Objectives Prognostication is an essential ability to clinicians. Nevertheless, it has been shown to be quite variable in acutely ill patients, potentially leading to inappropriate care. We aimed to assess the accuracy of physician’s prediction of hospital mortality in acutely deteriorating patients referred for urgent intensive care unit (ICU) admission.

Methods Prospective cohort of acutely ill patients referred for urgent ICU admission in an academic, tertiary hospital. Physicians’ prognosis assessments were recorded at ICU referral. Prognosis was assessed as survival without severe disabilities, survival with severe disabilities or no survival. Prognosis was further dichotomised in good prognosis (survival without severe disabilities) or poor prognosis (survival with severe disabilities or no survival) for prediction of hospital mortality.

Results There were 2374 analysed referrals, with 2103 (88.6%) patients with complete data on mortality and physicians’ prognosis. There were 593 (34.4%), 215 (66.4%) and 51 (94.4%) deaths in the groups ascribed a prognosis of survival without disabilities, survival with severe disabilities or no survival, respectively (p<0.001). Sensitivity was 31%, specificity was 91% and the area under the receiver operating characteristic curve was 0.61 for prediction of mortality. After multivariable analysis, severity of illness, performance status and ICU admission were associated with an increased likelihood of incorrect classification, while worse predicted prognosis was associated with a lower chance of incorrect classification.

Conclusions Physician’s prediction was associated with hospital mortality, but overall accuracy was poor, mainly due to low sensitivity to detect risk of poor prognosis.