ADAPT@AGENT.

HOSPITAL: Agent-based Optimization and Management of Clinical Processes

University of Hohenheim, Department of Information Systems II, Schwerzstrae 35, D-70599 Stuttgart, Germany

University of Wrzburg, Chair of Computer Science VI, Am Hubland, D-97074 Wrzburg, Germany

Abstract: There are several continuing challenges within the healthcare domain. On the one hand, there is a greater need for individualized, patient oriented processes in diagnostics, therapy, nursing and administration, and on the other hand, hospitals have extremely distributed decision processes and strong local (individual) autonomy with a high degree of situational dynamics. This research focuses on developing an information system that can substantially increase the efficiency of hospital process management. We present a system for agent based simulation and support for clinical processes. In particular, we describe the architecture of the system and its functionalities and the integration of existing FIPA, DICOM and HL7 standards. We discuss an example scenario for "clinical trials" to illustrate how the system supports distributed clinical process management. This system interacts with other multiagent systems within the Agent.Hospital Framework and hospital information systems in the eHealth Lab. This research is part of the German Priority Research Program (SPP) 1083 "Intelligent Agents and their application in business scenarios".

Keywords: multi-agent systems, simulation, health care, clinical trials

An earlier version of the manuscript received the Best Paper award for the Intelligent Agent and Multi-Agent Systems in Business mini-track at the Americas Conference on Information Systems (AMCIS 2003), Tampa, FL, August 4-6, 2003.

[email protected]

Introduction

Over the last few years, the cost of health services in Germany has become an important area of economic analysis. The budget for the German health care increased to 218 Billion during the year 2000 with a growth of 5-10 % every year after that. Thus, the current expense of health care is forcing the German government to take a closer look at the macroeconomic cost structures within this industry and find new ways to minimize the overall cost. Information technology is expected to play a major role in the reduction of expenses and the stabilization of health care budgets. With a view to curb expenditure, recently, the biggest IT project for German health care was started, which has introduced a new electronic eHealth-Card until 20061. One of the major problems that need to be addressed is hospital logistics, which is rather complex and highly distributed. Considerable research is underway in this area that is designed to analyze the domain as well as develop software systems, particularly for management and control of information flows and business processes. The approaches to this domain are two-fold. On the one hand, simulation systems are used for analyzing the domain, planning, and re-engineering of business processes. On the other hand, development of management and control systems for information flow and business processes is of interest. The increasing importance of information systems is interlinked with specifics of the (german) healthcare domain, such as highly fragmented, disparate professional groups, which complicates or particularly blocks the integration of innovative service processes as well as new and more efficient information systems. Several important questions have been examined by the German Priority Research Program (SPP 1083)2 since 2000. The main goal of SPP 1083 is to investigate the utilization of agent technologies within large realistic business scenarios and the identification of further research needs. Basic supposition of the SPP is that agent-based development and connection of decentralized information systems generates important benefits by supporting inter-organizational business processes and organizational flexibility. Examination of these hypotheses is supported by the development of large agent-based software systems.

http://www.dimdi.de http://www.realagents.org

[email protected]

In this paper we introduce the essential questions and the research approach addressed by the ADAPT project, which is part of the previously mentioned SPP 1083. We discuss the systems developed for simulation and application and their integration into the Agent.Hospital Framework (REF), which is part of the Agentcities research network. In addition we address questions on how to realize large scaled agent-based distributed service networks that support healthcare standards like HL7 or DICOM. An example scenario on clinical trials illustrates our approach. At the end of the paper we discuss the shortcomings and give an outlook on further research steps and implementation activities.

Research Problem & Main Objectives

The overall objective of our research is to substantially increase the efficiency of hospital processes. The main problem in optimization is that the processes as well as the systems are inherently distributed and the effects of changes are difficult to foresee. Under these circumstances, agent technology is the appropriate solution. Optimization by agents can basically be reached in two ways: a) optimization potential can be identified by agent-based simulation experiments, and b) flexible agent based information systems can optimize the real system. Integrating both of these components, simulation models can be used as a test bed for the development of agent applications. Our empirical studies have identified the following critical success factors that help achieve the objectives of being able to optimize the system: Completeness and topicality of information available to (human) actors Knowledge about the actual patient status helps trigger other changes Adequate representation of the variety and the dynamics of inter-/intra process interactions Based on our experience, we contend that a lot of domain knowledge and inter dependencies is necessary to create an appropriate agent-based system. On the other hand, even more knowledge would be needed by conventional information systems. The adaptive abilities of agents allow dealing with even uncertain conditions and distributed knowledge. In cooperation with the clinical partners, the ADAPT project (Heine et al., 2003) tried to optimize the processes as well as the information flow between the participating oncology and radiation therapy departments. 3

[email protected]

As a result of this, resource allocation, time scheduling and tactical planning should improve with respect to efficiency and control. The experiences gained should allow us to draw conclusions about the usability as well as the advantages of agent based software. The specific motives and goals of this research are: Improvement of distributed appointment scheduling Appointment scheduling for treatment and examination tasks in hospitals is inherently distributed between various organizational units and there are a lot of interdependencies between the processes and the actors. This fact makes it very difficult to determine the effects of single action and to optimize the process. The scheduling and negotiation strategies used in general practice are mainly dominated by simple ad hoc solutions, which can be optimized. Therefore, simulation and analysis of scheduling scenarios and finding more sophisticated processes is an important objective of the ADAPT project. Using clinical trials on chrono-modulated medication, an agent-based system was designed and implemented in order to support the trial scheduling and coordination of patients, medical staff, and equipment. Support for decisions about participation in clinical trials Clinical trials obligate the participating hospitals to perform prescribed treatments and examinations. As one of the benefits, the cost calculation should show whether the incurred liabilities are covered or not. According to this, the ADAPT system should allow cost-benefit estimations based on simulations which take into account medical considerations and individual preferences. Agent technology is used to handle the high level of environmental dynamics as well as the complex and restrictive requirements of clinical trials. Utilization and trial specific statements are supported by realistic simulation results. Operationalization of study protocols Due to the complexity and regimentation of trials, a tremendous effort is needed for the coordination of all the actors and equipments involved. A continuing goal of the ADAPT project is to advance the agent based simulation system towards a real-time assistance system that enables online resource allocation. Supporting the operational processes by coordinating the flow of information (adapted to the individual medical pathway) using agent technology should result in a higher level of process efficiency.

[email protected]

Agent.Hospital Framework

To evaluate the agent systems that have been developed and to show that they interoperate in a sophisticated way, a complex and realistic evaluation scenario is needed. Therefore, the hospital logistics working group of SPP 1083 including the ADAPT-project developed an extensive model called Agent.Hospital (Kirn et al., 2003). This model is based on an open framework for numerous healthcare actors. The basic architecture of the framework is shown in Figure 1. The Agent.Hospital framework contains several detailed healthcare models, different kinds of service agents, and agent-based platforms. Within Agent.Hospital, one can evaluate modeling methods or examine configuration problems as well as agent-based negotiation strategies and coordination algorithms. In the context of the Priority Research program (SPP 1083) the research partners involved in the hospital logistics working group integrated the partial hospital logistics models created by the individual projects. Due to the high number of participating research groups, a wide spectrum of relevant clinical processes could be offered. Relevant organizational structures, processes and necessary data models were analyzed, formalized and modeled at several hospitals. To be able to integrate a number of separate partial models, it was necessary to define numerous gateways between these models (see figure 1) and to develop a common ontology for interaction through these gateways. Additionally, basic process patterns had to be defined (for instance planning and execution of clinical trials with oncological patients). At the conceptual level, Agent.Hospital consists of partial models, process patterns, gateway specifications and shared ontologies. The resulting framework is an overall conceptual model based on the integration of project specific partial models. In Agent.Hospital several clinics, departments and wards are defined as service provider and consumer units. In our framework, the services offered and the services requested are denoted simply as services. For a meaningful interaction between the service provider and the service consumer unit, there needs to be a well defined interface based on existing agent communication languages, interaction protocols, content languages, as well as common ontologies.

[email protected]
Kliniken & Stationen Internal Medicine Uni Hohenheim/ Hosp. Sindelfinden (Referring) Physicians TU Ilmenau/ Physicians Thuringia, OnkoNet Reception Ward Uni Hohenheim/ Hosp. Sindelfingen Gastro-Enterology Uni Wrzburg/ Univ.-Hosp. Wrzburg Cancerology TU Ilmenau/ Univ.-Hosp. KIM II Jena Surgery Ward Uni Mannheim/ Hosp. Pegnitz, Hosp. Kulmbach N.N. (/)

Radiology & Radiotherapy Conv., CT, MRT Cardiological Laboratory Uni Mannheim/ Hosp. Pegnitz & Hosp. Kulmbach Transport Serv. Uni Freiburg/ Univ.-Hosp. Freiburg Uni Mannheim/ Hosp. Pegnitz, Hosp. Kulmbach Radiotherapy Uni Wrzburg/ Univ.-Hosp. Wrzburg Radiodiagnosis Uni Freiburg/ Univ.-Hosp. Freiburg N.N. (/)

Operating Theatre Surgery Block Ambulance TU Ilmenau/ Ambulances Thuringia Emergency Dept. TU Berlin/ Hosp. Charit Berlin Univ Trier/Hosp. Barmh. Brder Trier Dept. of Surgery Uni Mannheim/ Hosp. Pegnitz & Hosp. Kulmbach Op.-Planning TU Berlin/ Charit Berlin N.N. (/) Intensive Care Unit TU Berlin/ Hosp. Charit Berlin

Figure 1: Application diagram of the Agent.Hospital framework with several selected supply chains Currently the following supply chains of the overall framework shown in Figure 1 have been implemented: clinical trials and radio therapeutics (ADAPT), rescue patient (AGIL), lung cancer treatment (ASAinlog), angina pectoris (MedPAge), gallstone therapy, surgery processes (Policy Agents), and radiological service processes (EMIKA). More details and references to these projects are provided via the RealAgentS website, the public groupware tool of the SPP (http://www.realagents.org). For the successful development of Agent.Hospital framework, the use of existing and established standards was very important. In agent research, the Foundation for Intelligent Physical Agents (FIPA) plays the leading role. During the last few years, the FIPA organization has proposed primary standards for the design of agent platforms as well as standards for communication and interaction. To be able to integrate the different agent systems it was necessary to define numerous gateways using these standards (see Figure 2) and to develop common ontologies for interactions among the respective gateway agents. Intelligent software agents in any application have to be integrated into an existing (often proprietary) information system infrastructure. Therefore, the aim of the hospital logistics working group is the implementation, evaluation und documentation of agent based health care services, which are supposed to be the foundation for future FIPA-Application Specifications. Since 2002, the priority research program 1083 is a 6

[email protected]

member of FIPA and has been tasked to develop and evaluate examples and to refine the existing application specifications in health care.
Gateway Agent A Gateway Agent B

MAS A

MAS B

MAS Interaction

FIPA-comliant MAS platform (i.e. JADE)

FIPA-comliant MAS platform (i.e. FIPA-OS)

Figure 2: Gateway Agent Concept (Krempels et al 2003) Besides working with FIPA, the SPP 1083 has initiated cooperation with the association of information system producers in health care (VHGit), to develop specifications for interfaces between specialized applications (practice systems, patient record, etc.). Further more, the SPP contributes to Agentcities (Willmott et al., 2003), an open, worldwide net of FIPA-compliant agent platforms, using which different agent based services - often in the development stage - are provided. Currently there are more than 140 agent platforms that are members of Agentcities with a high concentration in Europe and also several platforms in the USA, Australia and the near east. Within Agentcities, a lot of different application domains exist, for instance eHealth, manufacturing control, digital libraries, travel services, and so on. The ultimate aim of this initiative is to support the commercial as well as academic efforts for the development of agent based applications and to make it possible to compose dynamic, intelligent and autonomous agents as well as complex service agents. One goal of Agent.Hospital was to become a part of Agentcities. Therefore, new Agentcities platforms have been set up in five German cities. They are connected by a central directory service (Agent.HospitalDF) in Aachen, which registers all participating service agents. Different infrastructural services and tools have been set up or developed to realize Agent.Hospital: Agent.Hospital Directory Facilitator (AHDF3): a common and centralized yellow pages service for the agents within Agent.Hospital. It provides a web interface for online monitoring and several other extended functions like availability testing.

http://www-i4.informatik.rwth-aachen.de/agentcities/RWTH_Agentcities_Web_files/hospital.html

[email protected]

Agent.Hospital Event Service (AHES): an event service realized by an agent. It can be used for event based simulation of multi agent systems in the healthcare domain. Agent.Hospital Actor Agent (AHAA): another common component to all the involved projects is the Actor agent. Instances of the actor agent represent real actors in Agent.Hospital scenarios. Agent.Hospital Simulation Model (AHSim): an executable simulation model describing a typical patient process providing interfaces for agent based application systems. Agent.Hospital Ontology Repository (AHOR): a repository providing healthcare specific task ontologies (Becker et al 2002) on demand. The shared ontologies are base for mutual understanding of the systems. Agent.Hospital CVS (AHCVS): a concurrent version system containing the source files of the service and interface agents.

Technical Base and Prototypical Solutions

To optimize scheduling and coordination of clinical processes, intelligent agents for clinical care were developed within the ADAPT-Project. A set of personal assistance agents was developed for solving coordination problems of different actors (physicians, patients, nurses) in the hospital. These agents support their specific users in different tasks such as the administration of the personal datebook, making decisions or scheduling of shifts or medical actions. The basic foundation for the A4Care assistance system is the FIPA-compliant platform JADE. The central directory service of Agent.Hospital supports the process of finding suitable interaction partners. The agents of the assistance system have the following responsibilities: Physician agent A physician agent can compose single tasks to a treatment plan and assign it to a specific patient. In doing so, it can specify order restrictions and possible parallel sequences. As an extension, we plan to integrate a library of predefined clinical pathways. Patient agent: This assistance agent takes over the automatic or semi-automatic scheduling of the assigned plan. It acts respecting the whishes of its owner and presents the results in the personal datebook. The user can do manual rescheduling and define new policies including personal restrictions and preferences. The actual negotiation is then processed in the background by connecting the assistance agents to the functional units.

[email protected]

Functional unit agent: This agent is a proxy person for a functional unit. It coordinates the incoming appointment requests and sends appointment proposals automatically. Here the current schedule, opening times and additional restrictions (e.g. for reduced personal during lunch) are respected. A collateral monitoring of appointment execution starts an automatic rescheduling if discrepancies between the planned and actual schedules occur. Users can grant different grades of autonomy to their agents and decide, when an agent is allowed to act autonomously and when it has to make further enquiries to the users. A hight grade of autonomy on the one hand increases the overall performance of the system, on the other hand, it requires that preferences of the users be formalized very exactly and updated regulary. Additional agents to the ones described above are realized for the medical staff and wards to solve the problem of making fair shift schedules. In addition to the development of these agent-based services running on FIPA-Platforms, we need to develop a realistic environmental model. This model is well populated with agents, this time agents representing human actors. An integral part of the ADAPT project is the development of the agent based simulation system called SeSAm (ShEll for Simulated Agent systeMs) (Klgl et al., 2001). Originally developed for the simulation of social insects (Klgl et al., 1998), it was used to implement environmental simulations for the ADAPT project as well as for comprehensive Agent.Hospital framework. SeSAm is an integrated environment for modeling and simulating multi-agent systems and provides powerful modeling functions for the easy construction of complex models. As a special feature, it offers visual modeling of all the tasks to support domain-experts that do not have programming skills in creating simulation models. It allows the visual construction of agents and their environment respecting their properties, abilities and behaviour. Different agent classes can be created and their properties and abilities can be defined by just adding agent specific attributes and modular features.

[email protected]

Figure 3: Visual modeling and simulation with the SeSAm-Environment There are several useful functions available within SeSAm, for example movement, evolution and scheduling and the given set can be extended by the user. Using these features and the basic set of primitive actions (e.g. Move, ChangeImage, SendMessage) and functions (e.g. GetAllObjectsOnPosition) the behavior of the agent can be described. The basic representation of the agents behavior is the UML-activity diagram (see Figure 3). They are easily understandable and a common notation well known to system developers. Activity diagrams consist of activities and transition rules. The activities can be seen as states containing a series of actions, which are executed until an exit rule activates the next activity. After modelling the agent classes, simulation situations can be created on a two dimensional map. For the simulation of a scenario, this model representation is compiled and can be run directly from within the environment. During simulation it is possible to watch the animated map, pause and resume the simulation and to control agents states. Recently, time several additional features to support handling of more complex models like healthcare scenarios are integrated in the environment (Oechslein et al., 2002). Among various other features, the environment allows hierarchical modeling at the level of function description, as well as at the level of behavior

10

[email protected]

description. Basic function primitives can be composed to more complex user defined primitives and parts of the activity graph can be combined together to form compound activities. To simplify testing and finding errors in the model, it is possible to create rules for invariant states, similar to assertions in modern programming languages. Even refactoring methods are supported, which are also quite popular in traditional programming. Experienced modellers can get the power of a programming language without having to learn any syntactic notions. To support the connection to external agent systems additional extensions were required. Therefore, new features to support communication as well as ontologies were added. Here, general standards were adhered in order to ensure interoperation with other systems. The new plug-ins provide FIPA-compliant communication primitives and a possibility to import ontologies modeled with the ontology tool Protege 2000.

Figure 4: Engineering agent systems in a simulated environment vs. application in real world In our view, agent-based applications are part of an environment containing real world agents like users and other systems (possibly agent systems). In agent software development isolated tests are usually very difficult, instead tests have to be carried out under the conditions of this environment. As mentioned above, our goal was to extend SeSAm in a way that it can be used as a test-bed for agent based software without remodeling the agent for interacting with the simulated environment (Klgl et al., 2003). Then, creating models of the global system becomes rather intuitive. Figure 4 shows a sketch of this approach. An agent based application is integrated in an environment consisting of additional information systems and users, which can also be seen as 11

[email protected]

agents. All agents from the real-world environment are replaced by simulated agents in SeSAm. The level of detail of the simulation is dependent on the requirements of the agent system. Due to the representation of the real-world in the simulated test-bed, the developed components can be tested both under realistic as well as extreme conditions. This usually cannot be done in the real world for reasons of security and prohibitive costs. Especially in the healthcare domain, practical testing of prototypical software is usually dangerous. In this context, SeSAm offers furthermore powerful capabilities for experimentation support like distribution of simulation runs on several hosts and online graphical analysis of the simulation results like the ones shown in Figure 5.

Figure 5: Online graphical analysis of simulated processes

A Scheduling Scenario

The example scenario discussed here relates to scheduling in the radiation therapy units. Based on empirical investigations we have implemented a detailed SeSAm model describing the patient scheduling in a clinic for radiation therapy. The clinic consists of a reception, where the patients appointments are planed, waiting rooms and treatment rooms as well as radiation devices. Functional units and devices have specific opening times and a current schedule. One of the typical patient processes for radiation therapy is assigned by the doctor after a patients arrival. Often, these processes are also referred to as clinical pathways. Once the pathway is assigned, the necessary medical actions (treatment and diagnosis) have to be planned and the patient

12

[email protected]

has to visit the functional units. A typical problem is that the schedule of the real treatment execution always differs from the preplanned schedule. Here we wanted to find new scheduling strategies to improve the process taking into consideration predefined quality criterions like waiting time of patients and utilization ratio of resources. Extensions of the model can also simulate late patients and uncertain treatment times.

Figure 6 Flexible configuration of the simulation. Number of agents and opening times of the functional units may vary. Different settings representing the situation in other clinics can be created very easily. The number and kinds of functional units as well as their opening times can be varied (see Figure 6). Alternative clinical pathways can be assigned to patients. The probability of a new patients arrival, as well as potential tardiness are adjustable. All this makes the model to be a flexible testing environment

13

[email protected]

Figure 7 shows a screenshot of a running simulation. Some patients are waiting for admission. The selected patient is waiting in the queue in front of the tumor localization room (Called Lok). This unit has currently opening hours (shown by the green light) whereas the radiation therapy units are still closed (indicated by the red light). Figure 7 also shows the schedule of the selected patient and the appointment for Vorlok, which is also known as the localization room.

Figure 7: Weekly schedules of patients and radiation therapy devices. Based on this model it is possible to evaluate different strategies, for example, first-come first-served or a preplanned schedule that takes into consideration priorities of examinations. Different hypotheses of dynamic changes (patients that come too late, additional emergencies) can be evaluated with their effects on evaluation criterions and plan stability.

14

[email protected]

As a first step, our experiments were able to show plausible reactions of the system that are similar to the current real world behavior. In the next step we want to connect the simulation engine to the negotiation based reactive planning system A4Care, which is very effective in a distributed dynamic environment (Herrler et al. 2002).

A Clinical Trial Scenario

An important field in health care is the medical treatment of cancer. Typically, these treatments are very expensive and new clinical trials are conducted to test the effectiveness of various new drugs and treatment protocols. Clinical trials are carefully controlled studies in which oncology experts evaluate better and cost effective ways to treat, prevent, or diagnose cancer. Increasing the number of qualified patients who are enrolled in clinical trials, as they represent an opportunity for patients with cancer to receive the best possible care is really important. Clinical trials are also extremely valuable because they answer important questions that will help to continually improve cancer care and decrease the risk of cancer development. Before a new therapy or medical treatment can be put into daily clinical practice, a lot of trials have to be run. These trials are detailed plans for medical treatments, for instance a clinical trial protocol can describe at which point of time, in which quantum, and how medications or therapies have to be executed. But clinical trials are not performed in linearity; instead their concrete structures depend on specific patient constitutions, laboratory results, etc. Clinical trials can be described using the graph metaphor, each node representing a particular state of a particular study. Typical sample sizes for clinical trials require about 150 to 200 patients which make it necessary in general that several hospitals cooperate in performing such trials (multi-centric trials). Figure 8 exemplifies the process aspects of the integrated clinical trial scenario. The simulation according to the process description is directed by SeSAm and utilizes several agent-based services provided by the SPP 1083. At the beginning of a clinical trial various tasks for diagnosis and treatment have to be coordinated and resources have to be scheduled and if necessary communicated to the corresponding units. Figure 8 shows an example with CT- (computer tomography) and MRT- (magnetoresonance-tomography) examinations as well as the execution of a surgery. 15

[email protected]

Request study

ADAPT SeSAm

documentalist

Create study plan

7.1 Patient Information RequestStudyPlan ADAPT SeSAm

Ereignis

ward

Receive findings

Local study center

arrange Medium term appointment

ADAPT Daisiy

7.1 Study Plan Result Patient accepted (y/n) ADAPT SeSAm

Funktion/ Aktivitt

Physician (ward)

Interpret findings Out-patient clinic

Hospitalization necessary

Organisations einheit

x
Dismiss patient from study Patient does not meet requirements Patient meets requirements

Admit patient (administration)

1.1.1 Patient Information AddNewDocument

Person

ward Admit patient (ward)

Konnektor

ASAinlog
Documentary service Draw up test plan

Give orders for MRI (chest, abdomen, pelvis) Request MRI (chest, abdomen, pelvis) Receive request for MRI (chest, abdomen, pelvis) 2.5.3.1 Patient Information AddPatient(Treatment List) Give orders for Ultrasound (US) (abdomen) 1.8.1.1 Patient Information AddPatient(Treatment List)

Physician (ward)

Give orders for surgery 1.9.1.1 Patient Identification requestTask(Prefere nce List,Negotiation Weight)

Physician (ward)

Physician (ward)

ward

Request surgery

ward

ward

ASAinlog

ASAinlog

Request US (abdomen)

ASAinlog

Surgical ward

Receive request for surgery

radiology

Internal medicine

Receive request for US (abdomen)

Surgical ward

Scheduling

radiology

Scheduling

Internal medicine

Scheduling

1.9.1.2 Surgical ward Return Date & Time Patient Information setAppointment (Schedule) radiology Return Date & Time

PolicyAgents

MedPAge

2.5.3.2 Patient Information GetPatientSchedule

Internal medicine

Return Date & Time

MedPAge

1.8.1.2 Patient Information GetPatientSchedule

ward

Receive Date & Time

Patient Information modifyDocument

ward

Receive Date & Time

Patient Information modifyDocument

ward

ASAinlog

Receive Date & Time

Patient Information modifyDocument

ASAinlog

Date & Time for surgery is known

ward

ASAinlog

Take patient to examination

ward

Take patient to examination

ward

Take patient to surgical ward

Patient is in examination room

Specialist is not in examination room

ADAPT SeSAm

Patient is in examination room

Specialist is not in examination room

ADAPT SeSAm

Surgical ward

Perform surgery radiology

6.3 Doctor Identification Notify(Message) Internal medicine

6.4 Doctor Identification Notify(Message)

Call Specialist

Call Specialist

Surgical ward

Write surgery report

EMIKA

EMIKA

Surgical ward send surgery report to ward Surgical ward Return patient to ward radiology

Examine patient

Specialist is in examination room Internal medicine

Examine patient

Specialist is in examination room

ward

Receive surgery report

ward

Treat patient

radiologist

Document findings

Internal med. specialist

Document findings

Release patient from hospital

radiology

Send findings to ward

Internal medicine

Send findings to ward

Figure 8: Exemplary part of the clinical trial scenario with interactions between different multiagent systems

16

[email protected]
SRM^S01-S11^SRM_S01
MSH ARQ [ APR ] [ { NTE } ] [ { PID [ PV1 ] [ PV2 ] [ { OBX } [ { DG1 } } ] { RGS [ { AIS [ APR ] [ { NTE } ] [ { AIG [ APR ] [ { NTE } ] [ { AIL [ APR ] [ { NTE } ] [ { AIP [ APR ] [ { NTE } ] }

Schedule Request Message

] ]

} ]

Resource Group Segment Appointment Information - Service Appointment Preferences Notes and Comments

} ]

Appointment Information - General Resource Appointment Preferences Notes and Comments

} ]

Appointment Information - Location Resource Appointment Preferences Notes and Comments

} ]

Appointment Information - Personnel Resource Appointment Preferences Notes and Comments

Figure 9: HL7 Schedule Request Message structure (SRM) in the clinical trial scenario. At first, the suitability of the patient for the clinical study will be checked (age, gender, blood count etc.). If the patient fulfils the preconditions for the trial for instance, an MRT examination is necessary. Planning of this examination means that different service agents try to negotiate mandatory appointments (see exemplary the gateways 2.5.3.1/2.5.3.2 in Figure 8). But these agent-based negotiations are constrained by the existing timetable of the corresponding hospital information system. To interact with this information system (precondition is HL7 compliance) the HL7 agent generates specific HL7 Schedule Messages like SRM (Schedule Request Message). There are several health care standards development efforts currently underway throughout the world. Health Level Seven is one of several American National Standards Institute (ANSI) accredited Standards Developing Organizations (SDOs) operating in the healthcare arena. The application level addresses definition of the data to be exchanged, the timing of the interchange, and the communication of certain errors to the application. It supports such functions as security checks, participant identification, availability

17

[email protected]

checks, exchange mechanism negotiations and, most importantly, data exchange structuring4. Figure 9 shows the HL7 Schedule Request Message structure. Based on this message structure, the actual Schedule Message (SRM^S01) is generated, which is shown in Figure 10. The responsible HL7 Agent is able to receive, edit, process and send such HL7 Messages to standards compliant hospital information systems (HIS). Currently, detailed tests with different HIS have been planned. The agents communicate in a FIPA compliant way and transmit the HL7 as well as the DICOM Messages as content of the FIPA-ACL messages (Agent Communication Language). ACR (the American College of Radiology) and NEMA (the National Electrical Manufacturers Association) formed a joint committee to develop a Standard for Digital Imaging and Communications in Medicine. This Standard is developed in liaison with other Standardization Organizations including CEN TC251 in Europe and JIRA in Japan, with review also by other organizations including IEEE, HL7 and ANSI in the USA. The DICOM Standard pertains to the field of Medical Informatics. Within that field, it addresses the exchange of digital information between medical imaging equipment and other systems. Because such equipment may interoperate with other medical devices, the scope of this Standard needs to overlap with other areas of medical informatics. However, the DICOM Standard does not address the breadth of this field5.

MSH|^~\&|||||20030101||SRM^S01|1234|1234|2.4 ARQ|0001|0002|||||0010^radiological
examination^ROUTINE|NORMAL|120|min|200303150800^200303201700|P

rioritt|||4711^Mayer^Hans^^Dr.||||4712^Otto^Karl^^Dr. PID|1111|2222|3333|4444|Mustermann^Hans|Mustermann|19500521|M DG1|001|I9|786.5|CRANIAL PAINS|199401010730|W DG1|002|I9|412|OLD MYOCARDIAL INFARCTION|199401010730|W

Figure 10: SRM Message for an MRT examination. A Screenshot of the HL7 Message Agent interface is shown in Figure 11. Relevant message segments and fields can be edited by the user and sent to the next responsible agent or hospital information system.

http://www.hl7.org http://medical.nema.org/

18

[email protected]

Agent based Application

HIS

PACS

RIS

eHealth Lab Environment

Figure 11: HL7 Message Agent for interactions with HL7-compliant HIS. After the MRT examination, several HL7 messages have to be sent pertaining to the results of the examination in the current scenario. But the transfer of radiological image data is historically grown and still not supported by HL7. The transfer of image data for instance from the MRT modality to viewing workstations, digital archives or a remote radiologist for a second opinion is supported by the DICOM standard (our basic assumption is that all systems are DICOM compliant with specific conformance statements). The following interaction diagram (Figure 12) shows a simplified process of the system interactions within Agent.Hospital. This diagram depicts just the essential steps and the interactions for error handling are not included (e.g. notunderstood, refuse, failure, etc.).

19

[email protected]
ADAPT ADAPT SeSAm SeSAm ADAPT ADAPT DAISIY DAISIY ASAinlog ASAinlog

addPatient(patient)

reject(patient)

accept(patient) request(study plan) inform(study plan) addDocument(study plan) inform-done

Figure 12: Simplified interaction diagram in the integrated clinical trials scenario The Screenshot of the DICOM Message Agent interface shows the viewing and editing (if allowed) functions (Figure 13). The DICOM image data with all header information can be sent by the agent to the next responsible agent or archiving system.

Agent based Application

HIS

PACS

RIS

eHealth Lab Environment

Figure 13: DICOM Message Agent for interaction with DICOM-compliant PACS/RIS etc.

20

[email protected]

All these interactions have first been tested in a simulated environment provided by SeSAm and in the second step implemented and tested in connection with real hospital information systems in the secure laboratory environment called eHealth Lab6.

Summary & Outlook

Agent technology is able to handle the high level of environmental dynamics as well as complex and restrictive requirements of clinical processes. The aim of ADAPT is to optimize the planning and management of clinical trials as well as provide decision support in this context. In order to accomplish this, realistic simulation results are needed. In this research we have employed a multi-agent based simulation because conventional simulation has been shown to be not sufficient (Klgl et al. 2002). As far as multi agent modeling is concerned, we have improved existing modeling techniques to handle complex models with intelligent actors. Another unique aspect of our approach to materializing realistic simulation scenarios is the integration of partial models within the Agent.Hospital initiative. FIPA compliance and subsequent use of ontologies facilitate this interoperation and also enables the development of agent systems, which can be deployed from the simulation environment to an existing information system infrastructure. The prototype system development has been completed. There are already demo systems, which can explicate the possibilities of our approach. While the initial feedback is very encouraging, we still have to demonstrate the practical use of our system. Our immediate goal is to bring the agent systems to a controlled real life experiment and show that the results of simulation and the system can be practically used in the management and control of clinical trials. We are at the point, where we have gotten the initial results from the simulation and now we can evaluate the experimental agent systems and deploy them to existing information system infrastructure in the near future. Both applications (simulation/management) will be evaluated with respect to specific advantages of agent-based software over traditional systems.

http://www.ehealthlab.de/index_e.html

21

[email protected]

References

Becker, M., Heine, C., Herrler, R. & Krempels, K.-H. (2003). OntHoS - an Ontology for Hospital Scenarios. In: Applications of Software Agent Technology in the Health Care Domain (Moreno, A., Nealon, J. Eds.). Whitestein Series in Software Agent Technologies (WSSAT), Birkhuser Publisher Basel, pp. 87-103. Heine, C., Herrler, R., Petsch. M. & Anhalt, C. (2003). ADAPT Adaptive Multi Agent Process Planning & Coordination of Clinical Trials. In: AMCIS 2003 Proceedings, Tampa/Florida (Best Paper Award). Herrler, R., Heine, C. & Klgl, F. (2002) Appointment Scheduling Among Agents: A Case Study In Designing Suitable Interaction Protocols. In: AMCIS 2002 Proceedings, S. 1456-1463, Dallas. Kirn, S., Heine, C., Herrler, R. & Krempels, K.H. (2003). Agent.Hospital agent-based open framework for

clinical applications. In Kotsis, G., Reddy, S. (eds.): Twelfth IEEE International Workshops on Enabling Technologies: Infrastructure for Collaborative Enterprises (WET ICE 2003 PostProceedings). CS Press, Los Alamitos, CA
Klgl, F., Herrler, R. & Oechslein, C. (2003). From Simulated to Real Environments: How to use SeSAm for software development. In: Multiagent System Technologies (Schillo, M. et al. Eds.) - 1st German Conference MATES03, (LNAI 2831). Klgl, F.; Puppe, F. ; Raub U. & Tautz, J. (1998). Simulating Multiple Emergent Behaviors - exemplified in an Ant Colony in Proc. of "Artificial Life VI", Los Angeles, June 26-29, 1998; C. Adami, R. Belew, H. Kitano, and C. Taylor, eds., MIT Press. Klgl, F. (April 2001). Multagentensimulation Konzepte, Werkzeuge, Anwendungen (Multi-Agent Simulation Concepts, Tools, Application) Addison Wesley Klgl, F., Oechslein, C., Puppe, F. & Dornhaus A. (2002). Multi-Agent Modelling in Comparison to Standard Modeling In: AIS'2002 (Artificial Intelligence, Simulation and Planning in High Autonomy Systems), F. J. Barros and N. Giambiasi (Eds.) SCS Publishing House.

22

[email protected]

Krempels, K.-H., Nimis, J., Braubach, L., Pokahr, A., Herrler, R. & Scholz, T.(2003). Entwicklung intelligenter Multi-Multiagentensysteme - Werkzeuguntersttzung, Lsungen und offene Fragen. (Development of intelligent Multi-Agent Systems Tool-Support, Solutions and open Questions.) In: INFORMATIK 2003 Innovative Informatikanwendungen (Dittrich, K., Knig, W., Oberweis, A., Rannenberg, K., Wahlster, W. Eds.), GI-Edition Lecture Notes in Informatics, P-34. Kllen Publisher, Bonn, Vol.1. Oechslein, C., Klgl, F., Herrler, R. & Puppe, F. (2002). UML for Behavior-Oriented Multi-Agent Simulations In: Dunin-Keplicz, B., Nawarecki, E.: From Theory to Practice in Multi-Agent Systems LNAI; Second International Workshop of Central and Eastern Europe on Multi-Agent Systems, CEEMAS 2001 Cracow, Poland, September 26-29, 2001, Revised Papers Springer 2002, Heidelberg, ISBN 83-915953-0-7 Willmott, S., Constantinescu, I., Dale, J., Somacher, M., Marinheiro, R., Mota, L., Bothelo, L., Bonnefoy, D., Picault, J., Poslad, S., Tan, J. & Bothelo, L. (2003) Agentcities Network Architecture Recommendation. http://www.agentcities.org/rec/00001/actf-rec-00001a.pdf, [2003-01-27].

23