Analysis of the expenses linked to hospital stays: how to detect outliers.

When hospital financing depends on a budget which in turn depends on the pathologies being treated, it is necessary to detect hospital stays which show discrepancies between the resources they consume and the medical characteristics they present. Deterministic nonparametric frontier models are used to rank hospital stays according to their expenses taking into account the severity of the patients' conditions. As these models are very sensitive to the extreme stays, a robust frontier model, the order-m frontier is used. The too-efficient stays are highlighted and described. The mean expenses are estimated after excluding too-efficient and inefficient stays. This mean is higher than the mean estimated by using classical trimming rules.