机器学习模型与国家早期预警评分用于预测患者病情恶化风险的表现 - 单中心急诊入院研究

Aug 6, 2026

Increasing operational pressures in emergency departments require fast, accurate identification of patients who need urgent clinical action. With electronic health records enabling large, feature-rich datasets, modern machine learning can be applied to deterioration risk prediction. This single-site retrospective study assesses transformer-based and tree-based models using structured data and free-text triage notes, and compares them with the National Early Warning Score (NEWS). Results show superior precision for models incorporating free-text triage notes and lower alert rates while still capturing most high-risk patients missed by NEWS.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

摘要

研究目的

研究设计

研究场景与方法

研究结果

结论与要点

引言

This article has been accepted for publication in BMJHCI following peer review. The definitive copyedited, typeset version is available online at 10.1136/bmjhci-2024-101088 The Performance of Machine Learning Versus the National Early Warning Score for Predicting Patient Deterioration Risk: A Single-Site Study of Emergency Admissions Matthew Watson1, Stelios Boulitsakis Logothetis2, Darren Green3,4, Mark Holland5, Pinkie Chambers6, and Noura Al Moubayed7,8 1 Postdoctoral Research Associate, Department of Computer Science, Durham University, Durham, DH1 3LE, UK 2 PhD Student, Department of Public Health and Primary Care, University of Cambridge, Cambridge, CB2 0SR, UK 3 Professor, Division of Cardiovascular Sciences, University of Manchester, Manchester, M13 9PL, UK 4 Department of Renal Medicine, Northern Care Alliance NHS Foundation Trust, Manchester, M6 8HD, UK 5Associate Professor, School of Clinical and Biomedical Sciences, University of Bolton, Bolton, BL3 5AB, UK 6Clinical Academic, School of Pharmacy, University College London, London, WC1N 1AX, UK 7Associate Professor, Department of Computer Science, Durham University, Durham, DH1 3LE, UK 8 Evergreen Life Ltd, Manchester, M3 5NA, UK ABSTRACT Objectives Increasing operational pressures on emergency departments (ED) make it imperative to quickly and accurately identify patients requiring urgent clinical intervention. The widespread adoption of electronic health records (EHR) makes rich feature patient data sets more readily available. These large data stores lend themselves to use in modern machine learning (ML) models. This paper investigates the use of transformerbased models to identify critical deterioration in unplanned ED admissions, using freetext fields, such as triage notes, and tabular data, including early warning scores (EWS) . Design A retrospective machine learning study. Setting A large ED in a UK university teaching hospital. Methods We extracted rich feature sets of routine clinical data from the EHR and systematically measured the performance of tree-and transformer-based models for predicting patient mortality or admission to critical care within 24 hours of presentation to ED. We compared our proposed models to the National Early Warning Score (NEWS) . Results Models were trained on 174,393 admission records. We found that models including freetext triage notes outperform structured tabular data models, achieving an average precision of 0 .92, compared to 0.75 for treebased models and 0.12 for NEWS. Conclusions Our findings suggests that machine learning models using freetext data have the potential to improve clinical decision making in the ED; our techniques significantly reduce alert rate, whilst detecting most high-risk patients missed by NEWS. Summary Box What is already known on this topic Increasing operational pressures on emergency departments (ED) make it imperative to quickly and accurately identify patients requiring urgent clinical intervention. Current track and trigger systems use relatively small amounts of parameters to identify physiologically unstable patients, but the widespread adoption of electronic health records means that richer patient details are now available. However, the utility of this data, particularly freetext triage note data, for the use of early warning scores was unclear. What this study adds Our study shows that, when used with transformer-based machine learning techniques, the rich patient data collected in electronic health records (including freetext triage notes) can significantly outperform the National Early Warning Score when predicting patient deterioration. Our work highlights the efficacy of machine learning for clinical decision support tools and the currently untapped information contained in freetext triage note data. 1 Introduction Early recognition and intervention of deteriorating patients is vital to prevent avoidable hospital deaths. [1] Track and trigge

FAQs

该研究的核心目的是什么？

评估机器学习模型在预测急诊非计划入院后24小时内病情恶化风险方面的表现，并与国家早期预警评分（NEWS）进行比较。

研究中使用了哪些类型的数据特征？

从电子健康记录中提取结构化数据（含早期预警评分等表格数据），并使用包含分诊记录的自由文本（free-text triage notes）特征。

主要结果显示了什么差异？

包含自由文本分诊记录的模型优于仅结构化数据的模型：平均精确度达到0.92，而树模型约为0.75，NEWS约为0.12。

Aug 6, 2026

Upload by Ava Thompson