Evaluating the National Early Warning Score (NEWS) in triage: A machine learning perspective

Background

The National Early Warning Score is widely used in Emergency Departments for triage, primarily to predict mortality. However, its effectiveness in assessing additional clinical outcomes relevant to triage, such as patient urgency and severity, remains unclear and warrants further investigation.

Methods

This prospective observational study was conducted from June to December 2023 at the Merano Emergency Department, enrolling 1,270 patients. The National Early Warning Score was categorized into three risk groups: low (0–4), intermediate (5–6), and high (≥7). Outcomes assessed included 30-day mortality, Intensive Care Unit hospitalization, general hospitalization, physician-defined urgency, and physician-defined severity. A decision tree analysis identified the most significant predictors for each outcome. Receiver Operating Characteristic (ROC) curve analysis evaluated the predictive performance of the National Early Warning Score across these outcomes.

Results

Among the enrolled patients, 88.7 % had a National Early Warning Score of 0–4, 6.1 % had 5–6, and 5.2 % had ≥ 7. The National Early Warning Score demonstrated good predictive ability for 30-day mortality (ROC = 0.82) and Intensive Care Unit hospitalization (ROC = 0.72). Decision tree analysis identified the score as the strongest predictor for mortality and Intensive Care Unit admissions. However, for physician-defined urgency and severity, it was less reliable, often misclassifying seriously ill patients into lower-risk categories.

Conclusions

The National Early Warning Score proves valuable in triage for predicting mortality and Intensive Care Unit admissions. However, its limitations in evaluating broader clinical outcomes, as revealed by decision tree analysis, suggests that it should be complemented rather than replace traditional triage assessments.