The nurse scheduling problem: a combinatorial problem, solved by the combination of constraint programming and real users heuristics.

Nurse scheduling consists of assigning shifts and rest to nurses for each day, on a time schedule, taking in account legal and collective constraints, and individual wishes. The Nurse Scheduling Problem (NSP) is a highly difficult and complicated problem that has already be explored using several operational research methods. Nevertheless, those methods are not flexible enough to match the individual requests from nursing staff and they don't allow for effective management of unforeseen absences. This paper explores the use of a Constraint Programming (CP) to build Gymnaste, which is since June 1997 a available package, after five years of a slow research and development process. We describe the way the problem is modeled, the constraints typology, and the interface. In fact, the NSP is also an ill defined problem, making the man computer interface critical to let the user chose the best heuristics. Gymnaste is currently beta-tested on several pilot sites. Schedules have been generated very quickly, typically in less than a minute for 20-30 persons over 4 weeks. An evaluation process is running, which preliminary results also show some interesting sociological and organizational aspects.