ObjectiveScoring systems that predict mortality do not necessarily predict prolonged length of stay or costs in the intensive care unit (ICU). Knowledge of characteristics predicting prolonged ICU stay would be helpful, particularly if some factors could be modified. Such factors might include process of care, including active involvement of full-time ICU physicians and length of hospital stay before ICU admission.DesignDemographic data, clinical diagnosis at ICU admission, Simplified Acute Physiology Score, and organizational characteristics were examined by logistic regression for their effect on ICU and hospital length of stay and weighted hospital days (WHD), a proxy for high cost of care.SettingA total of 34 ICUs at 27 hospitals participating in Project IMPACT during 1998.PatientsA total of 10,900 critically ill medical, surgical, and trauma patients qualifying for Simplified Acute Physiology Score assessment.InterventionsNone.ResultsOverall, 9.8% of patients had excess WHD, but the percentage varied by diagnosis. Factors predicting high WHD include Simplified Acute Physiology Score survival probability, age of 40 to 80 yrs, presence of infection or mechanical ventilation 24 hrs after admission, male sex, emergency surgery, trauma, presence of critical care fellows, and prolonged pre-ICU hospital stay. Mechanical ventilation at 24 hrs predicts high WHD across diagnostic categories, with a relative risk of between 2.4 and 12.9. Factors protecting against high WHD include do-not-resuscitate order at admission, presence of coma 24 hrs after admission, and active involvement of full-time ICU physicians.ConclusionsPatients with high WHD, and thus high costs, can be identified early. Severity of illness only partially explains high WHD. Age is less important as a predictor of high WHD than presence of infection or ventilator dependency at 24 hrs. Both long ward stays before ICU admission and lack of full-time ICU physician involvement in care increase the probability of long ICU stays. These latter two factors are potentially modifiable and deserve prospective study.