# A New Early Warning System for Predicting Adverse Outcomes in Hospitalized Patients With Neurological Conditions: A Retrospective Cohort Study

## TL;DR

This study develops a new early warning system for hospitalized neurological patients to better predict adverse outcomes like death or urgent transfers.

## Contribution

The paper introduces NEURO-EWS, a novel early warning score specifically tailored for neurological and neurosurgical patients.

## Key findings

- Standard EWS models showed lower predictive accuracy in neurological patients compared to general populations.
- The new NEURO-EWS model achieved an AUC of 0.782, outperforming existing models in predicting adverse outcomes in neurological patients.

## Abstract

Introduction

Early Warning Scores (EWS) are widely used to identify acute clinical deterioration in hospitalized patients, particularly when integrated with rapid response teams (RRTs) as part of hospital safety systems. Although EWS have demonstrated benefits in reducing mortality and facilitating timely intensive care transfers, existing models have been primarily validated in general populations. Neurological and neurosurgical patients, despite being a significant subgroup of emergency admissions, are often underrepresented in these validation studies. Their unique clinical presentations - often subtle or atypical - pose additional challenges for early detection. This study aims to assess the predictive accuracy of various EWS for in-hospital mortality and urgent transfers among patients with neurological and neurosurgical diagnoses, and to develop a new, neurological-specific score (NEURO-EWS) tailored to this population.

Methods

This retrospective cohort study analyzed clinical records of adult patients with neurological conditions, divided into neurological and neurosurgical subgroups. The primary outcome was a composite of in-hospital mortality or urgent transfer to critical care; secondary outcomes analyzed these components separately. The predictive performance of EWS was evaluated at multiple time points using the area under the receiver operating characteristic curve (AUC), and calibration was assessed through probability plots. A novel scoring model, NEURO-EWS, was developed using multivariate logistic regression and validated in a separate cohort.

Results

Standard EWS showed lower predictive performance than expected. The modified Rapid Emergency Medicine Score (mREMS) achieved the highest accuracy among existing models (AUC: 0.68 in the overall cohort, 0.754 in neurological, and 0.646 in neurosurgical patients). The newly developed NEURO-EWS demonstrated superior performance in neurological patients (AUC: 0.782), outperforming all other scores.

Conclusion

Disease-specific EWS improves the detection of clinical deterioration in neurological patients. While mREMS remains the most reliable general model, NEURO-EWS provides enhanced predictive value for neurological populations and may support better clinical decision-making.