ObjectivesTo analyze the utilization of intensive care unit (ICU) days in a Canadian medical-surgical ICU and to identify ICU patients with prolonged ICU length of stay (LOS).DesignProspective descriptive study.SettingA Canadian tertiary care medical-surgical ICU.PatientsConsecutive patients admitted to an adult medical-surgical ICU. Neurosurgical, cardiac surgical, and coronary care unit patients were excluded.Measurementsor=to14 days. Among the four LOS groups, the number of ICU days and observed and predicted death rates were compared. Admissions were also stratified by risk of death into five probability range quintiles. Among the five risk groups, ICU LOS was compared between survivors and nonsurvivors.Results0.8 (predicted to die) or <0.2 (predicted to live) whose outcomes were opposite to that predicted had twice the ICU LOS compared with patients whose outcomes were consistent with prediction.Conclusionor=to14 days accounted for 7.3% of total admissions but consumed 43.5% of total ICU days. Identification of patients with prolonged ICU LOS who would ultimately die in the ICU may lead to earlier withdrawal of therapy in these patients, resulting in a substantial reduction in suffering and cost savings. In our study population, outcome prediction using the APACHE II Equation didnot provide sufficient power to accurately discriminate between nonsurvivors and survivors. (Crit Care Med 1999; 27:1319-1324)