Aims and Objectives Patients who attend the Emergency Department (ED) often present with undifferentiated pathology. Patients with acute illness can deteriorate resulting in mortality, Intensive Care Unit (ICU) admission, ward admission or unscheduled return to ED due to untreated pathology. The National Early Warning Score (NEWS) is an established scoring system and has been extensively validated in the inpatient setting. This study aims to investigate the utility of NEWS in predicting short-term morbidity and mortality outcomes in patients presenting to the ED.

Method and Design Emergency attendances in 2019 from three East London EDs were included in this study. Routinely collected structured data were extracted from electronic health records. Patients were stratified into low (NEWS 0-4), medium (5-6) or high (7 or more) NEWS categories based on initial physiological parameters. A data model with NEWS, age and sex was derived from a dataset of patients attending one of three EDs across East London. Logistic regression analysis was performed to determine the primary outcome of 24-hour mortality and secondary outcomes of 48-hour mortality, inpatient admission, ICU admission or 7-day reattendance to the ED. A binary classification model was developed using a training cohort with logistic regression and evaluated against a test set to derive areas under the receiver operating characteristic (AUROC) curves.

Results and Conclusion 256,701 patients were included finally, consisting of 247,842 (96.55%), 5,847 (2.28%) and 3,012 (1.17%) in low, medium, and high NEWS categories. The primary outcome of mortality within 24-hours was comprised of 205 (0.08%) patients. The logistic regression model performance achieved AUROC curves between 0.65 - 0.95 (24-hour mortality 0.95, 48-hour mortality 0.94, ICU admission 0.82, inpatient admission 0.68, 7-day reattendance 0.65). Within this study of an undifferentiated population of emergency care attendances across East London, increasing NEWS was associated with increased mortality, admission and reattendance. Of the outcomes investigated, prediction of 24-hour mortality achieved the highest performance.