use of multi-agents in computerizing clinical...

Use of Multi-Agents in Computerizing ClinicalGuidelines

INDERJEET SINGH DOGRA

University of Windsor

[email protected]

Clinical guidelines define standard procedures for the diagnosis and treatment of various dis-

eases, which should be followed by medical practitioners around the world. But adherence tothese guidelines becomes a difficult task considering their frequent versioning, changes required

in them to cater to specific health conditions of patients, their effective management etc. Addedto this is the increasing cost of medical services, in the healthcare domain, due to increasing ag-

ing population, which calls for optimizing the use of medical resources. This survey summarizes

research work on computerizing clinical guidelines using multi-agent technology, to enable theircompliance in medical processes and to provide a decision support system that assists profession-

als in taking decisions through various situations. This survey contains detailed annotations of

research publications describing different approaches used to computerize clinical guidelines, theireffective management, modification, verification and usage in remote monitoring.

Categories and Subject Descriptors: A.1 [Introductory and Survey]: ; D.l.3 [Program-

ming Techniques]: Concurrent Programming—Distributed programming; I.2.11 [ComputingMethodologies]: Distributed Artificial Intelligence—Multiagent systems ; I.6.0 [Simulation

and modeling]: General

General Terms: clinical guidelines, computerizing guidelines

Additional Key Words and Phrases: clinical decision support system, agent-based computerization

of clinical guidelines, remote monitoring using conputerized clinical guidelines, management andcustomization of computerized clinical guidelines

Contents

1 Introduction 2

2 Survey of Research 32.1 Agent-Based Approaches for Computerizing Clinical Guidelines . . . 3

2.1.1 Distributed guideline-based health care system . . . . . . . . 32.1.2 Computer-based management of clinical guidelines . . . . . . 42.1.3 Using BDI software agents for representing medical guidelines 52.1.4 Agent-based process-critiquing decision support system . . . 62.1.5 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Verifying Computerized Clinical Guidelines . . . . . . . . . . . . . . 82.2.1 Spin model checking for the verification of clinical guidelines 82.2.2 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3 Interpreting Clinical Guidelines using Ontology . . . . . . . . . . . . 102.3.1 An ontology-driven agent-based clinical guideline execution

engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3.2 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

ACM Journal Name, Vol. V, No. N, Month 20YY, Pages 1–0??.

2 · INDERJEET SINGH DOGRA

2.4 Remote Monitoring using Computerized Clinical Guidelines . . . . . 112.4.1 Saphire: A multi-agent system for remote healthcare moni-

toring through computerized clinical guidelines . . . . . . . . 112.4.2 Agent-based execution of personalised home care treatments 122.4.3 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.5 Managing and Customizing Clinical Guidelines . . . . . . . . . . . . 142.5.1 An agent-based cognitive approach for healthcare process man-

agement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.5.2 A semiotic multi-agent modeling approach for clinical path-

way management . . . . . . . . . . . . . . . . . . . . . . . . . 152.5.3 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Concluding Comments 16

4 Annotations 184.1 Bosansky et al. 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . 184.2 Isern et al. 2004b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194.3 Isern et al. 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204.4 Isern et al. 2007b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214.5 Isern et al. 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224.6 Laleci et al. 2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244.7 Li et al. 2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254.8 McGrory et al. 2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . 264.9 Terenziani et al. 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . 274.10 Wu et al. 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1. INTRODUCTION

Clinical guidelines are standard procedures that suggest actions, directions andprinciples for treatment of various diseases. These procedures are established byexpert committees, which should be followed by the medical practitioners in health-care domain, but considering the specific health condition of each patient, thechanging nature of guidelines and their effective management owing to their dy-namic participation in medical processes, the abidance of clinical guidelines in anyhealthcare setting becomes a difficult task. This survey presents the work doneby various researchers in computerizing clinical guidelines using the multi-agentapproach, with the goal of developing a clinical decision support system that iscapable of assisting and guiding the medical professionals through various situa-tions. The survey also highlights the use of multi-agent concept in handling variousaspects of computerized clinical guidelines, which forms part of the overall decisionsupport system. These aspects include verification, modification and managementof computerized guidelines.

The research papers for this survey were found using Google Scholar, ACM,LNCS and IEEE.

This survey is based on 10 research papers, of which 7 are conference papers,2 are journal papers and one of them is a series paper. Most of these papers aretheoretical in terms of the ideas they present, which they intend to demonstrateusing specific case studies. The work done by Isern et al. [2003] appears to be

ACM Journal Name, Vol. V, No. N, Month 20YY.

Use of Multi-Agents in Computerizing Clinical Guidelines · 3

the first paper directed at computerizing clinical guidelines, using the multi-agentapproach and it also forms the basis for the work done in [Isern et al. 2004b; Isernet al. 2006; Isern et al. 2007b; Isern et al. 2011], where each deals with a differentissue, involving the use of computerized clinical guidelines. Isern et al.’s [2006]work present a comparative study of the relevant systems in the field, which addsvalue to this survey as it gives an overview of the different systems available forrepresenting and executing clinical guidelines electronically.

The first section in this survey discusses papers presenting different agent-basedapproaches for computerizing clinical guidelines. The next section presents paperson verification of computerized clinical guidelines, which involves testing a guidelineagainst its different execution paths. The next section reviews papers that high-lights the role of ontology in interpreting computerized clinical guidelines. The nextset of research papers underlines the application of computerized guidelines in re-mote monitoring of patients. The last group of papers illustrate various approachesfor effectively managing and customizing computerized guidelines.

Despite the availability of various approaches for computerizing medical processesin healthcare domain, none of them have been completely realized in a real clinicalsetting. This is due to the requirement of smooth integration of guideline executionwith the daily workflow of activities within any medical centre, which in itself isa big concern. Furthermore it has also been noticed that agent technology is oneof the best approaches for the development of computerized guideline executionsystems, due to the autonomous behaviour offered by the agents, the capability ofdynamic co-ordination among the agents and the handling of distributed environ-ment in an efficient way. The research papers reviewed in this survey also indicatesthat the implementation of guideline execution system can standardize the medicaltreatments of diseases and reduce the medical costs by optimizing the utilizationof medical resources.

2. SURVEY OF RESEARCH

2.1 Agent-Based Approaches for Computerizing Clinical Guidelines

The research papers covered in this section present different approaches for com-puterizing the representation and execution of clinical guidelines, using the multi-agent paradigm. The prime concern behind computerizing the clinical guidelines isto reduce the cost of medical services, by optimizing the use of available medicalresources, and ensure the compliance of clinical guidelines by medical practitioners.The section also includes a survey by Isern at el. [2006], which compares some ofthe relevant systems, available for computerization of clinical guidelines.

2.1.1 Distributed guideline-based health care system. Considering the increasingcost of medical services in the healthcare domain, the efficient usage/managementof medical resources has become a critical issue. Isern et al. [2004b] address theproblem of optimizing the use of medical resources by proposing the integration ofmulti-agent system and the medical guidelines thereby improving the coordinationbetween the medical services and the humans. By using autonomous agents torepresent different entities and medical services in healthcare domain, the authorsalso address the problem of interaction and cooperation among different processes



within a healthcare organization.

Isern et al. [2004b] refer to previous work by Isern et al. [2003].

Isern et al. [2004b] state that their previous work does not consider the use ofmedical guidelines and also lacks the use of dedicated agents for various medicalservices.

Isern et al. [2004b] propose a new version of the multi-agent system architecturethat is an improvement over their previous work HeCaSe, in order to provide abetter coordination and cooperation between humans and medical services using themedical guidelines. The proposed architecture incorporates the medical guidelinesin the system by using a guideline agent which is responsible for accessing, storingand modifying the medical guidelines. The architecture consists of a medical centreagent which supervises the user’s access to the main system and provides all therequired information to the user through the user agent. A user agent maintainsprofile of the user and allows the same to be accessed by the doctor so as toanalyse the user’s medical records before the visit. The user agent is linked tothe main system through the broker agent that helps user retrieve the requiredinformation about the system. Each of the medical centre agents is responsiblefor monitoring the department agents and various service agents. The departmentagents are further composed of the doctor agents, more specific service agents andthe guideline agents. Finally the medical records of all the patients are stored inthe database by a medical record agent which lies at the bottom of the architectureand provides a secure access to the stored data.

The system also provides scheduling of services in which an appointment betweenthe patient and the doctor and between the patient and the required service isarranged. The scheduling mechanism takes into consideration different constraintssuch as the patient’s availability, timetable of the services etc, before setting upan appointment. The authors describe the scheduling process using a simplifiedexample of appendicitis.

Isern et al. [2004b] do not describe any experiments conducted.

Isern et al. [2004b] state that the proposed architecture incorporates the featureof autonomy and pro-activity in the system, by modelling various entities in thehealthcare domain as the agents, with the supposition that the modelled entities areautonomous having their own beliefs, desires and intentions. The authors also statethat their system allows the addition of new agents and features and uses standardlanguages for the agent communication and the representation of ontologies.

The research papers [Isern et al. 2006], [Isern et al. 2007a], [Isern et al. 2007b]and [Isern et al. 2011] refer to Isern et al.’s [2004b] work.

2.1.2 Computer-based management of clinical guidelines. Computerized execu-tion of the clinical guidelines has become an important area of research and many(semi-) automatic systems are being developed nowadays for this purpose. Isern etal. [2006] examines several systems, available for the representation and executionof clinical guidelines, on the basis of a set of properties and present a comparativestudy for the same.



Isern et al. [2006] do not refer to any previous survey.

Though this survey is not exactly on topic, but the in-depth analysis of eightrelevant systems presented by the authors, including their own system, HeCaSe2,makes an important contribution towards the main topic and the comparative studyalso adds value to our survey.

Isern et al. [2006] examined eight relevant systems in their survey, the names forwhich are SAGE, GLARE, GLEE, NewGuide, Arezzo, DeGel, SpEM and HeCaSe2.These systems have been analyzed and compared against a set of properties whichincludes GL Repository, GL Editor, GL Representation Language, Agents, Coordi-nation, Run Time Engine, Access to EHR, Standard used and Security. For a betterunderstanding, the authors present the result of analysis in a tabular format.

Isern et al. [2006] state that the adoption of a single standard for the guide-line representation language can solve the interoperability problem, as most of thesystems compared in the survey use the representation languages that share thesame basic components. Another factor affecting the interoperability of differentsystems is the representation of an EMR (Electronic Medical Record), for which abest representation is yet to be finalized. The authors also state that none of thecompared systems except the Arezzo, which is used for some specific tasks only, isbeing used in any medical centre. This is due to the reason that smooth integra-tion of the execution of computerized guidelines with the workflow of activities, ina healthcare institute, is a big issue which still needs to be resolved. Consideringthe agent-based design of the HeCaSe2 and Arezzo systems, the authors also claimthat the agent-based approach enables the designing of flexible and interoperablesystems, which can be extended to include the services and resources particular toa specific healthcare institute.

Isern et al. [2006] claim that the multi-agent technology is one of the best ap-proaches for the development of computerized clinical guideline execution systems.The authors base their claim on the autonomous behaviour offered by the agents,the effective handling of the distributed environment, the dynamic co-ordinationprovided among the agents, the various communication algorithms available foragent interaction, etc. Finally to assert their claim, the authors describe a map-ping between the properties of intelligent agents and the requirements of a guidelineenactment system.

There are no specific references to Isern et al.’s [2006] work by other researchersin this survey.

2.1.3 Using BDI software agents for representing medical guidelines. The com-puterization of clinical guidelines using the traditional CIG (Computer-Interpretable-Guidelines) model works by decomposing a clinical guideline into separate workflowactivities, which are managed by a centralized inference engine. This fragmentationresults in the loss of the actual context represented by a guideline and increasesthe size and complexity of the system, with the addition of more guidelines. Mc-Grory et al. [2008] address the problem of computerizing the clinical guidelinesusing the BDI (Belief-Desire-Intention) agent paradigm, thereby accounting for theissues highlighted in the CIG models.



McGrory et al. [2008] do not refer to any previous work.

McGrory et al. [2008] present a BDI agent approach for computerizing the clinicalguidelines, in which each guideline is represented by a single autonomous agent.This approach retains the original context represented by a guideline and reducesthe complexity of the system, as each agent has its own inference engine in contrastto having a centralized engine. For the BDI agent shell, the authors tend to usethe modified version of the Jadex agent shell known as the Autonomous SocialisingKnowledge agent (ASK-agent). In the ASK-agent model, a guideline is decomposedinto workflow activities with each guideline having a dedicated inference engine formaking decisions. These guidelines communicate with other resources in the systemby exchanging messages.

An agent in the ASK-agent model captures the information attributes, the de-liberative and action attributes and the motivational attributes of a guideline in itsbelief, intention and desire components respectively. This information is extractedfrom the guidelines using the Medical Logic Module (MLM) of the Arden Syntax.A particular ASK-agent thus embeds all the information contained in a guidelineand serves as a knowledge source for the other interacting agents in the system.The authors describe the step-by-step procedure for converting a guideline to itscorresponding ASK-agent.

McGrory et al. [2008] do not describe any experiments conducted.

McGrory et al. [2008] claim that their approach allows a guideline’s content to beembedded in an autonomous standalone agent module, having a dedicated inferenceengine. This agent can then be used as a knowledge source for the other agentssearching for the information related to the guideline. The authors also state thattheir concept provides a solution to the management of clinical guidelines in theway similar to that provided by the narrative guidelines.

There are no specific references to McGrory et al.’s [2008] work by other re-searchers in this survey.

2.1.4 Agent-based process-critiquing decision support system. Bosansky et al.[2009] address the problem of using medical guidelines/processes and the multi-agent system approach in a decision support system, in order to simulate the med-ical processes and ensure their abidance by the practitioners. In such a system,the occurrence of inconsistencies while following the medical guidelines/processescan be avoided with multiple co-operating agents monitoring the execution of thesemedical guidelines/processes.

Bosansky et al. [2009] refer to the previous work by Isern et al. [2007c].

Bosansky et al. [2009] criticize Isern et al.’s [2007c] system for interpreting asingle guideline at a given time, as this can result in missing out crucial informationthat could have been provided by the simultaneous execution of related guideline(s).They also criticize the approach for using a single agent for interpretation of aguideline. This can lead to a problem if the agent working on a particular guidelineis required to take on some high priority task, in which case the agent will have toeither suspend the execution of current guideline or avoid taking the high prioritytask. Both the cases highlight the need for multiple agents interpreting a guideline.



Bosansky et al. [2009] present a new multi-agent system architecture in whichthe agents work with the medical guidelines/processes, thereby allowing their sim-ulation and monitoring. They propose to use a set of co-operating agents for inter-preting a medical guideline/process, thereby overcoming the problem of suspendingexecution of a guideline due to the need for the agent to take up a higher prioritytask. The proposed architecture also allows the concurrent execution of a set ofguidelines and does not depend on the physician to select the guideline, hence pro-viding a better analysis and control for a wider domain of related diseases. Theydescribe a critiquing decision support system as an applications of the proposed ar-chitecture which monitors the correct execution of the medical guidelines/processesand alerts the physician in case of inconsistencies.

In their architecture, each step in the guideline/process is handled by a processagent which simulates and predicts the output of the corresponding step, matchesit with the physician’s output and raises an alert in case of inconsistency. Thedifferent roles in a healthcare domain are maintained by the role agent and theconcrete physicians, nurses, patients and other employees working in a healthcaredomain are represented by the execution agent. All the patient data is maintainedby the patient health record agent which can be accessed through the environmentagent. Finally a process director agent answers the queries of other agents regardingthe information about medical guidelines/processes.

Bosansky et al. [2009] do not describe any experiments conducted.

Bosansky et al. [2009] claim that their framework is flexible and agile as it allowsthe simultaneous interpretation of a set of medical guidelines/processes therebycovering a wider domain of related diseases and avoid overlooking of crucial infor-mation by the physician. They state that their system is based on general principlesand hence does not depend on a specific process language and can also be appliedto other organizational processes. Finally they claim that the system also offersseveral possible ways to alert the healthcare personnel, in case of inconsistencies.

The research paper [Lhotska et al. 2011] refers to Bosansky et al.’s [2009] work.



2.1.5 Summary.

Year Author Title of Paper Major Contribution

2004b Isern and

Moreno

Distributed

guideline-based healthcare system.

Incorporates medical guidelines in

HeCaSe architecture, which canbe accessed, stored and modified

by a dedicated guideline agent.

2006 Isern and

Moreno

Computer-Based

Management of ClinicalGuidelines: A Survey.

Compares eight relevant systems,

available for computerizedrepresentation and execution of

clinical guidelines.

2008 McGrory,

Grimson, Clarke

and Gaffney

Software agents

representing medical

guidelines.

Introduces BDI agent, as an

improvement over the traditional

CIG model, for thecomputerization of medical

guidelines.

2009 Bosansky and

Lhotska

Agent-based

process-critiquingdecision support system.

Uses multiple co-operating agents

for interpreting a clinical guidelineand allows the parallel execution

of a set of guidelines.

Table I. Major Contribution in Computerizing Clinical Guidelines

2.2 Verifying Computerized Clinical Guidelines

The single research paper in this section deals with the verification of clinical guide-lines against their different execution paths, using the agent-based paradigm. Thepaper was written in 2006.

2.2.1 Spin model checking for the verification of clinical guidelines. The verifi-cation of various properties of computerized clinical guidelines and their applicationin the healthcare domain becomes an important aspect in order to ensure their cor-rect implementation and execution. The verification of clinical guidelines is alsorequired as they undergo frequent changes during their real time application indiagnosis and treatment processes. Moreover, for a given scenario, the verificationprocess can automatically test the guideline for all its possible execution paths.Terenziani et al. [2006] address the problem of verification of clinical guidelinesalong with their specification, following the agent-based paradigm.

Terenziani et al. [2006] do not refer to any previous work.

Terenziani et al. [2006] present an agent-based approach for modelling the clini-cal guidelines, which uses the GLARE (Guideline Acquisition, Representation andExecution) system as the basis for acquiring the guidelines through a user interface.These guidelines, including the agents interacting with them, are then modelled asprocesses in the specification language Promela. This specification language formsthe part of the model checker SPIN, that the authors propose to use as an exten-sion to the GLARE architecture for the specification and verification of the clinicalguidelines. The clinical guidelines acquired using the GLARE system are mapped



to processes in Promela, using their XML specifications in GLARE. The patientrecord database and all the agents in the system are also modelled as Promelaprocesses. The model checker SPIN then transforms each Promela process into afinite automaton and computes an asynchronous interleaving product of automata.This resulting automaton can be built on-the-fly during the verification processand depicts the global state space of the system i.e the model consisting of all thepossible executions of the guideline. The SPIN uses this global state space forthe verification of the guideline property. The SPIN converts the negation of thetemporal formula, corresponding to the input guideline, into a Buchi automatonand computes its synchronous product with the system global state space. Theempty language of the resulting Buchi automaton signifies that the guideline prop-erty holds true on all the possible executions of the clinical guideline, otherwise theverifier provides the execution path on which the property is false. The proposedapproach provides automata-based semantics for the clinical guidelines and allowsthe verification of different types of properties, concerning a given clinical guideline.

Terenziani et al. [2006] illustrate the working of their proposed idea by using theverification process on the clinical guideline for ‘stroke’.

Terenziani et al. [2006] claim that the verification process found some discrep-ancies in the expression of the ‘stroke’ guideline, which corresponds to informationthat was deliberately removed from the guideline.

Terenziani et al. [2006] state that their approach provides an automatic wayto verify the properties of clinical guidelines, using the model checker SPIN. Theyalso claim that the knowledge represented by the guidelines is modelled using thePromela formalism, which allows its active participation in the overall verificationprocess. The authors also claim to provide a more declarative and logical semantics,as compared to the automata-based semantics, for the clinical guidelines.

There are no specific references to Terenziani et al.’s [2006] work by other re-searchers in this survey.

2.2.2 Summary.


2006 Terenziani,

Giordano,Bottrighi, Montaniand Donzella

Spin model checking for

the verication of clinicalguidelines.

Simulates the verification of

properties and executionpaths of computerized clinicalguidelines using SPIN model

checker.

Table II. Major Contribution in Verifying Computerized Clinical Guidelines



2.3 Interpreting Clinical Guidelines using Ontology

The single paper in this section highlights the need of ontology as a basis for estab-lishing a common understanding of the healthcare domain concepts/terminology,among the disparate agents. Sharing of such understanding enables the co-ordinatingagents to interpret and implement the guidelines correctly.

2.3.1 An ontology-driven agent-based clinical guideline execution engine. Thecomputerization of clinical guidelines using a multi-agent system framework re-quires the agents to work in coordination over the complex activities defined in anyclinical guideline. For such a system to work efficiently the agents must share thesame understanding of the medical concepts/terminology involved in the healthcaredomain. Isern et al. [2007b] address the problem of filling the semantic gap betweenthe codification of clinical guidelines and their interpretation in the healthcare do-main to improve interoperability among the existing systems.

Isern et al. [2007b] cite previous work by Isern et al. [2004b].

Isern et al. [2007b] criticize their previous work for lacking the use of a sharedmedical ontology among the agents in the system.

Isern et al. [2007b] propose the inclusion of an application ontology, constructedas an external source, in their previous system HeCaSe2 to enable different entitiesin the system, represented as agents, to coordinate with each other using a sharedunderstanding of the medical concepts in the healthcare domain. The proposedontology covers the areas of representation of medical concepts/terminology used inhealthcare domain, modelling the healthcare entities including the relations amongthem and accumulating the semantic categories of these medical concepts. Theontology is encoded using the language OWL DL and categorizes all the conceptsinto three groups namely agent-based health care concepts, semantic types of theused concepts and medical concepts associated with the clinical guidelines.

The agent-based healthcare concepts comprise of all the concepts and propertiesrelated with internal structure of the organization which includes medical centres,departments, patients, practitioners and services. It also includes the mapping ofcomplex relations existing between these entities. The semantic types avoids thelanguage ambiguity by using UMLS semantic network. The semantic types definedin the UMLS are divided into two groups and organized as hierarchies of Entity andEvent. The Entity represents the meanings related with healthcare organizationsor the entities and the Event represents the meanings associated with events oractivities involved in a daily care flow. Finally the medical concepts handles thespecific vocabulary used in clinical guidelines and organize the knowledge relatedto the guidelines in terms of Diseases, Procedures and Personal Data. A set ofrelations is defined for all the concepts in the ontology to connect a concept withits identifier, its semantic type, the entity responsible for its execution and theproduced result. The authors also include an Ontology Agent in their previoussystem, HeCaSe2, to provide access to the designed medical ontology.

Isern et al. [2007b] do not describe any experiments conducted.

Isern et al. [2007b] state that the proposed medical ontology benefits the guideline-based execution system by helping identify an appropriate data source and the



suitable actors for accomplishing an action using it. They also claim that the de-signed ontology helps in the adaptation of the execution framework to a particularhealthcare organization without modifying the multi-agent system implementationand provides an application independent context.

The research papers [Isern et al. 2007a], [Isern et al. 2007c], [Bosansky andLhotska 2009] and [Isern et al. 2011] refer to Isern et al.’s [2007b] work.

2.3.2 Summary.


2007b Isern, Sanchez

and Moreno

An Ontology-Driven

Agent-Based ClinicalGuideline Execution

Engine.

Incorporates an application

ontology in HeCaSe2 model, toestablish a common

understanding of the medical

concepts.

Table III. Major Contribution in Interpreting Guidelines

2.4 Remote Monitoring using Computerized Clinical Guidelines

The research papers reviewed in this section describe the role of computerizedclinical guidelines in remote monitoring of the patients, having multiple healthproblems, and the elderly population at hospitals and at homes respectively. Themain objective is to reduce the medical cost, incurred due to the increasing grayingpopulation, and ease out the load on medical practitioners in healthcare domain.The section includes papers from 2008 and 2011.

2.4.1 Saphire: A multi-agent system for remote healthcare monitoring throughcomputerized clinical guidelines. Laleci et al. [2008] address the problem of reducingthe load on the healthcare system, in terms of delivering high quality medicalservices at a reasonable cost, by remotely monitoring patients at home and at thehospital. The authors also address the problem of interoperability between a clinicaldecision support system and the various heterogeneous clinical workflow systems,existing in a healthcare institute, which prevents the adoption of computerizedclinical guidelines by a healthcare institute.

Laleci et al. [2008] do not refer to any previous work.

Laleci et al. [2008] propose a clinical decision support system following the multi-agent approach, for the monitoring of chronic diseases both at hospital and in homeenvironment. The system is based on semantic infrastructure to account for theinteroperability problem and enable its smooth integration with the heterogeneouscomponents, existing within a healthcare institute. The proposed system is alsocapable of deploying and executing clinical guidelines in a distributed environment,consisting of different entities with diverse information systems.

For the system to work efficiently in a distributed environment various com-municating autonomous agents are defined, whose instantiation and elimination is



dynamically controlled based on their requirement. The agent factory agent is re-sponsible for creating a dedicated guideline definition corresponding to a particularpatient, which is then executed by the guideline agent. The access to the electronichealthcare records (EHR) of the patient is provided by the EHR agent, which ismodelled as a separate agent to restrict the access of other agents to EHR. Thesystem intends to solve the interoperability problem by using an ontology agent,which is responsible for mapping the parameters of medical web services, mappingthe parameters of sensor web services, mapping the terminologies used in clinicaldocument content and mapping the content of the electronic healthcare records ofthe patient. The monitoring agent monitors the execution of the clinical guidelinesand also provides an interface to the practitioners for visualizing it. During theexecution of a guideline, various alarms, notifications and reminders are issued tothe medical practitioners, which are handled by the alarm distribution agent. Thisagent is responsible for distributing the alarm messages to the marked recipientsusing an efficient and reliable method. The authors describe the detailed working ofeach agent using the SAPHIRE project that implements the proposed architecture.

Laleci et al. [2008] do not describe any experiments conducted.

Laleci et al. [2008] claim that the proposed system tends to standardize health-care processes, and reduces human error, by making them comply with clinicalguidelines. They also state that the proposed architecture has been implementedin the SAPHIRE project, which has two main applications, bedside monitoring ofsubacute phase of the patients suffering from myocardial infarction and homecaremonitoring of the rehabilitation of the cardiovascular patients undergone a revascu-larization therapy. With these applications, the authors claim to reduce the load onthe medical practitioners and the medical costs incurred in the healthcare processes.

There are no specific references to Laleci et al.’s [2008] work by other researchersin this survey.

2.4.2 Agent-based execution of personalised home care treatments. Isern et al.[2011] address the problem of reducing the cost of medical services in the health-care domain by improving and enhancing home care services, in which the elderlypatients with multiple co-morbid conditions are remotely supervised by the medicalprofessionals. They also address the problem of co-ordination among the medicalprofessionals having diverse skills and knowledge, for an efficient and smooth deliv-ery of the home care services. A patient suffering from multiple medical conditionsdemands for a personalised treatment, which cannot be taken care by the stan-dard clinical guidelines. Hence the authors also address the problem of creatingpersonalized treatment plans for each patient.

Isern et al. [2011] refer to previous work by Isern et al. [2004b].

Isern et al. [2011] state that the previous work by Isern et al. [2004b] does nottake into account the different medical conditions a patient may be suffering from,and hence lacks the capability of personalising the clinical workflows according toa patient’s needs and health status.

Isern et al. [2011] present a distributed system to handle the problems involvedin the delivery of home care services. This system has been realized within the



European K4Care project. The accessibility of the system through a web browsermakes it easy for the users to access it from anywhere. The system is capable ofautomatically adapting to the user’s personal profile and hence restricts the func-tionalities available to a user according to its role in the system. It also allows thecreation of personalized treatment plans, known as Individual Intervention Plans(IIP), by the team of medical professionals assigned to a particular patient. TheseIIPs are then executed by the multiple co-ordinating agents, which according tothe authors is one of the major tasks handled by the described architecture.

The architecture of the proposed system consists of three layers namely knowledgelayer, data abstraction layer and agent-based layer. The knowledge layer consistsof the electronic healthcare records of the patients in the form of hybrid SQL-XMLdatabase, the clinical processes employed in an organisation and the IIPs definedfor all the patients. This layer is also responsible for managing the medical andorganizational knowledge, which is represented using the ontologies. The data inthe knowledge layer comes from diverse sources and may be represented in differentformats such as XML, SQL, OWL etc, which makes it difficult for the agents toefficiently retrieve and interpret the data. Thus to handle the different languagerepresentation of the data and locate the required information in the database, thedata abstraction layer is provided above the knowledge layer. This layer makesthe execution of the home care processes independent of the way and the formatthe data is stored in the knowledge layer. The agent-based layer is responsiblefor co-ordinating all the activities among the agents, where an agent represents apatient or a healthcare professional. The agents in the system are semi-autonomousas some of the activities such as the creation of IIPs needs the intervention of themedical professionals. This layer also provides the SDA executor agents which areresponsible for the execution of the IIPs.

Isern et al. [2011] do not describe any experiments conducted.

Isern et al. [2011] claim that the proposed architecture enhances the flexibility,adaptability and reusability of the system, by making the representation of themedical knowledge independent of its usage in the home care medical services. Oneof the main features of the system, as stated by the authors, is the tools thatallow the medical professionals to define the personalised treatment plans for eachpatient. The authors also claim that their system provides a remote access to theuser for accessing the required medical knowledge and requesting necessary servicesfor the patients.

The research paper [Lhotska et al. 2011] refers to Isern et al.’s [2011] work.



2.4.3 Summary.


2008 Laleci, Dogac,Olduz, Tasyurt,

Yuksel and Okcan

Saphire: A multi-agentsystem for remote

healthcare monitoring

through computerizedclinical guidelines.

Describes a clinical decisionsupport system that is capable

of monitoring chronic diseases

both at hospital and at home.

2011 Isern, Moreno,

Sanchez, Hajnal,

Pedone and Varga

Agent-based execution of

personalised home care

treatments.

Introduces the remote

supervision of elderly patients,

with multiple co-morbidconditions, by creating

personalized treatment plans.

Table IV. Major Contribution in Remote Monitoring using Computerized Clinical Guidelines

2.5 Managing and Customizing Clinical Guidelines

The effective management and customization of clinical guidelines is an importantissue in order to address the changes that a guideline undergoes on daily basis andwith respect to a particular patient’s health condition. This section includes papersfrom 2009 and 2010 that deal with this issue.

2.5.1 An agent-based cognitive approach for healthcare process management.Wu et al. [2009] address the problem of effective management of dynamic clinicaland administrative processes in healthcare domain, to account for regular modifica-tions required in the clinical pathways during a patient’s diagnosis and treatment,and to efficiently handle the interactions between different functional units in thedomain.

Wu et al. [2009] do not refer to any previous work.

Wu et al. [2009] present a three-layer agent-based system architecture for mod-elling the execution and management of clinical processes. The three layers, inthe proposed architecture, consist of the agent layer, the database layer and theinterface layer. The agent layer defines a group of agents, that work in coordina-tion with each other to perform various healthcare tasks, using the defined processrules. The main responsibilities of this group of agents comprise of generating di-agnosis decision and recommending treatments, patient records management andcoordination of treatment processes, scheduling test between patient and service de-partment followed by report generation and delivery, managing operation processand maintaining information about patient’s state during treatment process. Thedatabase layer maintains all the medical information consisting of patient records,treatment records, examination records and diagnostic and treatment information.Each agent is provided a restricted access to the database and can only access asection of the data, relevant to its functionality. This provides a tight control overthe data accessibility and maintenance. Finally the interactions between humanusers and the agents are supported by the interface layer.



The medical knowledge, consisting of clinical pathways/guidelines, forms thebasis for effective working of the proposed architecture. This knowledge is acquiredand represented using an ontology, in which the pathways/guidelines are modelledas clinical algorithms. Each clinical algorithm consists of a scenario node, a decisionnode and an action node and based on these nodes various recommendations suchas messages, drugs and referrals are generated for a clinical process. Finally theauthors describe the working of their approach using the case study of treatmentof primary open angle glaucoma.

Wu et al. [2009] do not describe any experiments conducted.

Wu et al. [2009] state that the proposed system incorporates cognitive approachby modelling medical knowledge in terms of ontology, which facilitates the manage-ment of dynamic processes in health care. The authors also state that their systemimplements real-time dynamic decision making in healthcare processes, rule-basedknowledge engineering and continuous awareness of the environment.

There are no specific references to Wu et al.’s [2009] work by other researchersin this survey.

2.5.2 A semiotic multi-agent modeling approach for clinical pathway manage-ment. The application of clinical pathways/guidelines to the healthcare domainshould account for the dynamic and complex nature of the processes involved inthe domain, as in contrast to the execution of steps in a guideline, a completemedical process involves multiple interacting social agents with each one of themhaving a goal to accomplish. In such an environment, the effective managementof clinical pathways/guidelines thus becomes crucial to allow their dynamic par-ticipation in a medical process and account for the unpredictable situations. Li etal. [2010] address the problem of effective management and customization of clini-cal pathways/guidelines in the healthcare domain so as to increase their executionefficiency.

Li et al. [2010] do not refer to any previous work.

Li et al. [2010] present a new approach for agent based clinical pathway man-agement, in which they use a norm based agent, EDA (Epistemic, Deontic and Ax-iological) to model the internal architecture of the agent and a negotiation modelto effectively schedule services between the service provider agents and the servicerequester agents. The EDA model incorporates the semiotic principles and focuseson the social structure underlying the multi-agent cooperation. The authors alsodescribe the use of a semiotic approach in analysing the requirement of a clinicalpathway, which they identify as a crucial aspect prior to building any simulation.Following the semiotic approach, an agent’s behaviour in a clinical process is af-fected by the creation, exchange and usage of signs during the execution of theprocess. Every agent has two aspects of the semiotics, affordances and norms,stored in them, that enable them to decide and act autonomously. The affordance,which represent the collection of patterns of behaviour, and the norm, which de-scribe the constraints and rules an affordance must follow, are captured from theclinical pathways/guidelines and modelled using the semantic analysis and normanalysis respectively.



In the internal architecture of an agent, the EDA model serves as the knowledgebase consisting of three modules namely the epistemic module, the deontic moduleand the axiological module. The epistemic module stores all the beliefs and capabil-ities concerning the agent itself and its acquaintance agents. This knowledge can beacquired from the ontology chart, using the semantic analysis. The deontic modulestores all the affordances associated with the agent in the form of behavioural normsthat have direct effect on the modification of a clinical pathway. The axiologicalmodule is responsible for resolving conflicting situations of norms in the epistemicand deontic modules by establishing preferred sets of conflicting actions and goalsin the form of evaluative norms. The negotiation process is designed using thespeech act perspective and uses a set of norms instead of sequence of actions, asnorms are more flexible to changes than a sequence. The authors use the clinicalpathway of breast cancer to demonstrate their proposed approach.

Li et al. [2010] do not describe any experiments conducted.

Li et al. [2010] claim to have made a contribution towards the managementof clinical pathways, from the semiotic perspective, by acquiring the necessaryknowledge from the syntactic level to the social level. Through their proposedapproach, they also claim to have provided a conceptual basis for designing a multi-agent simulation model to optimize the clinical pathways or the management systemthat controls the execution of clinical pathways.

There are no specific references to Li et al.’s [2010] work by other researchers inthis survey.

2.5.3 Summary.


2009 Wu, Wang, Yun

and Jiang

An agent-based cognitive

approach for healthcareprocess management.

Introduces a three-layer

agent-based architecture forthe management of dynamic

medical processes.

2010 Li, Liu, Li and

Yang

A semiotic multi-agent

modeling approach for

clinical pathwaymanagement.

Uses EDA model-based agents

for guideline management.

Table V. Major Contribution in Managing and Customizing Clinical Guidelines

3. CONCLUDING COMMENTS

Isern et al. [2004b] appear to be the first to use multi-agent technology in comput-erizing the representation and execution of clinical guidelines. The authors’ work isan improvement over their previous model, HeCaSe, in which a dedicated guidelineagent is introduced to handle the storage, accessibility and modification of guide-lines. With the claim that the features of autonomy and pro-activity are supportedby their system, the authors also state that their system is versatile enough to be



adapted to other situations, other than the healthcare domain. McGrory et al.’s[2008] work handles the computerization of clinical guidelines using the BDI agentmodel and claims it to be an improvement over the traditional CIG model, byavoiding the decomposition of a guideline into separate workflow activities. Isernet al.’s [2004b] work is further improved by Bosansky et al. [2009] by allowing theinterpretation of a clinical guideline to be handled by a set of co-operating agentsand enabling the simultaneous execution of multiple related guidelines. This addsagility and flexibility to the system, as stated by the authors, and covers a widerdomain of related diseases, which is difficult for any physician to cover.

The work by Terenziani et al. [2006] deals with the verification of various exe-cution paths and properties of computerized guidelines, using the model checkingcapabilities of SPIN model checker. Through their approach, the authors claim toprovide a declarative and logical semantics for the clinical guidelines.

The research paper by Isern et al. [2007b] is another important piece of work inthe field, which uses the concept of ontology for building a vocabulary of medicalconcepts, in the healthcare domain. The authors state that having such an ontologycan help the execution framework to adapt easily to any healthcare organization,without requiring modification in the implementation of underlying multi-agentsystem. But the issue of manually creating such a powerful ontology, which requiresdomain expertise, still needs to be addressed.

The research papers by Laleci et al. [2008] and Isern et al. [2011] introduce theuse of computerized guidelines in remote supervision of patients, both at hospi-tal and at home. The authors of both of the papers aim at reducing the burdenon medical practitioners and lowering the costs of medical services, by remotelymonitoring patients suffering from chronic diseases or having multiple health con-ditions. Laleci et al. [2008] claim that their architecture has been implemented inthe SAPHIRE project, having the applications of bedside monitoring of subacutephase of the patients, suffering from myocardial infarction and homecare monitoringof the rehabilitation of the cardiovascular patients, undergone a revascularizationtherapy. Isern et al.’s [2011] work, on the other hand, enables the creation ofpersonalized treatment plans according to each individual patient’s health status,under the guidance of medical professionals. The authors claim that their systemprovide a remote access to the user for accessing the required medical knowledgeand requesting necessary services for the patients. In their future work, the authorsclaim to provide automatic assistance in the construction of personalized treatmentplans.

Wu et al. [2009] and Li et al. [2010] account for the effective management andcustomization of clinical guidelines, participating in dynamic medical processes.In their approach, Wu et al. [2009] uses a three-layer agent-based architecturethat incorporates cognition into agents, by modelling medical knowledge in termsof ontology and claim to implement real-time dynamic decision making. Li et al.[2010], on the contrary, base their approach on the semiotic principles, guidingthe behaviour of agents in the medical process and uses EDA model to design theinternal architecture of the agents. Through their proposed idea, the authors claimto optimize the management of clinical pathways.



4. ANNOTATIONS

4.1 Bosansky et al. 2009

Citation: Bosansky, B. and Lhotska, L. 2009. Agent-based process-critiquingdecision support system. In 2nd International Symposium on Applied Sciences inBiomedical and Communication Technologies, 2009. ISABEL 2009. 1 -6.

Problem. The authors address the problem of using medical guidelines/processesand the multi-agent system approach in a decision support system, in order tosimulate the medical processes and ensure their abidance by the practitioners. Insuch a system, the occurrence of inconsistencies while following the medical guide-lines/processes can be avoided with multiple co-operating agents monitoring theexecution of these medical guidelines/processes.

Previous Work. The authors refer to the previous work by Isern et al. [2007c].

Shortcomings of Previous Work. The authors criticize Isern et al.’s [2007c] systemfor interpreting a single guideline at a given time, as this can result in missing outcrucial information that could have been provided by the simultaneous executionof related guideline(s). They also criticize the approach for using a single agent forinterpretation of a guideline. This can lead to a problem if the agent working ona particular guideline is required to take on some high priority task, in which casethe agent will have to either suspend the execution of current guideline or avoidtaking the high priority task. Both the cases highlight the need for multiple agentsinterpreting a guideline.

New Idea/Algorithm/Architecture. The authors present a new multi-agent sys-tem architecture in which the agents work with the medical guidelines/processes,thereby allowing their simulation and monitoring. They propose to use a set of co-operating agents for interpreting a medical guideline/process, thereby overcomingthe problem of suspending execution of a guideline due to the need for the agentto take up a higher priority task. The proposed architecture also allows the con-current execution of a set of guidelines and does not depend on the physician toselect the guideline, hence providing a better analysis and control for a wider do-main of related diseases. They describe a critiquing decision support system as anapplications of the proposed architecture which monitors the correct execution ofthe medical guidelines/processes and alerts the physician in case of inconsistencies.

In their architecture, each step in the guideline/process is handled by a processagent which simulates and predicts the output of the corresponding step, matchesit with the physician’s output and raises an alert in case of inconsistency. Thedifferent roles in a healthcare domain are maintained by the role agent and theconcrete physicians, nurses, patients and other employees working in a healthcaredomain are represented by the execution agent. All the patient data is maintainedby the patient health record agent which can be accessed through the environmentagent. Finally a process director agent answers the queries of other agents regardingthe information about medical guidelines/processes.

Experiments Conducted. The authors do not describe any experiments conducted.

Results. The authors do not present any results.



Conclusions. The authors claim that their framework is flexible and agile asit allows the simultaneous interpretation of a set of medical guidelines/processesthereby covering a wider domain of related diseases and avoid overlooking of cru-cial information by the physician. They state that their system is based on generalprinciples and hence does not depend on a specific process language and can alsobe applied to other organizational processes. Finally they claim that the systemalso offers several possible ways to alert the healthcare personnel, in case of incon-sistencies.

Citations By Others. [Lhotska et al. 2011].

4.2 Isern et al. 2004b

Citation: Isern, D. and Moreno, A. 2004b. Distributed guideline-based healthcare system. In Proceedings of 4th International Conference on Intelligent SystemsDesign and Applications, ISDA. IEEE Press, Budapest, Hungary, 145-150.

Problem. Considering the increasing cost of medical services in the healthcaredomain, the efficient usage/management of medical resources has become a criticalissue. The authors address the problem of optimizing the use of medical resources byproposing the integration of multi-agent system and the medical guidelines therebyimproving the coordination between the medical services and the humans. By usingautonomous agents to represent different entities and medical services in healthcaredomain, the authors also address the problem of interaction and cooperation amongdifferent processes within a healthcare organization.

Previous Work. The authors refer to previous work by Isern et al. [2003].

Shortcomings of Previous Work. The authors state that their previous work doesnot consider the use of medical guidelines and also lacks the use of dedicated agentsfor various medical services.

New Idea/Algorithm/Architecture. The authors propose a new version of themulti-agent system architecture that is an improvement over their previous workHeCaSe, in order to provide a better coordination and cooperation between hu-mans and medical services using the medical guidelines. The proposed architectureincorporates the medical guidelines in the system by using a guideline agent whichis responsible for accessing, storing and modifying the medical guidelines. The ar-chitecture consists of a medical centre agent which supervises the user’s access tothe main system and provides all the required information to the user through theuser agent. A user agent maintains profile of the user and allows the same to beaccessed by the doctor so as to analyse the user’s medical records before the visit.The user agent is linked to the main system through the broker agent that helpsuser retrieve the required information about the system. Each of the medical cen-tre agents is responsible for monitoring the department agents and various serviceagents. The department agents are further composed of the doctor agents, morespecific service agents and the guideline agents. Finally the medical records of allthe patients are stored in the database by a medical record agent which lies at thebottom of the architecture and provides a secure access to the stored data.

The system also provides scheduling of services in which an appointment betweenthe patient and the doctor and between the patient and the required service is



arranged. The scheduling mechanism takes into consideration different constraintssuch as the patient’s availability, timetable of the services etc, before setting upan appointment. The authors describe the scheduling process using a simplifiedexample of appendicitis.



Conclusions. The authors state that the proposed architecture incorporates thefeature of autonomy and pro-activity in the system, by modelling various entitiesin the healthcare domain as the agents, with the supposition that the modelledentities are autonomous having their own beliefs, desires and intentions. The au-thors also state that their system allows the addition of new agents and featuresand uses standard languages for the agent communication and the representationof ontologies.

Citations By Others. [Isern et al. 2006; Isern et al. 2007a; Isern et al. 2007b;Isern et al. 2011].

4.3 Isern et al. 2006

Citation: Isern, D. and Moreno, A. 2006. Computer-Based Management ofClinical Guidelines: A Survey. In Proceedings of 4th Workshop on Agents Ap-plied in Health Care in conjunction with the 17th European Conference on ArticialIntelligence, ECAI 2006. IOS Press, Riva del Garda, Italy, 71-80.

Focus of Survey. Computerized execution of the clinical guidelines has become animportant area of research and many (semi-) automatic systems are being developednowadays for this purpose. The authors in this survey examine several systems,available for the representation and execution of clinical guidelines, on the basis ofa set of properties and present a comparative study for the same.

Previous Survey. The authors do not refer to any previous survey.

Overlap with Current Survey. Though this survey is not exactly on topic, butthe in-depth analysis of eight relevant systems presented by the authors, includingtheir own system, HeCaSe2, makes an important contribution towards the maintopic and the comparative study also adds value to our survey.

Overview of papers Surveyed. The authors examined eight relevant systems intheir survey, the names for which are SAGE, GLARE, GLEE, NewGuide, Arezzo,DeGel, SpEM and HeCaSe2. These systems have been analyzed and comparedagainst a set of properties which includes GL Repository, GL Editor, GL Represen-tation Language, Agents, Coordination, Run Time Engine, Access to EHR, Standardused and Security. For a better understanding, the authors present the result ofanalysis in a tabular format.

Results of Comparison. The authors state that the adoption of a single standardfor the guideline representation language can solve the interoperability problem, asmost of the systems compared in the survey use the representation languages thatshare the same basic components. Another factor affecting the interoperability of



different systems is the representation of an EMR (Electronic Medical Record), forwhich a best representation is yet to be finalized. The authors also state that noneof the compared systems except the Arezzo, which is used for some specific tasksonly, is being used in any medical centre. This is due to the reason that smooth inte-gration of the execution of computerized guidelines with the workflow of activities,in a healthcare institute, is a big issue which still needs to be resolved. Consideringthe agent-based design of the HeCaSe2 and Arezzo systems, the authors also claimthat the agent-based approach enables the designing of flexible and interoperablesystems, which can be extended to include the services and resources particular toa specific healthcare institute.

Claims/Conclusions. The authors claim that the multi-agent technology is oneof the best approaches for the development of computerized clinical guideline exe-cution systems. The authors base their claim on the autonomous behaviour offeredby the agents, the effective handling of the distributed environment, the dynamicco-ordination provided among the agents, the various communication algorithmsavailable for agent interaction, etc. Finally to assert their claim, the authors de-scribe a mapping between the properties of intelligent agents and the requirementsof a guideline enactment system.

Citations By Others. There are no specific references to this paper by otherresearchers in this survey.

4.4 Isern et al. 2007b

Citation: Isern, D., Sanchez, D., and Moreno, A. 2007b. An Ontology-DrivenAgent-Based Clinical Guideline Execution Engine. In 11th Conference on ArticialIntelligence in Medicine, AIME 2007, R. Bellazzi, A. Abu-Hanna, and J. Hunter,Eds. Lecture Notes in Articial Intelligence, vol. 4594. Springer Berlin / Heidelberg,Amsterdam, The Netherlands, 49-53.

Problem. The computerization of clinical guidelines using a multi-agent systemframework requires the agents to work in coordination over the complex activitiesdefined in any clinical guideline. For such a system to work efficiently the agentsmust share the same understanding of the medical concepts/terminology involved inthe healthcare domain. The authors in this paper address the problem of filling thesemantic gap between the codification of clinical guidelines and their interpretationin the healthcare domain to improve interoperability among the existing systems.

Previous Work. The authors cite previous work by Isern et al. [2004b].

Shortcomings of Previous Work. The authors criticize their previous work forlacking the use of a shared medical ontology among the agents in the system.

New Idea/Algorithm/Architecture. The authors propose the inclusion of an ap-plication ontology, constructed as an external source, in their previous systemHeCaSe2 to enable different entities in the system, represented as agents, to co-ordinate with each other using a shared understanding of the medical concepts inthe healthcare domain. The proposed ontology covers the areas of representation ofmedical concepts/terminology used in healthcare domain, modelling the healthcare



entities including the relations among them and accumulating the semantic cate-gories of these medical concepts. The ontology is encoded using the language OWLDL and categorizes all the concepts into three groups namely agent-based healthcare concepts, semantic types of the used concepts and medical concepts associatedwith the clinical guidelines.

The agent-based healthcare concepts comprise of all the concepts and propertiesrelated with internal structure of the organization which includes medical centres,departments, patients, practitioners and services. It also includes the mapping ofcomplex relations existing between these entities. The semantic types avoids thelanguage ambiguity by using UMLS semantic network. The semantic types definedin the UMLS are divided into two groups and organized as hierarchies of Entity andEvent. The Entity represents the meanings related with healthcare organizationsor the entities and the Event represents the meanings associated with events oractivities involved in a daily care flow. Finally the medical concepts handles thespecific vocabulary used in clinical guidelines and organize the knowledge relatedto the guidelines in terms of Diseases, Procedures and Personal Data. A set ofrelations is defined for all the concepts in the ontology to connect a concept withits identifier, its semantic type, the entity responsible for its execution and theproduced result. The authors also include an Ontology Agent in their previoussystem, HeCaSe2, to provide access to the designed medical ontology.



Conclusions. The authors state that the proposed medical ontology benefits theguideline-based execution system by helping identify an appropriate data source andthe suitable actors for accomplishing an action using it. They also claim that thedesigned ontology helps in the adaptation of the execution framework to a particularhealthcare organization without modifying the multi-agent system implementationand provides an application independent context.

Citations By Others. [Isern et al. 2007a; Isern et al. 2007c; Bosansky andLhotska 2009; Isern et al. 2011].

4.5 Isern et al. 2011

Citation: Isern, D., Moreno, A., Sanchez, D., Hajnal, A., Pedone, G., andVarga, L. Z. 2011. Agent-based execution of personalised home care treatments.Applied Intelligence 34, 2 (April), 155-180.

Problem. The authors address the problem of reducing the cost of medical servicesin the healthcare domain by improving and enhancing home care services, in whichthe elderly patients with multiple co-morbid conditions are remotely supervised bythe medical professionals. They also address the problem of co-ordination amongthe medical professionals having diverse skills and knowledge, for an efficient andsmooth delivery of the home care services. A patient suffering from multiple medicalconditions demands for a personalised treatment, which cannot be taken care bythe standard clinical guidelines. Hence the authors also address the problem ofcreating personalized treatment plans for each patient.



Previous Work. The authors refer to previous work by Isern et al. [2004b].

Shortcomings of Previous Work. The authors state that the previous work byIsern et al. [2004b] does not take into account the different medical conditions apatient may be suffering from, and hence lacks the capability of personalising theclinical workflows according to a patient’s needs and health status.

New Idea/Algorithm/Architecture. The authors present a distributed system tohandle the problems involved in the delivery of home care services. This system hasbeen realized within the European K4Care project. The accessibility of the systemthrough a web browser makes it easy for the users to access it from anywhere. Thesystem is capable of automatically adapting to the user’s personal profile and hencerestricts the functionalities available to a user according to its role in the system.It also allows the creation of personalized treatment plans, known as IndividualIntervention Plans (IIP), by the team of medical professionals assigned to a partic-ular patient. These IIPs are then executed by the multiple co-ordinating agents,which according to the authors is one of the major tasks handled by the describedarchitecture.

The architecture of the proposed system consists of three layers namely knowledgelayer, data abstraction layer and agent-based layer. The knowledge layer consistsof the electronic healthcare records of the patients in the form of hybrid SQL-XMLdatabase, the clinical processes employed in an organisation and the IIPs definedfor all the patients. This layer is also responsible for managing the medical andorganizational knowledge, which is represented using the ontologies. The data inthe knowledge layer comes from diverse sources and may be represented in differentformats such as XML, SQL, OWL etc, which makes it difficult for the agents toefficiently retrieve and interpret the data. Thus to handle the different languagerepresentation of the data and locate the required information in the database, thedata abstraction layer is provided above the knowledge layer. This layer makesthe execution of the home care processes independent of the way and the formatthe data is stored in the knowledge layer. The agent-based layer is responsiblefor co-ordinating all the activities among the agents, where an agent represents apatient or a healthcare professional. The agents in the system are semi-autonomousas some of the activities such as the creation of IIPs needs the intervention of themedical professionals. This layer also provides the SDA executor agents which areresponsible for the execution of the IIPs.



Conclusions. The authors claim that the proposed architecture enhances theflexibility, adaptability and reusability of the system, by making the representationof the medical knowledge independent of its usage in the home care medical services.One of the main features of the system, as stated by the authors, is the tools thatallow the medical professionals to define the personalised treatment plans for eachpatient. The authors also claim that their system provides a remote access to theuser for accessing the required medical knowledge and requesting necessary servicesfor the patients.



Citations By Others. [Lhotska et al. 2011].

4.6 Laleci et al. 2008

Citation: Laleci, G. B., Dogac, A., Olduz, M., Tasyurt, I., Yuksel, M.,and Okcan, A. 2008. Saphire: A multi-agent system for remote healthcare moni-toring through computerized clinical guidelines. In Agent Technology and e-Health,R. Annicchiarico, U. Corts, C. Urdiales, M. Walliser, S. Brantschen, M. Calisti, andT. Hemp ing, Eds. Whitestein Series in Software Agent Tech- nologies and Auto-nomic Computing. Birkhuser Basel, 25-44.

Problem. The authors in this paper address the problem of reducing the loadon the healthcare system, in terms of delivering high quality medical services at areasonable cost, by remotely monitoring patients at home and at the hospital. Theauthors also address the problem of interoperability between a clinical decision sup-port system and the various heterogeneous clinical workflow systems, existing in ahealthcare institute, which prevents the adoption of computerized clinical guidelinesby a healthcare institute.

Previous Work. The authors do not refer to any previous work.

Shortcomings of Previous Work. The authors do not mention any shortcomingsof the previous work.

New Idea/Algorithm/Architecture. The authors propose a clinical decision sup-port system following the multi-agent approach, for the monitoring of chronic dis-eases both at hospital and in home environment. The system is based on semanticinfrastructure to account for the interoperability problem and enable its smoothintegration with the heterogeneous components, existing within a healthcare in-stitute. The proposed system is also capable of deploying and executing clinicalguidelines in a distributed environment, consisting of different entities with diverseinformation systems.

For the system to work efficiently in a distributed environment various com-municating autonomous agents are defined, whose instantiation and elimination isdynamically controlled based on their requirement. The agent factory agent is re-sponsible for creating a dedicated guideline definition corresponding to a particularpatient, which is then executed by the guideline agent. The access to the electronichealthcare records (EHR) of the patient is provided by the EHR agent, which ismodelled as a separate agent to restrict the access of other agents to EHR. Thesystem intends to solve the interoperability problem by using an ontology agent,which is responsible for mapping the parameters of medical web services, mappingthe parameters of sensor web services, mapping the terminologies used in clinicaldocument content and mapping the content of the electronic healthcare records ofthe patient. The monitoring agent monitors the execution of the clinical guidelinesand also provides an interface to the practitioners for visualizing it. During theexecution of a guideline, various alarms, notifications and reminders are issued tothe medical practitioners, which are handled by the alarm distribution agent. Thisagent is responsible for distributing the alarm messages to the marked recipientsusing an efficient and reliable method. The authors describe the detailed working ofeach agent using the SAPHIRE project that implements the proposed architecture.





Conclusions. The authors claim that the proposed system tends to standardizehealthcare processes, and reduces human error, by making them comply with clini-cal guidelines. They also state that the proposed architecture has been implementedin the SAPHIRE project, which has two main applications, bedside monitoring ofsubacute phase of the patients suffering from myocardial infarction and homecaremonitoring of the rehabilitation of the cardiovascular patients undergone a revascu-larization therapy. With these applications, the authors claim to reduce the load onthe medical practitioners and the medical costs incurred in the healthcare processes.


4.7 Li et al. 2010

Citation: Li, W., Liu, K., Li, S., and Yang, H. 2010. A semiotic multi-agentmodeling approach for clinical pathway management. Journal of Computers 5, 2,266-273.

Problem. The application of clinical pathways/guidelines to the healthcare do-main should account for the dynamic and complex nature of the processes involvedin the domain, as in contrast to the execution of steps in a guideline, a completemedical process involves multiple interacting social agents with each one of themhaving a goal to accomplish. In such an environment, the effective managementof clinical pathways/guidelines thus becomes crucial to allow their dynamic par-ticipation in a medical process and account for the unpredictable situations. Theauthors in this paper address the problem of effective management and customiza-tion of clinical pathways/guidelines in the healthcare domain so as to increase theirexecution efficiency.



New Idea/Algorithm/Architecture. The authors present a new approach for agentbased clinical pathway management, in which they use a norm based agent, EDA(Epistemic, Deontic and Axiological) to model the internal architecture of the agentand a negotiation model to effectively schedule services between the service provideragents and the service requester agents. The EDA model incorporates the semioticprinciples and focuses on the social structure underlying the multi-agent cooper-ation. The authors also describe the use of a semiotic approach in analysing therequirement of a clinical pathway, which they identify as a crucial aspect prior tobuilding any simulation. Following the semiotic approach, an agent’s behaviour ina clinical process is affected by the creation, exchange and usage of signs during theexecution of the process. Every agent has two aspects of the semiotics, affordancesand norms, stored in them, that enable them to decide and act autonomously. Theaffordance, which represent the collection of patterns of behaviour, and the norm,



which describe the constraints and rules an affordance must follow, are capturedfrom the clinical pathways/guidelines and modelled using the semantic analysis andnorm analysis respectively.

In the internal architecture of an agent, the EDA model serves as the knowledgebase consisting of three modules namely the epistemic module, the deontic moduleand the axiological module. The epistemic module stores all the beliefs and capabil-ities concerning the agent itself and its acquaintance agents. This knowledge can beacquired from the ontology chart, using the semantic analysis. The deontic modulestores all the affordances associated with the agent in the form of behavioural normsthat have direct effect on the modification of a clinical pathway. The axiologicalmodule is responsible for resolving conflicting situations of norms in the epistemicand deontic modules by establishing preferred sets of conflicting actions and goalsin the form of evaluative norms. The negotiation process is designed using thespeech act perspective and uses a set of norms instead of sequence of actions, asnorms are more flexible to changes than a sequence. The authors use the clinicalpathway of breast cancer to demonstrate their proposed approach.



Conclusions. The authors claim to have made a contribution towards the man-agement of clinical pathways, from the semiotic perspective, by acquiring the neces-sary knowledge from the syntactic level to the social level. Through their proposedapproach, they also claim to have provided a conceptual basis for designing a multi-agent simulation model to optimize the clinical pathways or the management systemthat controls the execution of clinical pathways.


4.8 McGrory et al. 2008

Citation: McGrory, J., Grimson, J., Clarke, F., and Gaffney, P. 2008.Software agents representing medical guidelines. Dublin Institute of Technology.

Problem. The computerization of clinical guidelines using the traditional CIG(Computer-Interpretable-Guidelines) model works by decomposing a clinical guide-line into separate workflow activities, which are managed by a centralized inferenceengine. This fragmentation results in the loss of the actual context represented bya guideline and increases the size and complexity of the system, with the additionof more guidelines. The authors in this paper address the problem of computeriz-ing the clinical guidelines using the BDI (Belief-Desire-Intention) agent paradigm,thereby accounting for the issues highlighted in the CIG models.



New Idea/Algorithm/Architecture. The authors present a BDI agent approachfor computerizing the clinical guidelines, in which each guideline is represented by



a single autonomous agent. This approach retains the original context representedby a guideline and reduces the complexity of the system, as each agent has its owninference engine in contrast to having a centralized engine. For the BDI agent shell,the authors tend to use the modified version of the Jadex agent shell known as theAutonomous Socialising Knowledge agent (ASK-agent). In the ASK-agent model,a guideline is decomposed into workflow activities with each guideline having adedicated inference engine for making decisions. These guidelines communicatewith other resources in the system by exchanging messages.

An agent in the ASK-agent model captures the information attributes, the de-liberative and action attributes and the motivational attributes of a guideline in itsbelief, intention and desire components respectively. This information is extractedfrom the guidelines using the Medical Logic Module (MLM) of the Arden Syntax.A particular ASK-agent thus embeds all the information contained in a guidelineand serves as a knowledge source for the other interacting agents in the system.The authors describe the step-by-step procedure for converting a guideline to itscorresponding ASK-agent.



Conclusions. The authors claim that their approach allows a guideline’s contentto be embedded in an autonomous standalone agent module, having a dedicatedinference engine. This agent can then be used as a knowledge source for the otheragents searching for the information related to the guideline. The authors also statethat their concept provides a solution to the management of clinical guidelines inthe way similar to that provided by the narrative guidelines.


4.9 Terenziani et al. 2006

Citation: Terenziani, P., Giordano, L., Bottrighi, A., Montani, S., andDonzella, L. 2006. Spin model checking for the verification of clinical guide-lines. In ECAI 2006 Workshop on AI Techniques in Healthcare: Evidence-BasedGuidelines and Protocols.

Problem. The verification of various properties of computerized clinical guide-lines and their application in the healthcare domain becomes an important aspectin order to ensure their correct implementation and execution. The verification ofclinical guidelines is also required as they undergo frequent changes during theirreal time application in diagnosis and treatment processes. Moreover, for a givenscenario, the verification process can automatically test the guideline for all itspossible execution paths. The authors in this paper address the problem of verifi-cation of clinical guidelines along with their specification, following the agent-basedparadigm.





New Idea/Algorithm/Architecture. The authors present an agent-based approachfor modelling the clinical guidelines, which uses the GLARE (Guideline Acquisition,Representation and Execution) system as the basis for acquiring the guidelinesthrough a user interface. These guidelines, including the agents interacting withthem, are then modelled as processes in the specification language Promela. Thisspecification language forms the part of the model checker SPIN, that the authorspropose to use as an extension to the GLARE architecture for the specificationand verification of the clinical guidelines. The clinical guidelines acquired using theGLARE system are mapped to processes in Promela, using their XML specificationsin GLARE. The patient record database and all the agents in the system are alsomodelled as Promela processes. The model checker SPIN then transforms eachPromela process into a finite automaton and computes an asynchronous interleavingproduct of automata. This resulting automaton can be built on-the-fly duringthe verification process and depicts the global state space of the system i.e themodel consisting of all the possible executions of the guideline. The SPIN usesthis global state space for the verification of the guideline property. The SPINconverts the negation of the temporal formula, corresponding to the input guideline,into a Buchi automaton and computes its synchronous product with the systemglobal state space. The empty language of the resulting Buchi automaton signifiesthat the guideline property holds true on all the possible executions of the clinicalguideline, otherwise the verifier provides the execution path on which the propertyis false. The proposed approach provides automata-based semantics for the clinicalguidelines and allows the verification of different types of properties, concerning agiven clinical guideline.

Experiments Conducted. The authors illustrate the working of their proposedidea by using the verification process on the clinical guideline for ‘stroke’.

Results. The authors claim that the verification process found some discrepanciesin the expression of the ‘stroke’ guideline, which corresponds to information thatwas deliberately removed from the guideline.

Conclusions. The authors state that their approach provides an automatic wayto verify the properties of clinical guidelines, using the model checker SPIN. Theyalso claim that the knowledge represented by the guidelines is modelled using thePromela formalism, which allows its active participation in the overall verificationprocess. The authors also claim to provide a more declarative and logical semantics,as compared to the automata-based semantics, for the clinical guidelines.


4.10 Wu et al. 2009

Citation: Wu, B., Wang, M., Yun, H., and Jiang, H. 2009. An agent-basedcognitive approach for healthcare process management. In 8th IEEE InternationalConference on Cognitive Informatics, 2009. ICCI ’09. 441-447.

Problem. The authors in this paper address the problem of effective managementof dynamic clinical and administrative processes in healthcare domain, to account



for regular modifications required in the clinical pathways during a patient’s di-agnosis and treatment, and to efficiently handle the interactions between differentfunctional units in the domain.


Shortcomings of Previous Work. The authors do not highlight any shortcomingsof the previous work.

New Idea/Algorithm/Architecture. The authors present a three-layer agent-basedsystem architecture for modelling the execution and management of clinical pro-cesses. The three layers, in the proposed architecture, consist of the agent layer,the database layer and the interface layer. The agent layer defines a group ofagents, that work in coordination with each other to perform various healthcaretasks, using the defined process rules. The main responsibilities of this group ofagents comprise of generating diagnosis decision and recommending treatments,patient records management and coordination of treatment processes, schedulingtest between patient and service department followed by report generation anddelivery, managing operation process and maintaining information about patient’sstate during treatment process. The database layer maintains all the medical infor-mation consisting of patient records, treatment records, examination records anddiagnostic and treatment information. Each agent is provided a restricted access tothe database and can only access a section of the data, relevant to its functionality.This provides a tight control over the data accessibility and maintenance. Finallythe interactions between human users and the agents are supported by the interfacelayer.

The medical knowledge, consisting of clinical pathways/guidelines, forms thebasis for effective working of the proposed architecture. This knowledge is acquiredand represented using an ontology, in which the pathways/guidelines are modelledas clinical algorithms. Each clinical algorithm consists of a scenario node, a decisionnode and an action node and based on these nodes various recommendations suchas messages, drugs and referrals are generated for a clinical process. Finally theauthors describe the working of their approach using the case study of treatmentof primary open angle glaucoma.



Conclusions. The authors state that the proposed system incorporates cognitiveapproach by modelling medical knowledge in terms of ontology, which facilitatesthe management of dynamic processes in health care. The authors also state thattheir system implements real-time dynamic decision making in healthcare processes,rule-based knowledge engineering and continuous awareness of the environment.


REFERENCES

Bosansky, B. and Lhotska, L. 2009. Agent-based process-critiquing decision support system.

In 2nd International Symposium on Applied Sciences in Biomedical and Communication Tech-nologies, 2009. ISABEL 2009. 1 –6.



de Clercq, P., Blom, J., Korsten, H., and Hasman, A. 2004. Approaches for creating

computer-interpretable guidelines that facilitate decision support. Artificial Intelligence in

Medicine 31, 1, 1–27.

Isern, D. and Moreno, A. 2004a. Agent-based careflow using CPGs. In Research and De-velopment (Proceedings of Set Congrs Catal d’Intelligncia Artificial), CCIA 2004, J. Vitri,

P. Radeva, and I. Aguil, Eds. Recent Advances in Artificial Intelligence, vol. 113. IOS Press,

Barcelona, Catalonia, 293–300.

Isern, D. and Moreno, A. 2004b. Distributed guideline-based health care system. In Proceedingsof 4th International Conference on Intelligent Systems Design and Applications, ISDA. IEEE

Press, Budapest, Hungary, 145–150.

Isern, D. and Moreno, A. 2006. Computer-Based Management of Clinical Guidelines: A Survey.

In Proceedings of 4th Workshop on Agents Applied in Health Care in conjunction with the 17th

European Conference on Artificial Intelligence, ECAI 2006. IOS Press, Riva del Garda, Italy,71–80.

Isern, D., Moreno, A., Sanchez, D., Hajnal, A., Pedone, G., and Varga, L. Z. 2011. Agent-

based execution of personalised home care treatments. Applied Intelligence 34, 2 (April),

155–180.

Isern, D., Sanchez, D., and Moreno, A. 2007a. Agents and clinical guidelines: Filling thesemantic gap. In Proceedings of the 2007 conference on Artificial Intelligence Research and

Development. IOS Press, Amsterdam, The Netherlands, The Netherlands, 67–76.

Isern, D., Sanchez, D., and Moreno, A. 2007b. An Ontology-Driven Agent-Based Clinical

Guideline Execution Engine. In 11th Conference on Artificial Intelligence in Medicine, AIME2007, R. Bellazzi, A. Abu-Hanna, and J. Hunter, Eds. Lecture Notes in Artificial Intelligence,

vol. 4594. Springer Berlin / Heidelberg, Amsterdam, The Netherlands, 49–53.

Isern, D., Sanchez, D., and Moreno, A. 2007c. Hecase2: A multi-agent ontology-driven guide-

line enactment engine. In Proceedings of the 5th International Central and Eastern European

Conference on Multi-Agent Systems and Applications V. CEEMAS ’07. Springer-Verlag, Berlin,Heidelberg, 322–324.

Isern, D., Sanchez, D., Moreno, A., and Valls, A. 2003. HeCaSe: An Agent-Based System

to Provide Personalised Medical Services. In Proceedings of Workshop Agentes Inteligentes en

el Tercer Milenio en X Conferencia de la Asociacin Espaola para la Inteligencia Artificial,CAEPIA 2003. Donosti, Vasque Country, 1–10.

Laleci, G. B., Dogac, A., Olduz, M., Tasyurt, I., Yuksel, M., and Okcan, A. 2008. Saphire:

A multi-agent system for remote healthcare monitoring through computerized clinical guide-

lines. In Agent Technology and e-Health, R. Annicchiarico, U. Corts, C. Urdiales, M. Walliser,S. Brantschen, M. Calisti, and T. Hempfling, Eds. Whitestein Series in Software Agent Tech-

nologies and Autonomic Computing. Birkhuser Basel, 25–44.

Lhotska, L., Bosansky, B., and Dolezal, J. 2011. Integration of procedural knowledge in

multi-agent systems in medicine. In Proceedings of the Second International Conference onInformation Technology in Bio- and Medical Informatics. ITBAM’11. Springer-Verlag, Berlin,

Heidelberg, 82–95.

Li, W., Liu, K., Li, S., and Yang, H. 2009. An agent based approach for customized clinical

pathway. In Proceedings of International Forum on Information Technology and Applications,2009. IFITA ’09. Vol. 2. IEEE, Chengdu, China, 468 –472.

Li, W., Liu, K., Li, S., and Yang, H. 2010. A semiotic multi-agent modeling approach for clinicalpathway management. Journal of Computers 5, 2, 266–273.

McGrory, J., Grimson, J., Clarke, F., and Gaffney, P. 2008. Software agents representingmedical guidelines. Dublin Institute of Technology, No page numbers available.

Sutton, D. and Fox, J. 2003. The syntax and semantics of the proforma guideline modeling

language. Journal of the American Medical Informatics Association (JAMIA) 10, 5, 433–443.

Terenziani, P., Giordano, L., Bottrighi, A., Montani, S., and Donzella, L. 2006. Spin model

checking for the verification of clinical guidelines. In ECAI 2006 Workshop on AI Techniquesin Healthcare: Evidence-Based Guidelines and Protocols. No page numbers available.



Terenziani, P., Molino, G., and Torchio, M. 2001. A modular approach for representing and

executing clinical guidelines. Artificial Intelligence in Medicine 23, 3, 249–276.

Wooldridge, M. 2002. An Introduction to MultiAgent Systems. John Wiley and Sons.

Wu, B., Wang, M., Yun, H., and Jiang, H. 2009. An agent-based cognitive approach for health-care process management. In 8th IEEE International Conference on Cognitive Informatics,

2009. ICCI ’09. 441 –447.


use of multi-agents in computerizing clinical...

Documents