基于机器学习的院内心脏骤停预警模型研究进展
Research Progress of In-Hospital Cardiac Ar-rest Early Warning Model Based on Machine Learning
DOI: 10.12677/ACM.2024.141123, PDF, HTML, XML, 下载: 87  浏览: 180  科研立项经费支持
作者: 夏来百提姑·赛买提, 杨建中*:新疆医科大学第一附属医院急救?创伤中心,新疆 乌鲁木齐
关键词: 心脏骤停预警模型机器学习Cardiac Arrest Early Warning Models Machine Learning
摘要: 早期识别心脏骤停(CA)的预警症状和指标对患者的生存起到重要作用,由异常预测因子构成的临床预测模型作为风险的量化工具,为早期识别心脏骤停提供证据,近年来得到普遍应用。基于机器学习的心脏骤停预警模型具有灵活的预测算法,比传统的早期预警评分预测方法更准确、预测效能更高。国内外学者通过各种方法进一步提高了其预测能力,并实现了模型实时预测心脏骤停的功能。本综述复习相关心脏骤停预警模型的发展历程、模型方法和预测性能与总结模型发展中的局限性,探讨基于机器学习的心脏骤停预警模型对预防心脏骤停和心脏骤停后提供决策的研究价值以及对具有高预测能力的预警模型进行展望。
Abstract: Early identification of early warning symptoms and indicators of cardiac arrest (CA) plays an im-portant role in the survival of patients, and the clinical prediction model composed of abnormal predictors is used as a risk quantitative tool to provide evidence for early identification of cardiac arrest, and has been widely used in recent years. The early warning model of cardiac arrest based on machine learning has a flexible prediction algorithm, which is more accurate and more efficient than the traditional early warning score prediction method. Scholars at home and abroad have fur-ther improved their prediction ability through various methods, and realized the function of the model to predict cardiac arrest in real time. This review reviews the development history, model methods and prediction performance of relevant cardiac arrest early warning models, summarizes the limitations of model development, discusses the research value of cardiac arrest early warning models based on machine learning in preventing cardiac arrest and providing decision-making af-ter cardiac arrest, and prospects early warning models with high predictive ability.
文章引用:夏来百提姑·赛买提, 杨建中. 基于机器学习的院内心脏骤停预警模型研究进展[J]. 临床医学进展, 2024, 14(1): 871-876. https://doi.org/10.12677/ACM.2024.141123

1. 引言

心脏骤停(cardiac arrest, CA)是指心脏突然丧失功能。它可能突然出现,也可能出现前有预警症状。如果不立即采取适当的措施,心脏骤停通常是致命的 [1] 。仅在美国,每年院内心脏骤停发生就超过290,000例,生存率0%~36.2%之间 [2] 。研究表明心脏骤停在住院患者中发生率0.1%~0.6%,出院生存率在12%~25%之间,其最近报道的生存率增加 [3] 。我国的情况是不容乐观,每年大约有54万人发生心搏骤停,全国病死率为7.4% [4] 。

在CA发生前数小时甚至数天前就有异常生命体征等预警信号 [5] 。在院内发现患者病情恶化越晚死亡率越高。早期识别CA的预警指标有利于降低死亡率及患者生存中发挥重要作用 [6] [7] 。由异常预测因子构成的早期预警模型作为预测CA的量化工具,为早期识别提供直观及理性证据,对其探索与应用也越来越普遍 [8] ,Richard T. Carrick等 [9] 于JAHA发表的一项研究从1981年7月至2020年2月间发表的81个独特的CA临床预测模型研究报告:31个(38%)模型在推导时纳入的是欧洲人群队列,24个(30%)纳入了北美人群队列,17个(21%)纳入了亚洲人群队列。如今已有100多种预警系统可用于检测和管理患者的临床恶化 [10] ,主要分为基于传统和基于机器学习的预警模型。本文复习认识CA预警模型的发展,归纳各类模型发展的局限性以及展望构建和验证一种实时的、可解释的多变量基于机器学习的心脏骤停早期警预警模型。本文总结往年来文献发表的部分院内心脏骤停(in-hospital cardiac arrest, IHCA)预测模型,见图1

2. 院内CA预警模型的进展

目前用于早期预测院内心脏骤停的主要基于生命体征的模型包括:早期预警评分系统(early warning score,简称EWS) [11] ,是一种临床评估工具,根据这些参数与正常值的偏离程度进行评分,EWS被设计为床旁纸质图表上记录观察轻松计算结果,但无法解释随时间推移的趋势,通常产生假警报、警报疲劳,由此增加错过恶化患者的可能性 [12] 。C.P. Subbe等 [13] 于2001年提出的改良早期预警评分(Modified Early Warning Score,简称MEWS)是一种改良的用于危重患者的评分系统,用于急诊科识别早期危重患者,为急诊科分诊及救治提供决策。国家早期预警评分(national early warning score, NEWS) [14] 是于2012年制定的用于早期识别危重患者的标准化评分系统,NEWS在急诊科疑似脓毒症患者风险预测方面有更高的预测价值、敏感度和特异度 [15] [16] [17] [18] ,但在慢性低氧血症患者NEWS评分的敏感性存在疑问,NEWS评分重要参数中包含氧饱和度,慢性低氧血症患者的基础氧饱和度比正常值偏低,如若对慢性低氧血症患者仍使用NEWS评分来评估病情会增加假阳性率从而导致预测评分的预测能力降低 [19] [20] 。

Figure 1. Overview of the predictive models of cardiac arrest described in this paper

图1. 本文涉及的预测心脏骤停发生模型简介

近年开发了适合于疑似专科疾病中的CA预警评分系统。Jonas Faxén等 [21] 开发了风险评分模型(SAFER)来预测疑似非ST段抬高的急性冠状动脉综合征患者(NSTE-ACS)的院内心脏骤停。陈世浩等 [22] 开发了预测接受紧急血液透析的急诊科患者的院内心脏骤停评分系统,预测紧急血液透析患者院内心脏骤停的发生。上述预警评分可以帮助医疗保健提供者采取必要的预防措施并分配资源。尽管广泛引入了基于生命体征的早期预警评分系统,恶化仍未得到改善。因此通过机器学习算法开发了基于机器学习的早期预警系统(模型),以识别有恶化风险的住院患者 [23] 。

3. 基于机器学习的CA预警模型进展

3.1. 电子健康系统的开发为早期的机器学习预警模型奠定基础

机器学习( machine learning,简称ML)是一门使用计算机作为工具并致力于真实实时的模拟人类学习方式,并将现有内容进行知识结构划分来有效提高学习效率 [24] 。近十年来,它在医疗保健中的应用帮助推动了医生任务的自动化以及临床能力。从模型开发到模型部署,数据发挥着核心作用 [25] 。电子健康系统(Electronic Health Record, EHR)的开发为早期的机器学习预警模型奠定基础、尤其有助于收集庞大的多种变量数据。将实验室结果与生命体征测量相结合可以提高检测恶化患者的精度 [26] 。EHR指的是一个纵向的患者电子医疗信息搜集系统,可以记录患者在所有医疗机构产生的数据,包含患者的多种数据和信息,比如:患者的人口统计资料、病史、用药和过敏史、免疫情况、实验检查结果、影像学检查、生命体征、一般信息、医疗过程记录、支付信息等。EHR系统产生的数据和深度学习,机器学习等大数据领域技术进行融合。通过大数据等先进技术对EHR系统中的各类数据建模并做出分析,为临床诊疗以及个人疾病预防提供支持 [27] 。经不断探究发现机器学习在CA预警方面存在潜力。近期更有证据表明,ML可更准确地预测CA的发生,在特定情况下ML的预测性能优于传统统计学模型 [28] 。

3.2. 初步证实基于机器学习预测CA模型优于传统模型

Joon-Myoung Kwon等 [29] 提出了一种基于机器学习的预警系统,并证实了该系统明显优于修改后的预警模型。于2021年Yeon Joo Lee等 [30] 经多中心验证进再次证明了DEWS的预测性能均优于MEWS,这项研究证明了ML具有高效筛查工具的潜力。同年Oliver C Redfern等 [31] 使用常规采集的血液检测和生命体征预测CA:多变量模型的开发和验证。最终使用C统计量测试了LDTEWS:NEWS风险指数识别有恶化风险的患者的能力(CA),与单独使用NEWS相比,LDTEWS:NEWS风险指数提高了识别有恶化风险的患者的能力。尽管广泛引入了预警评分(EWS)系统和电子健康记录,但恶化仍未得到改善。

3.3. 基于机器学习CA预警模型初步实现实时预测CA

现基于机器学习CA预警模型经改变其参数或改变预算方法均能大大提高了预测能力,并它通过持续评估患者的生命体征、实验室检和病史的不断更新能够实时预测CA [23] 。Minsu Chae等 [32] 经对患者的生物信号、实验室数据和时间序列数据分析后,在EWS中原参数的基础上新加入的预测因子有效提升模型阳性预测值和敏感性。并通过机器学习实现了预算方法不同的模型之间进行了性能比较。

4. 基于机器学习的CA预警模型的预测性能比较分析

4.1. 基于机器学习的预测算法提升CA预警模型的预测性能

基于算法的预测比传统的预测系统取得了更好的性能 [33] [34] 。Churpek等人的研究 [35] 表明灵活的ML (即随机森林)算法(ROC曲线下面积,Area under the Curve of ROC, AUC为0.80)比传统的预测方法MEWS (AUC 0.70)更准确地预测临床恶化。这点由Marco A F Pimentel等 [11] 研究证实,并进一步证明基于机器学习模型HAVEN在24小时内预测能力最强。虽然该模型不适合急诊科,但是它的较高的预测能力和准确的预测方法给我们提供参考和曙光。通常,机器学习使用准确性作为分类的性能评估。在医疗保健领域,机器学习考虑了阳性预测值(Positive Predictive Value, PPV)和敏感性。Minsu Chae等 [32] 通过不同预算方法评估了心脏骤停预测模型性能。如决策树、随机森林、逻辑回归、LSTM模型、GRU模型和LSTM-GRU混合模型。然而,一些深度学习模型过度拟合。考虑了每种算法的最大PPV进行比较。证实决策树具有低PPV和灵敏度。随机森林的PPV最高,但敏感性较低。逻辑回归的PPV较低,但在浅层机器学习中具有最高的灵敏度。深度学习模型与逻辑回归的灵敏度相似,PPV高于逻辑回归。

4.2. 新的预测因子加入可能改善基于机器学习的CA预警模型性能

预测因子是临床预测模型的基础组成部分,探索新的预测因子有助于改善模型的预测准确度 [36] 。Minsu Chae等 [32] 人通过在EWS中参数的基础上加入新的预测因子从而在使用相同的预测方法的(LSTM)情况下升高了相应的PPV或灵敏度而证实基于深度学习的早期心脏骤停预测模型具有较高的PPV (如使用年龄,SBP,72小时内的最大SBP,72小时内的最小SBP,体温,每分钟呼吸数,血压、白蛋白、胆红素、肌酐、PLT、Hb、WBC、ALT等确认基于患者信息和生物信号数据的每个实验室数据使PPV或灵敏度增加了)。还经各类机器学习模型之间进行了比较证实了通过改变预算方法有效提升模型阳性预测值和敏感性。

5. 总结与展望

近20年来,国内外学者对CA预警模型进行了不断探索,并得到了简单易行的早期预警评分广泛使用至今,该模型确实降低了CA的发生率,但它存在的局限性(预测因子仅参考生命体征和评分系统需要医生经验丰富实时关注等)引起了PPA和敏感度降低,导致CA容易漏诊,CA的预防未得到根本改善。现许多外国学者借助了AI技术(机器学习等)的快速处理庞大复杂的数据、完全准确的运算和实时预测的优点筛选到了预测能力很高的新预测因子组合并不断探索多个变量实时快速反应的机器学习模型。如今CA发生率仍然高,已经成为重大社会公共健康问题。而用于临床的新型基于机器学习的CA预警模型很少。因此研发具有高预测能力的基于机器学习的CA预警模型并进行前瞻性多中心验证,便于临床应用,从而有效预防CA的发生或为医生提供决策。同时探究通过机器学习是否能预测心脏骤停的预测时间等,进而为实现健康中国2030重大战略目标而努力奋斗。

基金项目

科技援疆计划项目,项目编号:2022E02046;研究生创新创业项目,项目编号:CXCY2022009。

NOTES

*通讯作者。

参考文献

[1] Andersen, L.W., Holmberg, M.J., Berg, K.M., et al. (2019) In-Hospital Cardiac Arrest: A Review. JAMA: The Journal of the American Medical Association, 321, 1200-1210.
https://doi.org/10.1001/jama.2019.1696
[2] Alamgir, A., Mousa, O. and Shah, Z. (2021) Artificial Intelligence in Predicting Cardiac Arrest: A Scoping Review. JMIR Medical Informatics, 9, e30798.
https://doi.org/10.2196/preprints.30798
[3] Fernando, S.M., Tran, A., Cheng, W., et al. (2019) Pre-Arrest and Intra-Arrest Prognostic Factors Associated with Survival after In-Hospital Cardiac Arrest: Sys-tematic Review and Meta-Analysis. BMJ, 367, l6373.
https://doi.org/10.1136/bmj.l6373
[4] 吕智康, 程兆云, 孙俊杰, 等. 心搏骤停早期预警评分系统的研究现状和展望[J]. 中华危重病急救医学, 2022, 34(4): 440-443.
[5] Augutis, W., Flenady, T., Le Lagadec, D., et al. (2023) How Do Nurses Use Early Warning System Vital Signs Observation Charts in Rural, Remote and Regional Health Care Facilities: A Scoping Review. Australian Journal of Rural Health, 31, 385-394.
https://doi.org/10.1111/ajr.12971
[6] Li, Y.J., Ye, W.Y., Yang, K., et al. (2021) Prediction of Cardiac Arrest in Critically Ill Patients Based on Bedside Vital Signs Monitoring. Computer Methods and Programs in Biomedicine, 214, Article ID: 106568.
https://doi.org/10.1016/j.cmpb.2021.106568
[7] Zheng, K., Bai, Y., Zhai, Q.R., et al. (2022) Correlation between the Warning Symptoms and Prognosis of Cardiac Arrest. World Journal of Clinical Cases, 10, 7738-7748.
https://doi.org/10.12998/wjcc.v10.i22.7738
[8] Nishijima, I., Oyadomari, S., Maedomari, S., et al. (2016) Use of a Modified Early Warning Score System to Reduce the Rate of In-Hospital Cardiac Arrest. Intensive Care, 4, Article No. 12.
https://doi.org/10.1186/s40560-016-0134-7
[9] Carrick, R.T., Park, J.G., Mcginnes, H.L., et al. (2020) Clini-cal Predictive Models of Sudden Cardiac Arrest: A Survey of the Current Science and Analysis of Model Performances. Journal of the American Heart Association, 9, e017625.
https://doi.org/10.1161/JAHA.119.017625
[10] Chae, M., Han, S., Gil, H., Cho, N. and Lee, H. (2021) Prediction of In-Hospital Cardiac Arrest Using Shallow and Deep Learning. Diagnostics, 11, Article No. 1255.
https://doi.org/10.3390/diagnostics11071255
[11] Morgan, R., Williams, F. and Wright, M. (1997) An Early Warning Score for the Early Detection of Patients with Impending Illness. Clinical Intensive Care, 8, 100.
[12] Bedoya, A.D., Clement, M.E., Phelan, M., et al. (2019) Minimal Impact of Implemented Early Warning Score and Best Practice Alert for Patient Deterioration. Critical Care Medicine, 47, 49-55.
https://doi.org/10.1097/CCM.0000000000003439
[13] Subbe, C.P., et al. (2001) Validation of a Modified Early Warning Score in Medical Admissions. QJM: An International Journal of Medicine, 94, 521-526.
https://doi.org/10.1093/qjmed/94.10.521
[14] William, B., Alberti, G., Ball, C., et al. (2012) National Early Warm-ing Score (NEWS): Standardising the Assessment of Acute-Illness Severity in the NHs. Royal College of Physicians, London.
[15] Meylan, S. (2022) National Early Warning Score (NEWS) Outperforms Quick Sepsis-Related Organ Failure (qSOFA) Score for Early Detection of Sepsis in the Emergency Department. Antibiotics, 11, Article No. 1518.
https://doi.org/10.3390/antibiotics11111518
[16] Churpek, M.M., Snyder, A., Han, X., et al. (2017) Quick Sep-sis-Related Organ Failure Assessment, Systemic Inflammatory Response Syndrome, and Early Warning Scores for De-tecting Clinical Deterioration in Infected Patients outside the Intensive Care Unit. American Journal of Respiratory & Critical Care Medicine, 195, 906.
https://doi.org/10.1164/rccm.201604-0854OC
[17] Brink, A., Alsma, J., Verdonschot, R., et al. (2019) Predicting Mortality in Patients with Suspected Sepsis at the Emergency Department; A Retrospective Cohort Study Comparing qSOFA, SIRS and National Early Warning Score. PLOS ONE, 14, e0211133.
https://doi.org/10.1371/journal.pone.0211133
[18] Kostakis, I., Smith, G.B., Prytherch, D., et al. (2021) The Per-formance of the National Early Warning Score and National Early Warning Score 2 in Hospitalised Patients Infected by the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Resuscitation, 159, 150-157.
https://doi.org/10.1016/j.resuscitation.2020.10.039
[19] Hodgson, L.E., Bax, S., Montefort, M., et al. (2013) S67 COPD—In the NEWS! Thorax, 68, A36-A37.
https://doi.org/10.1136/thoraxjnl-2013-204457.74
[20] Lobo, R., Lynch, K. and Casserly, L.F. (2015) Cross-Sectional Audit on the Relevance of Elevated National Early Warning Score in Medical Patients at a Model 2 Hos-pital in Ireland. Irish Journal of Medical Science (1971-), 184, 893-898.
https://doi.org/10.1007/s11845-014-1216-y
[21] Faxén, J., Hall, M., Gale, C.P., et al. (2017) A User-Friendly Risk-Score for Predicting In-Hospital Cardiac Arrest among Patients Admitted with Suspected Non ST-Elevation Acute Coronary Syndrome—The SAFER-Score. Resuscitation, 121, 41-48.
https://doi.org/10.1016/j.resuscitation.2017.10.004
[22] Chen, S.H., Cheng, Y.Y. and Lin, C.H. (2021) An Early Predictive Scoring Model for In-Hospital Cardiac Arrest of Emergent Hemodialysis Patients. Journal of Clinical Medi-cine, 10, Article No. 3241.
https://doi.org/10.3390/jcm10153241
[23] Pimentel, M.A., Redfern, O.C., Malycha, J., et al. (2021) Detecting De-teriorating Patients in the Hospital: Development and Validation of a Novel Scoring System. American Journal of Res-piratory and Critical Care Medicine, 204, 44-52.
https://doi.org/10.1164/rccm.202007-2700OC
[24] 程燕, 王磊, 赵晓永. 根因分析研究综述[J]. 计算机应用研究, 2023, 40(4): 961-966.
[25] Zhang, A., Xing, L., Zou, J., et al. (2022) Shifting Machine Learning for Healthcare from Development to Deployment and from Models to Data. Nature Biomedical Engineering, 6, 1330-1345.
https://doi.org/10.1038/s41551-022-00898-y
[26] Kang, M.A., Churpek, M.M., et al. (2016) Real-Time Risk Prediction on the Wards: A Feasibility Study. Critical Care Medicine, 44, 1468-1473.
https://doi.org/10.1097/CCM.0000000000001716
[27] Tang, Q.H., Cen, X.X. and Pan, C.Q. (2022) Explainable and Efficient Deep Early Warning System for Cardiac Arrest Prediction from Electronic Health Records. Mathematical Biosciences and Engineering, 19, 9825-9841.
https://doi.org/10.3934/mbe.2022457
[28] 吴秋硕, 陆宗庆, 刘瑜, 等. 机器学习应用于心脏骤停早期预测模型的系统评价[J]. 中国循证医学杂志, 2021, 21(8): 942-952.
[29] Kwon, J.-M., Youngnam, L., Yeha, L., et al. (2018) An Algorithm Based on Deep Learning for Predicting In-Hospital Cardiac Arrest. Journal of the American Heart Association, 7, e008678.
https://doi.org/10.1161/JAHA.118.008678
[30] Lee, Y.J., Cho, K.J., Kwon, O., et al. (2021) A Multicentre Validation Study of the Deep Learning-Based Early Warning Score for Predicting In-Hospital Car-diac Arrest in Patients Admitted to General Wards. Resuscitation, 163, 78-85.
https://doi.org/10.1016/j.resuscitation.2021.04.013
[31] Redfern, O.C., et al. (2018) Predicting In-Hospital Mortal-ity and Unanticipated Admissions to the Intensive Care Unit Using Routinely Collected Blood Tests and Vital Signs: Development and Validation of a Multivariable Model. Resuscitation, 133, 75-81.
https://doi.org/10.1016/j.resuscitation.2018.09.021
[32] Chae, M., Gil, H.-W., Cho, N.-J. and Lee, H. (2022) Ma-chine Learning-Based Cardiac Arrest Prediction for Early Warning System. Mathematics, 10, Article No. 2049.
https://doi.org/10.3390/math10122049
[33] Pirracchio, R., Petersen, M.L., Carone, M., et al. (2015) Mortality Pre-diction in Intensive Care Units with the Super ICU Learner Algorithm (SICULA): A Population-Based Study. The Lan-cet Respiratory Medicine, 3, 42-52.
https://doi.org/10.1016/S2213-2600(14)70239-5
[34] Li, L., Ding, L., Zhang, Z., et al. (2023) Development and Validation of Machine Learning-Based Models to Predict In-Hospital Mortality in Life-Threatening Ventricular Arrhyth-mias: Retrospective Cohort Study. Journal of Medical Internet Research, 25, e47664.
https://doi.org/10.2196/47664
[35] Churpek, M.M., Yuen, T.C., Winslow, C., et al. (2016) Multicenter Compari-son of Machine Learning Methods and Conventional Regression for Predicting Clinical Deterioration on the Wards. Critical Care Medicine, 44, 368-374.
https://doi.org/10.1097/CCM.0000000000001571
[36] 文玲子, 王俊峰, 谷鸿秋. 临床预测模型: 新预测因子的预测增量值[J]. 中国循证心血管医学杂志, 2020, 12(6): 655-659.
https://doi.org/10.3969/j.issn.1674-4055.2020.06.04