Multi-agent management of Operating Theatres

The management of operations in a large hospital is performed jointly by several groups of people. These include the doctors, the coordinators of the pre-operation units, the managers of the different wards, and the administrative managers of the operating theatres and of the teams of operating rooms. The present study models this problem as a distributed assignment problem within the field of distributed constraints optimization problems (DCOPs) and proposes an innovative representation of constraints and a distributed search algorithm for solving it. Due to the size of the problem, the common approach is to use an incomplete search algorithm. The research reported in the present thesis focuses on the problem of assigning operation requests within a single medical ward in a large public hospital.There is a natural strong hierarchy among the groups that manage the monthly schedule of operations in a ward. The proposed algorithm takes advantage of the hierarchy, in representing the multiple preferences of the different agents. One class of agents performs an initial assignment of operation requests to rooms and days. The assignment incorporates multiple constraints and preferences of patients, of doctors, and of features of specific operations. Another class of agents, that represents the heads of medical wards or senior surgeons, needs to approve the schedule by checking it against a variety of additional medical considerations. This class can request specific changes in order to improve the partial assignment to conform better to its preferences. The distributed nature of the problem is intrinsic and stems from the fact that the medical preferences can only be stated with respect to daily or weekly assignments of operations to operating rooms and to times (computed by the pre-operation unit), due to the huge number of possible such assignments.The interaction among the classes of agents forms an interactive (or procedural) representation of constraints and preferences over assignments. This is a new form of constraints representation. The proposed algorithm enables a distributed multi-objective hill climbing, by using the strict hierarchy among the agents during the exchange of improvements and approvals of assignments.In order to evaluate the resulting assignments of the distributed hill climbing search, real hospital data is used. Tables of operation requests and their resulting assignments over 3 months in the general surgery ward.