A novel mortality prediction model for the current population in an adult intensive care unit

Background: The accurate and reliable mortality prediction is very useful, in critical care medicine. There are various new variables proposed in the literature that could potentially increase the predictive ability for death in ICU of the new predictive scoring model. Objective: To develop and validate a new intensive care unit (ICU) mortality prediction model, using data that are routinely collected during the first 24 h of ICU admission, and compare its performance to the most widely used conventional scoring systems. Methods: Prospective observational study in a medical/surgical, multidisciplinary ICU, using multivariate logistic regression modeling. The new model was developed using data from a medical record review of 400 adult intensive care unit patients and was validated on a separate sample of 36 patients, to accurately predict mortality in ICU. Results: The new model is simple, flexible and shows improved performance (ROC AUC = 0.85, SMR = 1.25), compared to the conventional scoring models (APACHE II: AUC = 0.76, SMR = 2.50, SAPS III: AUC = 0.76, SMR = 1.50), as well as higher predictive capability regarding ICU mortality (predicted mortality: 41.63 + 31.61, observed mortality: 41.67%). Conclusion: The newly developed model is a quite simple risk-adjusted outcome prediction tool based on 12 routinely collected demographic and clinical variables obtained from the medical record data. It appears to be a reliable predictor of ICU mortality and is proposed for further investigation aiming at its evaluation, validation and applicability to other ICUs. (C) 2017 Elsevier Inc. All rights reserved.