Predicting ICU admission and death in the Emergency Department: A comparison of six early warning scores

Aim: To compare the ability of the most used Early Warning Scores (EWS) to identify adult patients at risk of poor outcomes in the emergency department (ED). Methods: Single-center, retrospective observational study. We evaluated the digital records of consecutive ED admissions in patients ≥18 years from 2010 to 2019 and calculated NEWS, NEWS2, MEWS, RAPS, REMS, and SEWS based on parameters measured on ED arrival. We assessed the discrimination and calibration performance of each EWS in predicting death/ICU admission within 24 hours using ROC analysis and visual calibration. We also measured the relative weight of clinical and physiological derangements that identified patients missed by EWS risk stratification using neural network analysis. Results: Among 225,369 patients assessed in the ED during the study period, 1941 (0.9%) were admitted to ICU or died within 24 hours. NEWS was the most accurate predictor (area under the receiver operating characteristic [AUROC] curve 0.904 [95% CI 0.805-0.913]), followed by NEWS2 (AUROC 0.901). NEWS was also well calibrated. In patients judged at low risk (NEWS<2), 359 events occurred (18.5% of the total). Neural network analysis revealed that age, systolic BP, and temperature had the highest relative weight for these NEWS-unpredicted events. Conclusions: NEWS is the most accurate EWS for predicting the risk of death/ICU admission within 24h from ED arrival. The score also had a fair calibration with few events occurring in patients classified at low risk. Neural network analysis suggests the need for further improvements by focusing on the prompt diagnosis of sepsis and the development of practical tools for the measurement of the respiratory rate.