• No results found

Comparison between Agent-based Simulation and Discrete-

Chapter 2 Modelling Hospital-acquired Infections

3.4 Comparison between Agent-based Simulation and other Modelling

3.4.2 Comparison between Agent-based Simulation and Discrete-

Methodology

potential methods of modelling HAIs (see Section 3.3.2). ABS and DES are similar simulation techniques in many ways. For example, they both model individuals who change states over time. Individuals in both types of models may have attributes and states, and the individual characteristics will affect the state change of that particular entity (in DES) or agent (in ABS). Event-driven time advance mechanisms may be applied to both techniques. Despite the similarities, ABS and DES are two simulation methods which have many different features and these differences may affect their suitability and ability to model certain systems. There has been no systematic study to compare the two simulation techniques in the context of HAI modelling. In this section, the differences between the two simulation methods and the advantages and disadvantages of ABS relative to DES are discussed in this context.

Differences between Agent-based Simulation and Discrete-event Simulation

ABS and DES have different origins and for a long time they have been studied in different disciplines. ABS is traditionally associated with artificial intelligence and complex adaptive systems (North and Macal 2007) while DES is traditionally associated with operational research. It is only until recently that ABS is gradually assimilated by the operational research discipline. From the perspective of DES, many ABS models maintain a future events list and have the capability of event scheduling. From the perspective of ABS, many DES models, especially those involve humans and interactions, look similar to ABS models where agents are not fully developed and have limited behaviour rules. In order to compare the two simulation methods in the context of HAI modelling, the differences between the two techniques are discussed first.

Modelling approach and the role of entity/agent

ABS adopts a bottom-up modelling approach from the agent’s point of view and the agent is the single most important element of the model. DES adopts a top-down modelling approach from the overall system’s point of view and the entity is only one of the many essential elements of the model.

Entity/Agent representation and dynamics

Apart from attributes, agents in ABS can have complex states and adaptive behaviour rules. Agents may change the states reactively or proactively. Agents’ actual

Methodology

development and movement routes may be unpredictable and broadly defined by boundaries or rules. In DES, simple attributes, numerical, text or logical, are attached to entities. Entities normally change their states only passively and pass through the pre-defined system structure.

Traceability and control of entity/agent

In ABS, the agent is the core of the model and normally it is easy to trace a particular agent at any time during the simulation. Consequently, the control of an agent is direct and can be carried out at any time. While in DES, it may not be easy to trace an individual entity once it enters the system. Therefore, the control of an entity is indirect and is normally performed only when an event happens to the entity (e.g., enter or leave a queue/activity).

Events generation

Since the agent is the core of ABS and it is easily traceable and directly controllable, events in ABS can be generated by agents themselves and naturally associated with any specific agent. In DES, events are normally generated by and associated with other model elements (e.g., queues and activities) which constitute the model structure. As a result, it is difficult to associate events with a specific entity.

Handling of a number of concurrent state changes

In ABS, the model is built around agents and the focus of the model is to describe agents’ behaviour rules that change their states. Agents can be effectively and efficiently involved in a number of concurrent or simultaneous state changes. For example, the agent can change their states regarding spatial locations, and at the same time change states regarding the development of the infectious disease. If necessary, the patient agent can also be involved in other state changes that are relevant to the transmissions dynamics, such as the development of the underling illness and the aging process. The handling of a number of concurrent state changes can be easily implemented in ABS.

Methodology

changes in most DES software. Davies et al. (1993) developed a structure which overcomes this problem. The approach allows an entity to engage in different concurrent activities or wait in any number of queues, and activities can be de- scheduled, interrupted or delayed. The advantages of the approach in the health context were described by Davies and Davies (1994) and the method is referred to as POST (patient-oriented simulation technique). The approach has been applied in many healthcare areas (Davies et al. 2000; Cooper et al. 2002) including the transmission of community-based infections (Rauner et al. 2005). POST was coded in Pascal and subsequently in Delphi (Cooper et al. 2008) and is not easy to use. Similar facilities are not available in commercial software.

Spatial and movement representation

ABS models are normally spatially explicit and this is one of the reasons that modellers adopt ABS. It is common for agents to move freely (within boundary and according to rules) and be aware of its own as well as other agents locations. In DES, in order to represent spatial location and movements, a pre-defined model structure and states which represent physical locations are needed. However, the inclusion of the movement activity may prohibit the entity from engaging in other activities in DES (see the previous section on handling concurrent state changes).

Exchange messages

Sending and receiving messages are standard techniques adopted by ABS and agents can act accordingly based on the information received from other agents or from the system environment. In DES, it is normally difficult to exchange messages between specific entities due to the difficulty to trace and directly control individual entities.

Application domain

ABS best suits those systems which consist of autonomous agents who interact with each other and with the environment. Infection transmission among patients and HCWs in the hospital setting is a good example of such a system. DES, on the other hand, is best for modelling queuing systems. The differences between ABS and DES are summarised in Table 3.2.

Methodology

Table 3.2 Differences between agent-based simulation and discrete-event simulation

Agent-based simulation Discrete-event simulation Modelling approach

and the role of entity/agent

Bottom-up modelling approach from the agent’s point of view. Agent is the most important element of the model.

Top-down modelling approach from the system’s point of view. Entity is only one of the many essential elements of the model.

Entity/agent representation and dynamics

Agent can have complex states and behaviour rules and changes its states reactively or proactively. Agents’ actual development and movement routes may be unpredictable and defined by boundaries or rules.

Simple attributes (numerical or text) are attached to

entities. Entity changes its states only passively and passes through a pre-defined system structure.

Traceability and control of entity/agent

It is easy to trace a particular agent at any time during the simulation; Control of agent is direct and can be carried out at any time.

It is difficult to trace individual entity once it enters the system; control of an entity is indirect and is normally performed only when an event happens to the entity. Events generation Events can be generated by agents themselves and therefore

can be naturally associated with the agent who creates the event.

Events are normally generated by and associated with queues or activities, and difficult to be associated with a specific entity.

Handling concurrent state changes

Agents can effectively and efficiently handle a number of concurrent state changes which are embedded within the agent.

Entities can normally only handle one stream of activity which is embedded in the predefined modelling structure. Spatial and

movement representation

It is common for agents to move freely (within boundary and according to rules) and be aware of its own and other agents’ locations.

Entities can move within pre-defined states which

represent physical locations. This may cause problems for other activities.

Exchange messages Agents can exchange messages with each other and with the environment.

Entities can not exchange messages easily. Application domain Systems consist of autonomous agents who interact with

each other and with the environment.

Methodology

Relative Advantages of Agent-based Simulation

Based on the differences identified between ABS and DES, the relative advantages of ABS compared to DES can be summarised in the context of HAI modelling.

Natural choice to represent patients and infection transmissions

In general, compared to DES, ABS is a more natural choice to represent patients and infection transmission dynamics. In HAI models, patients are undoubtedly the most important element of the model. This is in line with the fact that agents in ABS are the core of the model. Furthermore, infection transmission occurs through the interactions among patients and between patients and HCWs; while ABS is deemed as most appropriate to model systems consisting of autonomous agents interacting with each other and with the environment.

Powerful tools to represent patient behaviour rules

Besides simple attributes and states, patients may have complex behaviour rules governing their infection development, movements and other aspects of state changes and activities. Compared to DES, ABS has more powerful tools to represent these behaviour rules, which can be further facilitated by the message exchange capability. For example, a patient’s behaviour may be triggered by the information received from other patients or HCWs.

Powerful tools to represent patient spatial location and movement

One particular advantages of ABS, compared to DES, is its ability to represent patients’ spatial locations and movements. The relative spatial location between susceptible and colonised patients may be critical to the successful transmission of a pathogen in the hospital setting. It is expected that the closer the two patients stay, the more likely contacts and consequently successfully transmissions may occur. ABS simulation has well established methods to represent spatial locations and various rules to govern the movements of agents.

Powerful tools to represent multiple concurrent state changes

Methodology

Relative Disadvantages of Agent-based Simulation

The relative disadvantages of ABS compared DES include:

• Compared to DES which has a wide range of well-developed and user-friendly software packages, there are only a few software packages that support ABS and most of these have a relatively short history and are less user-friendly to develop, debug and implement ABS models. This may be a disadvantage for potential modellers when choosing between ABS and DES.

• Compared to DES, ABS is less effective in representing queuing systems which can be an important aspect of the HAI models when the patient pathway needs to be modelled.

3.4.3 Comparison between Agent-based Simulation and General Individual-