an ED system, however, can suffer from poor

scalability in large state spaces since a lookup table

is used to store a separate value for every possible

state-action pair. The size of such a table increases

exponentially with the number of state variables of

the system making it challenging to use in real

applications (Tesauro, Jong, Das, & Bennani, 2006).

In addition, poor performance in live systems

implementing this approach can be observed due to

the long learning periods that may be necessary.

AN AUTONOMIC COMPUTING FRAMEWORK FOR SELF-MANAGED EMERGENCY DEPARTMENTS

Figure 3: Example of ED QN Model.

5.3 Model-driven Implementation

in the ED

The work by Menascé and Bennani (2003) describes

creating autonomic computing models for computer

systems based on the notion of queuing network

(QN) models (Menascé & Bennani, 2003). Much

like computer systems, an

ED has many shared

resources (e.g., beds, X-ray machines, doctors, and

nurses). The performance of such a system can,

hence, be conveniently represented by a Queuing

Network (QN) model that represents the flow of

patients and their contention for these resources.

When a patient arrives at the ED for treatment, the

patient alternates using the different resources in the

ED such as the nurse, bed, and X-ray machine, quite

likely more than once. At any point in time, a patient

can be treated by a nurse and another using the X-

ray machine, while other patients are waiting to be

treated by the nurse or use the X-ray machine. Thus,

the nurse and X-ray machine can each be

characterized as a queue with a waiting line. An

example of a QN model of an ED system is

illustrated in Figure 3. In the diagram, the different

resources of the system are represented by queues.

This is a common representation in computer

systems that our approach may adapt for self-

managed ED systems including humans as

resources. For example, nurses, doctors, and X-ray

technicians are resources used by patients who flow

through the system as indicated in Figure 3. The

flow of a patient from one resource to another gives

this model the network nature.

Some parameter and policy optimization

approaches commonly used for QN models are hill

climbing and beam-search algorithm among others.

These approaches are referred to as combinatorial

search techniques. The use of exhaustive searches of

all possible configurations of the ED system is not

feasible due to the complexity of such systems.

Consequently, using a combinatorial search

technique will find a close-to-optimal configuration

so that the utility function of the new configuration

is as close as possible to the desired QoS level

(Babaoglu, et al., 2005) (Menasce, Bennani, &

Ruan, 2005). In this case, the ED autonomic

manager will use combinatorial search techniques

such as hill-climbing to find the close-to-optimal

configuration. A state space represents possible

configurations of the system, as shown in Figure 4.

Each point in the space represents a configuration of

controlled parameters and the numerical value

associated with each point represents the value of

the utility function.

Figure 4: Example of State Space Search.

The figure shows that the current configuration is

point A with value .10, which is obtained by

computing the utility function using the

measurements obtained from the sensors in the ED

system. Through hill-climbing search, all

‘neighbour’ configurations are examined and a new

configuration with the highest value of the utility

function is selected. The search is repeated at each

new point visited until either the value of the QoS

does not improve, or a threshold on the number of

points traversed has been exceeded. Through hill-

climbing, a new close-to-optimal configuration,

point B with value .35, is found. The value of any

point in the search space can be computed through

the use of QN models that can predict the value of

the ED QoS metrics for configurations different than

………….

Doctor

X-ray

Technician

Discharge

Staff Nurse

Waiting Area

Emergency Department

Triage Nurse

Charge Nurse

………….

HEALTHINF 2011 - International Conference on Health Informatics

the current one. The QoS values are then used to

compute the value of the utility function for that

point in the search space.

The configuration parameters for each point in

the space can be represented as v=(v

, …,v

). As an

example, in the ED, possible configuration

parameters to be changed by the autonomic

controller are: the maximum number of doctors

), the maximum number of nurses (M

), the

queuing discipline at the X-ray machine (d

X-ray

), and

the queuing disciplines at the CT-Scan (d

CT-Scan

Thus, the configuration point can be defined as

v=(M

, d

X-ray

CT-Scan

). Based on these

parameters, the combinatorial search technique of

choice will start at the current configuration C

examine all the neighbour configurations to C

and

move to the one with the highest QoS value. A

neighbour configuration is defined as one in which

the parameters values of M

, and M

changes by ±1

and the parameter values of d

X-ray

, and d

CT-Scan

changes from First Come First Served (FCFS) or

Priority Queuing, for example. The search is

repeated at each new point until either the value of

utility function does not improve, or a threshold on

the number of points traversed has been exceeded.

The autonomic manager will then send an action

operation comprised of the new optimal

configuration to the ED system to change the ED’s

current configuration in order to achieve improved

operations and decreased service cost.

6 CONCLUDING REMARKS

Inspired by biology, autonomic computing has

evolved as a discipline to create software systems

and applications that self-manage in an attempt to

overcome the complexities and inability to maintain

current and emerging systems effectively.

The implementation of autonomic computing has

been increasingly emerging in fields including

power management, data centers, clusters, and

GRID computing systems, and ubiquitous

computing. These applications are already

demonstrating their feasibility and value (Huescher

& McCann, 2008). However, there is no evidence

that autonomic computing has been implemented in

healthcare Emergency Department (ED) systems

where not only hardware and software comprise the

system resources, but also human beings. Our

framework extends the autonomic computing

concepts to create a self-managed ED system to

reduce the dependency on human intervention to

maintain such a complex system, thus, improve the

ED operations, and decrease the ED service cost.

We are currently in the process of investigating a

live implementation of the framework proposed in

this paper in an ED environment. Specifically, we

plan to compare the results on an ED with and

without an autonomic computing system in order to

investigate whether performance and cost

improvements can be obtained.

REFERENCES

ACEP, 2010. Meeting the Challenge of Emergency

Department Overcrowding/Boarding. Washington:

American College of Emergency Physicians.

AHRQ. (2010). Emergency Severity Index, Version 4.

Retrieved October 20, 2010, from Agency for

Healthcare Research and Quality: http://

www.ahrq.gov/research/esi/esi1.htm

Ashby, W. R., 1960. Design for a Brain. Chapman & Hall

Ltd.

Babaoglu, O., Jelasity, M., Montresor, A., Fetzer, C.,

Leonardi, S., Moorsel, A. v., et al., 2005. Self-Star

Properties in Complex Information Systems. Springer

Verlag: Lecture Notes in Computer Science.

Gracanin, D., Bohner, S. A., & Hinchey, M., 2004.

Towards a Model-Driven Architecture for Autonomic

Systems. Proc. 11th IEEE Intl. Conf. Engineering of

Computer-Based Systems, (pp. 500-505).

Hashimoto, F., & Bell, S., 2007. Improving Outpatient

Clinic Staffing and Scheduling with Computer

Simulation. Journal of General Internal Medicine ,

182-184.

Huescher, M. C., & McCann, J. A., 2008. A survey of

Autonomic Computing—Degrees, Models,and

Applications. New York: ACM.

IBM, 2010. Autonomic Computing - The Solution.

Retrieved June 30, 2010, from Autonomic Computing:

http://www.research.ibm.com/autonomic/index.html

Kephart, J., & Chess, D. (2003). The Vision of Autonomic

Computing. IEEE Internet Computing , 41-50.

Kim, M., Massaguer, D., Dutt, N., Mehrotra, S., Ren, S.,

Stehr, M.-O., et al. (2008). A Semantic Framework for

Reconfiguration of Instrumented Cyber Physical

Spaces. St. Louis: IEEE.

Mainsah, E., 2002. Autonomic computing: The next era of

computing. Electronic Communications Engineering

Journal , 14, 1, 2–3.

Menascé, D. A., & Bennani, M. N., 2003. On the Use of

Performance Models to Design Self-Managing

Computer Systems. Computer Measurement Group

Conference. Dallas: Computer Measurement Group.

Menascé, D. A., Bennani, M. N., & Ruan, H., 2005. On

the Use of Online Analytic Performance Models in

Self-Managing and Self-Organizing Computer

Systems. Springer Verlag: Lecture Notes in Computer

Science.

AN AUTONOMIC COMPUTING FRAMEWORK FOR SELF-MANAGED EMERGENCY DEPARTMENTS

Menascé, D., Almeida, V., & Dowdy, L., 2004.

Performance by Design: Computer Capacity Planning

by Example. Upper Saddle River: Pearson Education,

Inc.

Mengwasser, M. J., & Berger, M. A., 2009, February.

Hospital Capacity Planning Model. Noblis Sponsered

Research Projects , pp. 20-21.

Moss, J., & Xiao, Y., 2004. Improving Operating Room

Coordination: Communication Pattern Assessment.

The Journal of Nursing Administration , 93-100.

Poulymenopoulou, M., Malamateniou, F., &

Vassilacopoulos, G., 2008. Emergency healthcare

process automation using workflow technology and

web services. Athens: Informatics for Health and

Social Care.

Rohr, M., Boskovic, M., Giesecke, S., & Hasselbring, W.,

2006. Model-driven Development of Self-managing

Software Systems. Proc. 9th Intl. Conf. Model-Driven

Engineering Languages and Systems. Springer.

Russel, S., & Norvig, P., 2003. Artificial Intelligence: A

Modern Approach 2nd Ed. Prentice Hall.

Sloman, M., 1994. Policy driven management for

distributed systems. J. Netw. Syst. Manag.

Tesauro, G., Jong, N. K., Das, R., & Bennani, M. N.,

2006. A Hybrid Reinforcement Learning Approach to

Autonomic Resource Allocation. 3rd IEEE

International Conference on Autonomic Computing

(ICAC), (pp. 65-73). Dublin, Ireland.

Welch, S., Augustine, J., Camargo, C. A., & Reese, C.

(2006). Emergency Department Performance

Measures and Benchmarking Summit. Academic

Emergency Medecine , 1074–1080.

Wise, A., Cass, A. G., Lerner, B. S., Call, E. K.,

Osterweil, L. J., & Jr., S. M., 2000. Using Little-JIL to

coordinate agents in software engineering. Automated

Software Engineering. ASE.

HEALTHINF 2011 - International Conference on Health Informatics