Agent-based Modeling and Simulation Software Architecture

for Health Care

Karam Mustapha and Jean-Marc Frayret

Polytechnic University of Montreal, Mathematical and Industrial Engineering Department

2500, Chemin de Polytechnique, Montreal, H3T 1J4, Canada

Keywords: Health Care, Agent-based Simulation, Colon and Colorectal Cancer.

Abstract: Health Care (HC) organizational structure and related management policies are essential factors of HC sys-

tem. They can be tested through simulations in order to improve HC performance. To simplify the design of

these simulations we have proposed a modelling approach based on an additional structure. The modelling

approach considers the complexity of the modelling process, where in the various models are developed.

This approach is organized according to two main abstraction levels, a conceptual level and a simulation

level. We developed a computer simulation environment of patient care trajectories using the agent in order

to evaluate new approaches to increase hospital productivity and adapt hospital clinical practice conditions

for the elderly and patients with multiple chronic diseases. For that, we have developed a multi-agent

framework to simulate the activities and roles in a HC system. This framework can be used to assist the col-

laborative scheduling of complex tasks that involve multiple personals and resources. In addition, it can be

used to study the efficiency of the HC system and the influence of different policies.

1 INTRODUCTION

Health Care (HC) is a rich domain for multifaceted

simulation studies. The conceptual and architectural

modeling is challenging due to the diverse and com-

plex dimensions. In this domain, simulation general-

ly aims at experimenting and testing management

policies or organizational designs in a controlled

environment in order to understand their economi-

cal, human and environmental consequences. This

paper deals with the simulation of cancer patients’

pathways.

Almost 88% of the Canadian population over the

age of fifty1 (41 % women and 46 % men) will de-

velop some form of cancer during their lifetime.

Lung, breast, colon, rectal and prostate cancers rep-

resent more than half of all new cancer cases (52 %).

Colon and rectal cancers are the third most common

cancers among men and women and are considered

the second leading cause of cancer death among men

and the third among women.

With the aging population and the intricacy of

http://www.cancer.ca/~/media/cancer.ca/CW/cancer%20informat

ion/cancer%20101/Canadian%20cancer%20statistics/canadian-

cancer-statistics-2013-FR.pdf

the medical system, the management of HC activi-

ties has become increasingly complex. Therefore,

simulation is a relevant tool to model this complexi-

ty and improve its operations. In particular, agent-

based modeling and simulation significantly extend

the capabilities of simulation approaches such as

discrete-event simulation as discussed in the next

section.

Providing high-quality care is a priority among

health professionals. However, resources are limited

and their utilization must be optimized in order to

meet high quality standards and patients’ unique pro-

files. Therefore, the challenge faced by HC providers

and managers is to design organizational and medical

processes that deliver the right treatment, to the right

patient, at the right time using the right resources.

Factors, such as socio-demographic and environmen-

tal characteristics, as well as the characteristics of the

organizational and decision-making systems, can be

used to simulate patient care trajectories, from their

diagnosis to the end of the treatment.

In this paper, we propose to study the efficiency

of organization decisions which aims at: i) describ-

ing the HC organization; ii) modelling and simulat-

ing the behaviours and decisions of its actors and iii)

implementing these decisions and observe their local

Mustapha, K. and Frayret, J-M.

Agent-based Modeling and Simulation Software Architecture for Health Care.

DOI: 10.5220/0005972600890100

In Proceedings of the 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2016), pages 89-100

ISBN: 978-989-758-199-1

and global effect on the HC, and iv) supporting each

step with specific conceptual and software support.

This paper presents different objectives, the first

objective presented requires the agent-based model-

ling and simulation of complex behaviours, decision-

making processes and interactions between hospital

staff and patients. The most appropriate technology to

simulate these complex mechanisms is Agent-Based

modelling and Simulation (ABS). The second objec-

tive is therefore to create and validate the patient

agent model, which includes a physiological model of

how the cancer evolves in time in response to specific

treatments. Also, to simulate a large number of pa-

tients treated simultaneously with the same resources

of the hospital; this step of the project is only con-

cerned with the general behaviour of the patient agent,

and how well it can be configured in order to simulate

colon and colorectal cancer patients with different

attributes. As for third objective, healthcare decision

makers need reliable tools to support them in decision

making for adapting policies to help cut costs or re-

duce waiting time, and to provide visualization which

allows them to rehearse innovative ideas before they

are implemented.

Contributing to aforementioned objectives, we

aim to developing a computer simulation environ-

ment of patient care trajectories using the agent in

order to evaluate new approaches to increase hospi-

tal productivity and adapt hospital clinical practice

conditions for the elderly and patients with multiple

chronic diseases. Ultimately, the simulation model

will include: the physical health of the patient; the

cognitive state of the patient; the psychosocial state

of the patient; the hospital resources, staff and phy-

sicians. For that, we have developed a multi-agent

architecture to simulate the activities and roles in a

HC system. This architecture can be used to assist

the collaborative scheduling of complex tasks that

involve multiple personals and resources. In addi-

tion, it can be used to study the efficiency of the HC

system and the influence of different policies.

First, this paper describes the general scope of

this simulation project and presents an up to date

ABS. Next, the general conceptual model of the

simulation is described and finally simulation results

are presented.

2 LITERATURE REVIEW

Many research projects are based on the agent para-

digm to model and/or simulate complex systems.

Indeed, this paradigm provides a tailored approach

to model complex systems by explicitly addressing

the study of the interactions and behaviours of their

components. The design of HC agent-oriented mod-

els is a difficult task that requires the use of specific

knowledge and skills. This section defines ABS and

introduces a detailed analysis of ABS applications in

the medical domain. Finally, this section also pre-

sents different ABS development framework.

2.1 Agent-based Simulation

ABS is an abstract representation of reality that in-

volves the elaboration of a descriptive model, which

reproduces the behaviour of the system by modelling

its components, including their decision-making

capabilities and interactions patterns, as agents. An

agent can be defined as an entity, theoretical, virtual

or physical, capable of acting on itself and on the

environment in which it evolves, and capable of

communicating with other agents (Jennings et al.,

1998).

Research in ABS is prolific. It is known under

different labels, including multi-agent simulation,

individual-based models and agent-based models.

These tools are part of a more generic technology

known as multi-agent systems; this domain of appli-

cations is much larger than simulation. In the litera-

ture, the concept of agent is generally defined as

(Jennings et al., 1998) "…a computer system situat-

ed in an environment, which is a way autonomous

and flexible to achieve the objectives for which it

was designed."

In practice, the multi-agent paradigm is used at

two levels: for modelling and for simulating. At the

first level, it is required to create multi-agent models

that (1) reproduce the naturally distributed structure

of the studied systems, or (2) propose a representa-

tion of complex problems. Such models can be used

for developing reactive, deliberative or hybrid agent

models. The second level involves the simulation

(i.e. experimentation with these models). Such a

simulation may or may not be based on distributed

software architecture. In other words, the operational

simulation model is not necessarily multi-agent. It

may be object-oriented or translated into other simu-

lation languages (e.g. DEVS) (Quesnel et al., 2007)

2.2 Agent-based Simulation in the

Health Care Domain

HC operation management is a domain that is well

suited to ABS because it involves many people in-

teracting with their own decision-processes. With

agent-based modelling, it is possible to explicitly

model these individuals and their interactions. How-

SIMULTECH 2016 - 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

ever, although ABS is growing in the medical do-

main, applications to the real world are still rare

((Nealon and Moreno, 2003), (Devi and Mago,

2005)).

In the medical domain, (Mustafee et al., 2010)

identifies 200 papers, in which simulation is used.

More than 70% of these applications used Monte

Carlo simulation, while 20% used Discrete-Event

Simulation, less than 9% used System Dynamics,

and finally only 1% used ABS.

For instance, (Stainsby et al., 2009) uses ABS to

reorganize hospital emergency departments. Recent-

ly, several simulation techniques have been used in

conjunction to capture different dimensions. (Knight

et al., 2012) use DES and ABS to model a healthcare

system, in which patients choose their hospital based

on a linear additive service function of three factors

(i.e., hospital reputation, travel distance, waiting

time). Finally, (Figueredo and Aickelin, 2011) pro-

poses one of the first systematic studies aiming at

comparing SD and ABS based on a simple mathe-

matical model of interactions between a tumour and

immune cells. The authors concluded that both mod-

elling paradigms are not always equivalent.

In most organizational simulations in the medical

field, agents, whether patients, doctors or nurses are

of reactive type and their behaviour is very specific

to the purpose of the simulation. In (Kazar et al.,

2008), the author discussed the introduction of a

multi-agent system into the medical field, which

helps the management take decisions and actions,

and also ensures the communication and coordina-

tion by reducing the errors of diagnosis and treat-

ment, and by improving time required for the medi-

cal resources, and other medical departments. How-

ever, (Kanagarajah et al., 2010) (Laskowski et al.,

2009) use simulation in order to analyse the perfor-

mance of an emergency department in different con-

figurations. In these studies, agents are used to mod-

el resources that move through the hospital with

predefined process time. In (Jones and Evans, 2008),

modelling deals mainly with the different types of

treatment associated with their time and resource

requirements, which then become predefined in the

simulation. Only patient’s arrival time and resource

availabilities change dynamically. In these models,

the agents travelling times within the hospital is pre-

defined. However, it can also be dynamically com-

puted in the simulation as in (Zhang and Yao, 2010),

which models the evacuation of a hospital undergo-

ing a fire, or in (Krizmaric et al., 2005) that use sim-

ulation to study different transport configurations for

clean and dirty equipment in the hospital.

Also, some authors proposed the concept of an

online medical service system for internet users using

a multi agent system, the user can get access to the

details of the closest and best health care system such

as hospital, medical clinic, etc. (Gupta et al., 2012).

However, (Han et al., 2006) used the medical sensor

modules with combinations of wireless telecommuni-

cation technology based in the multi-agent system.

The papers (Iantovics, 2008) (Gupta and Mukho-

padhyay, 2012), proposed a hybrid system with hu-

man and artificial agent members. (Gupta and

Mukhopadhyay, 2012) proposed an operational algo-

rithm to describe the operations of a hybrid multi

agent system based intelligent medical diagnosis sys-

tem called Clinical Diagnosis System (CDS) (Gupta

and Pujari, 2009). Also, (Mahmud et al., 2009) pre-

sented hybrid architecture of a multi-agent consulta-

tion system for obesity oriented health problems.

Some authors propose a multi-agent oriented

learning environment aimed at learning using a posi-

tive approach to perform diagnostic reasoning and

modelling of a domain (Rosa et al., 2003). In (Chao

and Wong, 2009), the authors proposed the model of

practical data mining diagnostic which intends to

support real medical diagnosis by two emerging

technologies - data mining (Zhang et al., 2005) and

multi-agent system (Foster et al., 2005)(Klusch et

al., 2003). In the next section we present the patient

agent models.

3 PATIENT AGENT MODELS

In the literature review, we have presented various

research based on the definition of methodologies to

guide the designers in the development of multi-

agent models in general. However they present a

number of weaknesses related to modelling HC, and

their simulation, for example, at present, there is no

generally accepted health care ontology for generat-

ing and analysis of medical or health care infor-

mation. This makes it difficult to communicate be-

tween several systems developed in different areas.

Also, other limitations related to the framework can

be synthesized by the following: (i) the absence of

an approach which ensures the passage of the con-

ceptual level to the implementation level; (ii) the

transition from design to implementation is costly in

time and development efforts; (iii) consideration of

the organization; (iv) multi-modelling and v) time

management.

In this study we proposed a modelling approach

based on an additional structure to consider the

complexity of the modelling process, where in the

various models is developed. The real system is first

Agent-based Modeling and Simulation Software Architecture for Health Care

represented by the HC domain and then the overall

modelling approach is based on an incremental ap-

proach in which different models are developed. As

shown in Figure 1, the expert’s fields of intervention

are specified including different models: conceptual

modelling and simulation oriented agents.

Figure 1: A schematic of the Modelling Process.

In the process of modelling and simulation patient

trajectory, the model distinguishes three main steps:

the conceptual modelling, the conceptual agent

modelling, and operation modelling (Simulation

Oriented Agents - SOA).

The patient is the central actor of the HC system

or real system. It interacts with many resources, in-

cluding physicians, nurses and equipment. Its dy-

namic condition is the main driver of resource utili-

zation, and its reaction to treatment defines the sys-

tem quality level. In order to design such an agent,

different models are proposed to describe its place in

the overall system, and its complex behaviour.

Conceptual modelling is based on several mod-

els specifying the nature of the agents and the archi-

tecture of multi-agent system. In the following, it is

for the programmer to operationalize the conceptual

model agent. Each agent identified at the conceptual

level is specified and implemented according to the

constraints related to the development environment.

It is always for the programmer, to take into account

the technical constraints ignored at the simulation.

Thus, the multi-agent system will be deployed in a

software environment enabling its execution to con-

duct simulation experiments.

The conceptual modelling and conceptual agent

modelling are described next. With regard to operat-

ing step, it will be addressed in the next section on

the architecture simulation support.

3.1 Conceptual Model

The general conceptual model proposed in this study

defines the main interactions between the patient and

its environment (Figure 2). It is composed of four

dimensions and includes different aspects of the

patient, its environment, and the HC system. These

dimensions are related to physiology of the patient,

the psychosocial state and support of the patient, the

decision processes and the resources used to treat the

patient. The links between the different aspects iden-

tified within these four dimensions represent their

mutual dependencies. The central part represents the

patient agent. The other parts represent the hospital

staff involved in the treatment selection, as well as

patient support (e.g., family members, nurses) (for

more details see the reference (Gilli et al., 2014)).

Psychosocial dimension: The psychological

dimension includes an emotional model of the pa-

tient agent and its social influences, especially in the

form of support from family members and nurses.

This model describes a response to specific situa-

tions and will eventually contribute to measuring the

patient quality of life during treatment.

Physiological Dimension: this dimension in-

cludes both the patient’s health model (its general

physical and health condition) and its cancer evolu-

tion model. Both are affected by treatment in differ-

ent manners, while influencing each other. In prac-

tice, this dimension includes on the one hand, the

absolute physiological state of the patient and can-

cer, and, on the other hand, the perception of this

state obtained from observations (e.g., analysis,

scans, and biopsies). While the initial information is

not necessarily known, the subsequent information

can be out-dated, and more or less accurate. Finally,

in this model, the patient health model is influenced

by his or her emotional model.

Decision Dimension: This dimension includes

both the patient’s and the physician’s decision mod-

els. It represents the main actors’ decision-making

processes and preferences that contribute to treat-

ment selection and treatment implementation. It is

the part of the conceptual model that directly con-

tributes to the decision and implementation of pa-

tient care trajectories. Here, the patient decision

model is influenced by its health and emotional

models, while the physician decision model is influ-

enced by the patient cancer and health models. The

patient decision model also contributes to plan each

individual treatment according to the system re-

source availabilities.

System Dimension: The system dimension rep-

resents the virtual hospital resources and processes.

SIMULTECH 2016 - 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

Treatment

plan

Resource

Model

Physician decision

model

Patient decision

model

Cancer evolution

model

Patient health

model

Patient emotional

model

Support

Treatment

implementation

Decision

Physiology

Psychosocial

Patient a

ent

Physician agent

System

Figure 2: General conceptual model.

When a physician requests a type of treatment,

it must be planned according to the hospital priority,

the workload of the resources required for this kind

of treatment, as well as the preferences of the pa-

tient. The different sub-models of these dimensions

influence each other in order to emulate the general

relationships between the patient, his/her cancer, the

medical staff, and the patient's support. The relation-

ship between the patient and the hospital processes

and resources are addressed through the dynamic

specification of the treatment program into the care

trajectories, which defines how the patient interacts

with the different resources for his/her treatment and

tests/scans. The next section focuses on the concep-

tual agent model.

3.2 Conceptual Agent Model

The conceptual agent model must determine a num-

ber of properties of the previous conceptual model.

Focusing mainly on aspects of design and analysis,

the conceptual agent model integrates the major

concepts of agent, role, service and relationship,

defined as:

 The agent is an active entity of the environment;

 The role is played by the concept of an agent;

 A service is a function performed by an agent;

 A relationship is an interaction between entities.

The concepts behind the conceptual agent mod-

el are defined through a meta-model. This meta-

model defines as precisely as possible all the con-

cepts involved in a conceptual agent model and se-

mantic relationships. The conceptual agent meta-

model is formalized by the UML class diagram

shown in Figure below.

Environment

Agent

Reactive

Deliberative

Hybrid

Role

Service

Physician

Patient

Nurse

Informational

Simple

Complex

Active entities

Interaction

Role Description

Behavior

Interactionprotocol

Tas k

Figure 3: Conceptual agent model.

In this conceptual model an agent plays different

roles. The same role can be played by several agents.

A role provides services, while a service may require

a task. Relationships can develop between roles.

There are two sub-types of interactions, simple and

complex interactions (informational). The simple rela-

tionship is an exchange of information to complete

tasks, the distribution of tasks or the sharing of

knowledge and the complex relationship for example

assumes that agents must coordinate their actions in

order to combine their skills to solve complex tasks.

An interaction composed protocol. Finally, there are

several types of agents: reactive (If the simple behav-

iour is required, a type of stimulus-response behav-

iour is sufficient), deliberative (If decision making

and negotiation are needed, it will be the capacities of

a deliberative agent to perceive its environment and

the behaviour of other agents), hybrid (Reactive be-

haviour and deliberative behaviours are needed. For

example, an agent "smart" capable of interacting with

another agent when disruptive events occur). In the

next section we present the Simulation of Care Path-

ways for Patients (SiCaPP).

4 SIMULATION

METHODOLOGY (SICAPP)

The objective of this section is to present the soft-

ware solution restraint to accompany the process

design and Simulation of Care Pathways for Patients

(SiCaPP) for colon and rectal cancer treatment by

integrating the functional and software requirements,

and based on multi-agent modelling.

SiCaPP represents an implementation solution for

the conceptual agent model and is characterized by:

Agent-based Modeling and Simulation Software Architecture for Health Care

 Specification, the agents’ behaviour in appro-

priate languages to the granularity of agents, it

is to describe how the agent should behave dur-

ing the simulation without prejudging how they

will actually be implemented (language pro-

gramming, simulation language, environment,

etc.)

 The specification of interactions between

agents which results in dynamic simulation.

These interactions will have different imple-

mentation issues that are involved as agents of

a same environment.

The simulation environment aims both to facilitate

the handling of models and supervise their imple-

mentation in order to exploit their results. To support

the simulation design process different conceptual

software needs should be treated. These needs can

be summarized into two main categories:

 Needs related to the field study of HC

As mentioned in the previous section, the HC

complex systemimpose many constraints. It in-

volves modelling and simulating the system ac-

cording to their decision-making level and opera-

tional implementation. In fact, the processes na-

ture, calculations or decision-making implement-

ed processes method requires a large variability

of representation behavioural of entities in a HC.

This is translated by expressed models in adapted

modelling languages, which must be possible to

integrate, that is to say make or consistently

maintain at a conceptual and simulation level.

The multi-agent paradigm seems affordable.

 Needs related to simulation

The nature of the simulation of HC, as well as

taking into account the foregoing need, leads to a

distributed simulation load on one or more simu-

lators. Since the simulation word is set, the time

issue becomes unavoidable and is necessary in a

simulator. We need to synchronize the agents in

the simulator to avoid inconsistent behaviour of

the simulation as a whole and therefore the re-

sults of simulations erroneous (Fujimoto, 2000).

This problem is not specific to the simulation of

HC and has already been posed.

5 SICAPP ARCHITECTURE

SiCaPP architecture presents different services,

these services includes the following information:

agents’ management, time management, and inter-

agent communication.

The agents’ management provides all the func-

tions needed to manage the life cycle of agents ad-

dressing, functions such as launch and stop. It allows

for example, adding, changing, or deleting the

agents dynamically, it maintains a directory of these

agents taking particular account of the simulator in

which they operate. Secondly, the inter-agent com-

munication presents different communication lan-

guages like ACL message and provides the commu-

nication between agents in the environment. It can

also manage a directory like yellow pages integrat-

ing information on the capabilities and/ or agent

played roles of the simulation. Finally, the time

management is rarely mentioned in mutli-agent, of

the fact that the distributed nature of the simulation

is often more conceptual than software. Thus, time

management is implicitly centralized on the reactive

multi agent system and is not managed in the delib-

erative systems if not in relative terms.

In this architecture, we also define a different role

that includes the following information:

 A set of actions that can be performed, i.e. a pa-

tient role is to approve action prescribed by

physician.

 A set of protocols, which describe how this role

should interact with other roles.

 A set of goals.

The SiCaPP system is organized into different

packages (Figure 4), packages include:

 User interface: It is a GUI-based logic which

enables the user to generate, simulate patients

and show simulation results graphically.

 Database: It is used to register generated pa-

tients and the simulation results.

 Patient population: It generates patient’s popu-

lation; this information includes aspects of a

patient’s personnel information and physical

health such as treatment plan, medication and

diagnosis.

 Patient simulation: Controlled by physicians

who decide whether diagnostics are to be ac-

cepted, perform medical and surgical interven-

tions, provide prescriptions, and perform

chemotherapy and radiotherapy treatment in

collaboration with nurse.

 Treatments protocols: It describes a method to

be used during the treatment (e.g. drug, medi-

cal treatment) or a medical research study.

 Treatment plan: This package is used to choose

a treatment trajectory plan for patients based

on the epidemiological studies and real data.

SIMULTECH 2016 - 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

UserInterface

PatientPopulation

DataBase

PatientSimulation

TreatmentProtocols

EquationMethods

Gompertz and

Iwata

ClinicalProfil

Graphiccal/Results

Access

TreatmentPlan

Access

Figure 4: SiCaPP Architecture Diagram.

 Equation methods: it contains different sets of

mathematical equations used in our model,

e.g. gompertz model which describes the evo-

lution of the main tumour from the appearance

of the first cancerous cell to a larger tumour,

iwata (Iwata et al., 2000) (Gilli et al., 2014)

model which is used to describe the evolution

of metastases.

 Clinical profile: It is used to check the physical

and psychological state of the patient, it is

based on different notions e.g. depression, sta-

tus performance, sleep disturbance and fa-

tigue. We are using the epidemiological stud-

ies and Jewish Hospital real data.

6 SICAPP KERNEL

Medical information of a patient is one of the most

sensitive types of information; this information in-

cludes aspects of patient’s personnel information

and physical health such as treatments, medicines

and diagnosis. A patient may be treated by any

number of physicians or nurses but they must all

belong in the team which is responsible for this pa-

tient. A physician can treat any number of patients

and maintain the medical history for each patient.

The patient is considered as a composed class to

calculate tumour evolution using mathematical mod-

els. Tumour growth is based essentially on popula-

tion-based models (Verga, 2010). Also this class is

used to verify the stage before and during the treat-

ment (diagnosis step). Each patient has a medical

profile; this profile contains a record of all treat-

ments used within the medical group. If the patient

has been treated in any facility within the same med-

ical group, we will have an existing patient record

and a medical history for the patient; this may need

to be updated. A treatment instance is created for all

patients admitted and updated throughout the pa-

tient’s stay. The treatment will subsequently be add-

ed to the patients’ medical record upon patient dis-

charge.

6.1 Generate Population of Patients

using SiCaPP

Based on epidemiological studies and the real data,

two different methods are used to generate the popu-

lation of patient. Figures below show a state chart

for the class to generate a virtual patient population.

Firstly this class generates the age and gender using

the epidemiological studies (Figure 5) alternatively

we can extract this information from real data

(Figure 6). Secondly, based on the age and gender

we choose the cancer type (two types are available:

colon and rectal) and stage. Alternatively we can

also extract this information from real data. Finally,

we use the gompertz model to determine the tumour

size in mm and we can calculate the stage using the

iwata model. If the stage obtained is different using

both models we have to re-determine the stage again

and repeat the same procedures. In case stage results

are matching, population generated is registered in

the data base.

Star

Generate Age and Gender using

epidemiological studies

Determine

Cancer Type

Determine

Stage

Calculate

Tumor Size

Using Gompertz Model

(Free evolution)

[If stage

obtained is

different the

generated

Stage]

Population registered

in the data base

[If no]

Verify the

stage

Using Iwata

Model

Figure 5: Generate population using the epidemiological

studies.

Agent-based Modeling and Simulation Software Architecture for Health Care

Start

Extract patient

information from

data base

Extract Stage

[If simulated

stage different

the Stage of

patient]

[If no]

Ver i f y t h e

stage

Using Iwata

Model

Access

Calculate

Tumor Size

Using Gompertz Model

(Free evolution)

Population registered

in the data base

Figure 6: Generate population using the real data.

6.2 Select Treatment Plan using

SiCaPP

Diagnosis

Treatment

Start

Extract Patient

Information

Patient

Proposition

Treatment

If patient accept the treatment

If no

Treatment

selected

Select a

Figure 7: Select treatment plan.

Figure 7 shows a state chart for the treatment class

which is responsible for generating a treatment plan

for each patient. This treatment is defined by the

physician. The patient has to perform some diagno-

sis which enables treatment plan choosing. When the

patient approves the treatment, the following infor-

mation must be stored in the generated file to be

used in the simulation step. In case the patient re-

jects the treatment, the physician has to choose an-

other type of treatment in collaboration with the

medical team and patient.

6.3 Treatment Trajectory using

SiCaPP

The figure below shows a state chart of class used to

treat patients who have colon or rectal cancer. This

treatment is created by the physician. First of all, the

patient should be examined prior each treatment or

session of treatments such as radiotherapy or chemo-

therapy. This is needed to evaluate the physical and

psychological state of this patient and determine the

stage of the cancer. After this evaluation the physi-

cian will be able to verify the patient’s ability to

continue treatment or suspend it for some period

until the patients state is re-evaluated or the treat-

ment is adjusted ( for example change the dose of

the medication). During treatment, the patient may

need to undergo more examinations if it is necessary

or if the physician has any concerns.

Exam

Evaluation

and determine stage

Start

Extract Patient and

Treatment Information

Suspend

Discharged

Treatment

Exam required?

Yes

Adjust

treatment

Figure 8: Treatment phase.

7 VA L I D AT I O N

The first objective was to validate our modelling and

SIMULTECH 2016 - 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

simulation oriented agents; the results of the simula-

tion are presented above in figure 8. These simula-

tions should allow us to validate our simulation plat-

form for executing further simulations that involve

treating patients with colon and rectal cancer. The

input data of the simulation and the results are stored

in a database, which was added into our simulation

platform.

To do that, we carried out different experiments,

using the Java eclipse software package with a

3,5GHz Intel Core i7 processor and 32 Go of RAM.

More specific, we used the JADE platform (Java

Agent Development Framework). JADE it’s a MAS

development environment complies with the FIPA

very diffused and included a set of tools included

facilitating various MAS development phases (Ri-

massa et al., 1999). The experiment aims at as-

sessing the ability of the model to replicate the re-

sults of real studies with specific treatment proto-

cols. In order to compare the simulation results with

actual data, we used the results for the real patients

after treatments. The treatments results are classified

by survived or not.

7.1 Experiment and Generate the

Virtual Population

In this experiment, we must calibrate the model’s

parameters. In order to do this ABS, we use the Jew-

ish Hospital real data, which allows us to validate

our model during the different type of treatment. The

real data include 773 patients who have colon and

colorectal cancer. However, among these patients

there are just 56 patients that have a complete pro-

file, more precisely that they have the stage, type of

treatments, and the results after treatments which

characterized by survived or not. Each of these pa-

tients have different types of information (or differ-

ent profile), like stage, age, cancer type, type of

treatment, and the protocol received by patients for

different treatments. Patients in the protocol received

two daily doses of chemotherapy treatment continu-

ously without rest periods.

Figure 9 explains the contents presented in the

Jewish Hospital real data. Firstly we present the

percentage of patients (male and female) who have

Colon and Rectal Cancer in our real data, secondly

we present just the percentage of patients who have

a colon and rectal cancer with complete profile and

output results, and finally we present the percentage

of output results which is classified by Survived or

not.

To start our simulation, we must create several

populations of virtual patients based on the Jewish

Figure 9: Real Data Analyses.

Hospital real data. Firstly the user can be select

one real patient from the data base to generate the

following 100 virtual patients (or more) who have a

similar real patient profiles and same treatment plan.

During the generation of patient profiles we used the

Gompertz model which describes the evolution of

the main tumor from the appearance of the first can-

cerous cell to a larger tumor (we determine the stage

and the tumor size of the population generation).

Then, we use the iwata model to describe the evolu-

tion of metastases. In case the stage obtained and

real patient stages are matching, the population gen-

erated is registered in the data base and we can pre-

pare the simulation step. However, we must generate

the various parameters used by mathematics equa-

tions as Gompertz model and iwata model. The

model was calibrated for one protocol.

Table 1: Real data and Population generation information.

To validate our work, we select five patients with

different profiles to show the effectiveness of our

model. The table below represents the Jewish Hospi-

tal (patients selected) and the population generation.

This table presents the patient Number (P), the stage

(St), the type of cancer (CT: Colon (C) or Colorectal

(Col)), the different treatments (like surgery (Sug),

Radiotherapy (Rad) and Chemotherapy (Ch)) and

Agent-based Modeling and Simulation Software Architecture for Health Care

the Output results (Op) (Survived (S), or Not Sur-

vived (NS)). However, the real data and the popula-

tion generation have the same profile.

7.2 Calibration and Simulation Results

In order to calibrate the model for the configuration

of the real hospital data, we first need to estimate the

impact of each parameter on the results based on

their role in the model. For example, the percentage

of progressive disease is only defined by the pa-

rameter of the Gompertz evolution, the parameters

of the chemotherapy E0, and Absorption and Dose.

There are other parameters such as m, α and the

maximum and minimum size of the tumor in the

selection of the virtual population (Gilli, et al.,

2014). Thus, to calibrate the model, we proceed by

trial-and-error, using a dichotomy approach to set

each parameter and replicate the results of the hospi-

tal data as best as possible. Concerning the duration

of the simulated treatments, the median duration

reported in both studies was used for the correspond-

ing tests. The final values of the parameters for each

calibration are shown in Table 2. Concerning the

parameter Absorption, it has been set equal to its

defined in (Van Cutsem et al., 2000) value, while the

average value of α2 was taken in (Heun et al., 2011).

Table 2: Calibration Parameters.

In figure 10 below we compare our simulation

results with the results presented in the real hospital

data which are classified by survived or not.

Figure 10: Comparative analysis.

8 CONCLUSION AND FUTURE

WORK

The HCs is becoming increasingly complex. In the

search for their performance, modelling/simulation

becomes necessary. This modelling/simulation of

HCs needs multi-modelling with the use of different

formalisms or representation paradigms. The agent’s

oriented approach we showed to be relevant, includ-

ing the consideration of behaviour of various actors of

HC. Our research has focused on the definition of a

modelling approach for agent’s oriented simulation of

HC, with the main objective to allow a more organi-

zational modelling/agents oriented simulation of HC.

For this we have developed a simulation plat-

form for the implementation of the conceptual model

and implementation of multi-agent system. This

platform used a simulation platform based on a spe-

cific simulation environment (JADE). This simula-

tion allowed us to analyse the presented simulation

behaviour in the HC system. We have conducted

with our simulation platform several simulations of

the HC allowing the study of several relevant sce-

narios.

The validation phase described in this paper gives

very important results to reality reproduce, but it is

preliminary. Indeed, validation must be detailed with

more specific data for each patient and have a better

model calibrated than just on population averages,

before integration in the simulation platform. Thus,

validation with a more specific method, reflecting the

better use of the model in the simulation platform, is

required. This requires much more detailed data in the

treatment of each patient, to be provided by the Jew-

ish General Hospital in Montreal.

To complete the simulation platform, it will take

the next step in this focus on the most important part

will be the "Patient health model", because it will

determine the impact of patient treatment side ef-

fects that is an important aspect of treatment against

cancer. Indeed, the fight against cancer advanced by

chemotherapy can be seen as a balance between

enough drugs for reducing cancer, but not too much

to not kill the patient.

REFERENCES

Bernon C., Gleizes M. P., Picard G. and Glize P., 2002.

The Adelfe Methodology for an Intranet System De-

sign. In Proc. of the Fourth International Bi-

Conference Workshop on Agent-Oriented Information

Systems (AOIS), Toronto, Canada.

SIMULTECH 2016 - 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

Chao S. and Wong F., 2009. A Multi-Agent Learning

Paradigm for Medical Data Mining Diagnostic Work-

bench.

Devi M. S. and Mago V., 2005. "Multi-agent model for

Indian rural health care," Leadership in Health Ser-

vices, vol. 18, pp. 1-11.

Ferber J., Stratulat T. and Tranier J., 2009. Towards an

integral approach of organizations in multi-agents sys-

tems: the MASQ approach. Semantics and dynamics of

organizational models in Virginia Dignum.

Foster D., McGregor C. and El-Masri S., 2005. A Survey

of Agent-Based Intelligent Decision Support Systems

to Support Clinical Management and Research. In:

Proceedings of the 2nd International Workshop on

Multi-Agent Systems for Medicine, Computational Bi-

ology, and Bioinformatics, 16–34.

Figueredo G. P. and Aickelin U., 2011. "Comparing sys-

tem dynamics and agent-based simulation for tumour

growth and its interactions with effector cells," in Pro-

ceedings of the 2011 Summer Computer Simulation

Conference, pp. 52-59.

Fujimoto R., 2000. Parallel and distributed simulation

systems. John Wiley & Sons, Inc. USA.

Gaud N., Galland S. and Koukam A., 2008. Towards a

Multilevel Simulation Approach based on Holonic

Multi-agent. Published in the 10th International Con-

ference on Computer Modeling and Simulation

(EUROSIM/ UKSiM‟08), pp. 180–185, England. April

1–3.

Gilli Q., Mustapha K., Frayret J. M., Lahrichi N., and

Karimi E., 2014. Agent-Based Simulation of Colorec-

tal Cancer Care Trajectory: Patient Model. CIRRELT

(Interuniversity Research Centre on Enterprise Net-

works, Logistics and Transportation).

Gupta S., Sarkar A., Pramanik I., and Mukherjee B., 2012.

Implementation Scheme for Online Medical Diagnosis

System Using Multi Agent System with JADE.

Intemational Journal of Scientific and Research Publi-

cations, Volume 2, Issue 6,.ISSN 2250-3153, June.

Gupta S. and Mukhopadhyay S.. 2012 Multi Agent Sys-

tem Based Clinical Diagnosis Systm: An Algorithmic

Approach. International Journal of Engineering Re-

search and Applications (IJERA) ISSN: 2H8-9622.

Vol.2. Issue 5, pp. 1474-1477, September- October.

Gupta S. and Pujari S., 2009. A Multi-Aged: Based

Scheme for Heath Care and Clinical Diagnosis Sys-

tem, IAMA-09. ieeexplore. ISBN: 978-1-4244-4710-7.

July.

Han B-M., Song S.-J., Kyu Min Lee, Jang Kyung-Soo and

Shin Dong-Ryeol, 2006. Multi Agent System based

Efficient Healthcare Service. ICACT 2006, ISBN 89-

551 9-1 29-4.Feb. 20-22.

Heun J. M., Grothey A., Branda M. E., Goldberg R. M.,

and Sargent D. J., "Tumor status at 12 weeks predicts

survival in advanced colorectal cancer: Findings from

NCCTG N9741," The oncologist, vol. 16, pp. 859-867,

2011.

Hübner J. F., Sichman J. S., and Boissier O., 2007. Devel-

oping Organised Multi-Agent Systems Using the Moi-

se+ Model: Programming Issues at the System and

Agent Levels, Int. J. Accounting, Auditing and Per-

formance Evaluation, 1(3/4):370–395.

Iantovics B. 2008. The CMDS Medical Diagnosis System.

Ninth International Symposium on Symbolic and Nu-

meric Algorithms for Scientific Computing, IEEE.

Iwata K., Kawasaki K., and Shigesada N., 2000. "A dy-

namical model for the growth and size distribution of

multiple metastatic tumors," Journal of theoretical bi-

ology, vol. 203, pp. 177-186.

Jennings N. R., Sycara K., and M. Wooldridge, 1998. "A

roadmap of agent research and development," Autono-

mous agents and multi-agent systems, vol. 1, pp. 7-38.

Jones S. S. and Evans R. S., 2008 "An agent based simula-

tion tool for scheduling emergency department physi-

cians," in AMIA Annual Symposium Proceedings, p.

338.

Kanagarajah A., Parker D., and Xu H., 2010. "Health care

supply networks in tightly and loosely coupled struc-

tures: Exploration using agent-based modelling," In-

ternational Journal of Systems Science, vol. 41, pp.

261-270.

Kazar O., Sahnoun Z. and Frecon L.. 2008. Multi-agents

system for medical diagnosis. International Confer-

ence on Intelligent System and Knowledge Engineer-

ing, Vol. 1, Pages1265 – 1270.

Klusch M., Lodi S. and Moro G., 2003. The Role of

Agents in Distributed Data Mining - Issues and Bene-

fits. In: Proceedings of IEEE/WIC International Con-

ference on Intelligent Agent Technology (IAT’03),

211–217.

Knight V. A., Williams J. E., and Reynolds I., 2012.

"Modelling patient choice in healthcare systems: De-

velopment and application of a discrete event simula-

tion with agent-based decision making," Journal of

Simulation, vol. 6, pp. 92-102.

Krizmaric M., Zmauc T., Micetic-Turk D., Stiglic G., and

Kokol P., 2005. "Time allocation simulation model of

clean and dirty pathways in hospital environment," in

Computer-Based Medical Systems, 2005. Proceedings.

18th IEEE Symposium on, pp. 123-127.

Laskowski M., McLeod R. D., Friesen M. R., Podaima B.

W., and Alfa A. S., 2009. "Models of emergency de-

partments for reducing patient waiting times," PLoS

One, vol. 4, p. e6127.

Mahmud R., Aysa Abdul Halim Sithiq H., and Taharim H.

M., 2009. A Hybrid Technology for a Multi agent

Consultation System in Obesity Domain. World Acad-

emy of Science, Engineering and Technology.

Mustafee N., Katsaliaki K., and Taylor S. J. E., 2010.

"Profiling literature in healthcare simulation," Simula-

tion, vol. 86, pp. 543-558.

Nealon J. and Moreno A., 2003. "Agent-based applica-

tions in health care," in Applications of software agent

technology in the health care domain, ed: Springer,

pp. 3-18.

Picard G., 2004. Méthodologie de développement de sys-

tèmes multi-agents adaptatifs et conception de logi-

ciels à fonctionnalité émergente, thèse de doctorat,

Université Paul Sabatier de Toulouse III.

Agent-based Modeling and Simulation Software Architecture for Health Care

Quesnel G., Duboz R., Ramat E., and Traoré M.K., 2007.

VLE: A Multimodeling and Simulation Environment.

Proceedings of the Summer Simulation Multiconfer-

ence (SummerSim‘07), San Diego, California, USA,

July 15-18, pp. 367-374.

Rosa M. V., Cecilia D. Flores, Andre M. S., Louise J. S.,

Ladeira M. and Coelho H., 2003. A multi agent intel-

ligent environment for medical knowledge. Artificial

Intelligence in medicine, 335-366, Elsevier science

B.V.

Rimassa G., Bellifemine F., and Poggi A., JADE - A FIPA

Compliant Agent Framework, PMAA`99, p. 97-108,

Londres, 1999.

Stainsby H., Taboada M., and Luque E., "Towards an

agent-based simulation of hospital emergency depart-

ments," in SCC IEEE International Conference on

Services Computing, September 21, 2009 - September

25, 2009, Bangalore, India, 2009, pp. 536-539.

Verga F., 2010. "Modélisation mathématique de processus

métastatiques," Université de Provence-Aix-Marseille I.

Van Cutsem E., Findlay M., Osterwalder B., Kocha W.,

Dalley D., Pazdur R., et al., "Capecitabine, an oral

fluoropyrimidine carbamate with substantial activity in

advanced colorectal cancer: results of a randomized

phase II study," Journal of Clinical Oncology, vol. 18,

pp. 1337-1345, 2000.

Wooldridge M., Jennings N. and Kinny D., 2000. ‘The

Gaia methodology for agent –oriented analysis and de-

sign’, Autonomous Agents and Multi-Agent Systems,

vol. 3, n° 3, pp. 285-312.

Zambonelli F., Jennings N.R. and Wooldridge M., 2003.

‘Developing Multiagent Systems: The Gaia Method-

ology’, ACM Transactions on Software Engineering

and Methodology, vol. 12, n° 3, pp. 317-370.

Zhang W. and Yao Z., 2010. "A reformed lattice gas mod-

el and its application in the simulation of evacuation in

hospital fire," in IEEE International Conference on

Industrial Engineering and Engineering Management,

IEEM2010, December 7, Macao, China, pp. 1543-

1547.

SIMULTECH 2016 - 6th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

100