Nonparametric frontier model as a tool for exploratory analysis of hospital stays.

Diagnosis-related groups (DRG) were introduced in 1995 to the Belgian hospital financing system. Trimming rules are generally used when mean length of stay (LOS) is estimated by DRG. This paper proposes the use of frontier models instead of trimming rules. These models allow to take into account the characteristics of the patients, to rank hospital stays, and to indicate stays presenting discrepancy between the patient's characteristics and the resources consumed. The analysis is done with the nonparametric Free Disposal Hull (FDH) model and the method developed by Wilson to detect extreme observations, when defining the frontier is adapted to analyze large databases.