Hospital Staff Planning with Multi-Agent Goals

This thesis deals with creating software tools that can help hospital emergency room staff with making decisions that benefit the department as a whole. An emergency room is a complex organization with dynamic flexible processes and many autonomous actors involved. With a holistic simulation model one can predict potential bottlenecks across the department in various what-if scenarios. The complexity of the organization though makes it is difficult to both create and maintain a centralized model. Agent-based simulation is a simulation method that is based on having distributed agents with simple behavior models but when combined exhibit advanced behavior. Agent-based simulation has been used successfully to help improving local aspects of emergency room resource usage, in particular dimensioning of rooms, but it is still difficult to apply agentbased simulation for the whole department. To get a holistic model it could be beneficial to support both the distributed aspects of emergency rooms with decisions being made by many actors, and the centralized aspects of the overall procedures that the actors follow.
In this thesis I investigate achieving this by extending work in Artificial Intelligence on social agents. I investigate their use for simulation and apply it to the emergency room domain. I show how the emergency room procedures and roles can be represented using agent organizations, and how the different values and habits of staff members can be represented with social practices. I conduct a survey of use of formal frameworks and meta-models for programming multiagent systems in agent-based simulation, and implement a formal organizational meta-model for multi-agent systems in an agent-baseds imulation platform. I develop a simple model of an emergency room in the simulation platform by hand but also investigate how event logs collected from a real hospital can be used to generate organizational constraints for use in agent-based simulation. Finally I present perspectives on using agent organizations and multi-agent systems for testing distributed software and designing context-aware systems.