Power and limitations of daily prognostications of death in the medical intensive care unit

Objective: We tested the accuracy of predictions of impending death for medical intensive care unit patients, offered daily by their professional medical caretakers. Design: For 560 medical intensive care unit patients, on each medical intensive care unit day, we asked their attending physicians, fellows, residents, and registered nurses one question: Do you think this patient will die in the hospital or survive to be discharged? Results: We obtained > 6,000 predictions on 2018 medical intensive care unit patient days. Seventy-five percent of MICU patients who stayed >= 4 days had discordant predictions; that is, at least one caretaker predicted survival, whereas others predicted death before discharge. Only 107 of 206 (52%) patients with a prediction of death before discharge actually died in hospital. This number rose to 66% (96 of 145) for patients with 1 day of corroborated (i.e., > 1) prediction of death, and to 84% (79 of 94) with at least 1 unanimous day of predictions of death. However, although positive predictive value rose with increasingly stringent prediction criteria, sensitivity fell so that the area under the receiver-operator characteristic curve did not differ for single, corroborated, or unanimous predictions of death. Subsets of older (> 65 yrs) and ventilated medical intensive care unit patients revealed parallel findings. Conclusions: 1) Roughly half of all medical intensive care unit patients predicted to die in hospital survived to discharge nonetheless. 2) More highly corroborated predictions had better predictive value; although, approximately 15% of patients survived unexpectedly, even when predicted to die by all medical caretakers. (Crit Care Med 2011; 39: 474-479)