人工智能在脓毒症诊疗中的应用:创新、挑战与实践

彭嘉辰 ,  徐昉

重庆医科大学学报 ›› 2026, Vol. 51 ›› Issue (06) : 763 -769.

PDF (1032KB)
重庆医科大学学报 ›› 2026, Vol. 51 ›› Issue (06) : 763 -769. DOI: 10.13406/j.cnki.cyxb.004044
卓越医见:脓毒症的发生机制与临床治疗

人工智能在脓毒症诊疗中的应用:创新、挑战与实践

作者信息 +

Application of artificial intelligence in the diagnosis and treatment of sepsis: innovations,challenges,and clinical practice

Author information +
文章历史 +
PDF (1056K)

摘要

脓毒症的救治是阻碍重症医学临床诊疗水平提高的重要难题,关键问题在于其高度异质性导致的个体化治疗策略不足。人工智能(artificial intelligence,AI)的应用为破解这一难题提供了新的方向。本文梳理了脓毒症的定义和诊疗现状,剖析了AI对脓毒症诊疗的创新性推动作用与面临的挑战,探讨了AI赋能脓毒症诊疗的未来方向与发展前景。本文认为,推动临床医学与生物学、数据科学及计算机科学等多领域合作,借助AI工具整合多组学、多维度数据,实施富集策略破解脓毒症的生物学异质性,是开展精准医疗的重要前提,也是满足脓毒症早期诊疗与精准治疗需求的创新途径。

Abstract

The management of sepsis remains a critical challenge hindering the improvement of clinical diagnosis and treatment in critical care medicine,with the core issue of the inadequacy of individualized treatment strategies due to its high heterogeneity. The application of artificial intelligence (AI) provides a new direction for addressing this challenge. This article reviews the definition of sepsis and the current status of sepsis diagnosis and treatment,analyzes the innovative role of AI in promoting sepsis management and related challenges,and explores the future directions and prospects of AI‑empowered sepsis management. This article posits that promoting interdisciplinary collaboration among clinical medicine, biology, data science, and computer science, and leveraging AI tools to integrate multi-omics and multi-dimensional data for implementing enrichment strategies to decipher the biological heterogeneity of sepsis, constitute a crucial prerequisite for advancing precision medicine. It also represents an innovative approach to meet the demands for early diagnosis, treatment, and precise management of sepsis.

Graphical abstract

关键词

脓毒症 / 人工智能 / 异质性 / 精准医学

Key words

sepsis / artificial intelligence / heterogeneity / precision medicine

引用本文

引用格式 ▾
彭嘉辰,徐昉. 人工智能在脓毒症诊疗中的应用:创新、挑战与实践[J]. 重庆医科大学学报, 2026, 51(06): 763-769 DOI:10.13406/j.cnki.cyxb.004044

登录浏览全文

4963

注册一个新账户 忘记密码

重症医学的发展方向,正从对既有治疗方案、临床指南、共识定义及器官支持的单纯依靠,转向更具针对性的精准医疗[1]。脓毒症是一种异质性极强的综合征,其高死亡率的关键原因在于缺乏适合特定患者的个性化治疗[2]。关于脓毒症的早期管理,争议的焦点在于3个缺失:没有支撑早期识别和确定性治疗策略的权威定义、可靠的诊断标志物及能改变病程的确定性治疗方法[3]。重症监护病房(intensive care unit,ICU)汇集了大量复杂的高维医学数据,人工智能(artificial intelligence,AI)能够精确、高效地解析脓毒症的瞬时状态与病理生理机制,促进诊疗模式的转变。本文旨在回顾AI的应用在当前脓毒症诊疗模式下的特征与挑战,探讨利用AI赋能脓毒症诊疗的实践方向与前景。

1 脓毒症异质性与诊疗困境

1.1 脓毒症异质性与预测生物标志物筛选

脓毒症具有高度异质性,已被确认为过去临床试验失败的主要因素之一[4]。对脓毒症进行风险分层与亚型分类并开展精准个体化治疗,已成为应对脓毒症异质性的新方向[5-7]。然而,目前缺乏诊断脓毒症的“金标准”[8],早期识别受现有诊断工具与方法学限制[9],降低临床救治效能[10]。生物标志物被认为是脓毒症风险分层与预后预测的重要工具[11]。脓毒症生物标志物具有多样性,临床常用的C反应蛋白(C-reactive protein,CRP)敏感性高但特异性较低;降钙素原(procalcitonin,PCT)相对更具特异性。研究发现多种新型生物标志物具有预测效能:免疫细胞在感染时分泌高迁移率族蛋白B1(high mobility group box 1 protein,HMGB-1),其表达水平与脓毒症严重程度相关[12];长链非编码核糖核酸(long non-coding ribonucleic acid,lncRNA)和信使核糖核酸(messenger ribonucleic acid,mRNA)等[如:肺腺癌转移相关转录本1(metastasis-associated lung adenocarcinoma transcript 1,lnc-MALAT1)、微小核糖核酸-125a(micro ribonucleic acid-125a,miR-125a)][13]也能作为脓毒症生物标志物;白细胞介素(interleukin,IL)-35[14]、IL-38[15]、IL-27[16-17]等在脓毒症患者血清中的差异化表达为精准诊疗提供了多维度靶点。目前,高通量生物数据表型多用于识别脓毒症生物学亚型[18],用于描绘免疫图谱揭示脓毒症的解剖来源特异性免疫模式[19]。但由于临床上很少依赖单一指标指导治疗决策,因此,预测性生物标志物实际使用效能未能达到预期[20]。为克服这一局限,开发生物标志物组合并与临床体征相结合,更契合脓毒症的复杂性[21]。有研究发现6种生物标志物组合,在细菌性炎症的曲线下面积(area under the curve,AUC)明显大于单一标志物[22]。可见,整合生物标志物构建脓毒症生物标志物知识数据库,将有助于系统解析脓毒症异质性并构建精准医学模型[23]

1.2 脓毒症发展时相性与临床预警体系研发

脓毒症发生发展具有时相性,应通过随时间推移的纵向筛查予以确诊。而多种非感染性炎症(如:严重创伤、胰腺炎、缺血再灌注损伤、血管炎及药物不良反应)会出现与脓毒症相似的症状、体征与实验室结果[9]。研究发现,在初诊为脓毒症的患者中,有超过1/3的患者在后续的治疗中被证实为“非感染性疾病”[24]。可见,早期很难确定“感染”是否作为脓毒症的触发因素。临床上常用序贯器官衰竭评估(sequential organ failure assessment,SOFA)、快速序贯器官衰竭评估(quick sequential organ failure assessment,qSOFA)等工具协助脓毒症诊断。然而,SOFA评分在发生器官功能障碍前无法识别早期感染及其潜在器官功能损伤[9]。qSOFA评分[25]也并不是脓毒症特异性筛查工具。因而,将qSOFA评分和其他评分系统,或将全身炎症反应综合征(systemic inflammatory response syndrome,SIRS)和其他病理生理过程作为脓毒症的诊断依据进行比较的Meta分析,其结果会存在一定的误导性[9]。基于“生理紊乱通常先于可识别的临床恶化出现”的观点,研究者提出使用早期预警评分系统(early warning scoring system,EWS)筛查脓毒症和器官失代偿高风险患者。研究表明,英国国家早期预警评分(National Early Warning Score,NEWS)可在临床轨迹中跟踪患者状态并在预定阈值触发警报,从而提醒医务人员增加监测强度或调整治疗方案[26]。有研究发现疑似感染急诊(或住院)患者NEWS和改良早期预警评分(modified national early warning score,MNEWS)对院内死亡率的区分能力高于qSOFA与SIRS[27]。脓毒症和其他任何导致急性生理紊乱的疾病间必然存在重叠,而各参数异常相互关联[8],现有脓毒症EWS都有不同程度局限性。因此,高效整合疾病临床时相维度,提升生物标志物诊断效能,优化脓毒症预警体系需要依托新的检测、分析技术。AI已显示出具有根据脓毒症独特特征量身定制个性化治疗来优化治疗策略的潜力。

2 AI对脓毒症诊疗的创新性推动作用

2.1 基于电子病历开发的AI模型能大幅提前准确预测脓毒症发生时间

机器学习(machine learning,ML)模型训练数据应用于电子病历(electronic medical record,EMR)开发脓毒症个体化监测算法越来越受欢迎,主要集中在“大数据、数据驱动、疾病重要特征、疾病早期诊断、疾病风险预测”等领域。与单独使用结构化数据相比,联合临床文本进行ML或自然语言处理(natural language processing,NLP)来检测、识别、诊断或预测脓毒症不同阶段的发生发展更早、更准确[28]。Goh KH等[29]开发的脓毒症早期风险预警(sepsis early risk alert,SERA)算法可将脓毒症早期(发生前12 h)检出率提高32%,且降低假阳性率17%。脓毒症预测和优化治疗(sepsis prediction and optimisation of therapy,SPOT)算法自动触发脓毒症筛查比传统方法节省约6 h,死亡率下降9.9%[30]。使用靶向实时早期预警系统(targeted real-time early warning system,TREWS)后,3 h内评估和确认警报的脓毒症患者从“发出警报到首次使用抗生素”中位时间缩短了1.85 h[31]。可见,EWS可以通过ML算法提升脓毒症预测效能。

2.2 脓毒症阶段化治疗在AI辅助中获益

研究表明将微生物学变量指标纳入α、β、γ和δ 4种脓毒症亚表型后,入组比例增加,变量特征改变,更具临床价值[732]。在以病因学为基础的脓毒症分类管理方面,有研究开发并验证了AI能够有效预测严重创伤患者脓毒症发生风险(研究模型开发队列的发病率为43.44%)[33]。有研究总结分析了97个ML系统,85%数据源来自ICU最常见的细菌感染[34]。使用ML能大幅提升金黄色葡萄球菌、大肠杆菌和肺炎克雷伯菌等临床重要病原体抗生素耐药性识别能力与优化治疗[35]。需关注抗菌药物耐药性(antimicrobial resistance,AMR)基因突变日新月异的背景下ML辅助决策的可解释性[3436]。在脓毒症分层管理中,可通过AI优化亚表型分类,并针对不同病原学在风险评估、治疗策略优化等方面发挥作用。

2.3 新型多组学技术与大型数据集结合为探寻脓毒症有效生物标志物提供新途径

研究表明无监督共识聚类和ML分析白细胞全基因组表达谱可解析脓毒症患者4种内表型(Mars1-4)及其候选标志物[37]。基因转录变异的单核苷酸多态性(single nucleotide polymorphism,SNP)和表达数量性状位点(expression quantitative trait loci,eQTL)也可用于脓毒症反应特征(sepsis response signature,SRS)1或SRS2,以评估重症化与早期死亡风险[38]。有研究纳入急性感染诊断[39]、区分细菌或病毒感染[40]和脓毒症30 d死亡风险mRNA数据[41]开发的神经网络分类器用于分类筛查(入院<36 h)急性感染[42]。一些研究应用ML,将生物信息学数据集与脓毒症相关的分子事件进行关联分析,发现中性粒细胞胞外诱捕网(neutrophil extracellular traps, NETs)相关基因[43]、靶向中性粒细胞的关键基因[44]、胞吐作用相关基因[45]、免疫相关内质网应激基因[46]以及铁死亡相关基因[47]等具有作为脓毒症诊疗新型生物标志物的潜力。NOP2/Sun RNA甲基转移酶家族成员7(NOP2/Sun RNA methyltransferase family member 7,NSUN7)、核仁蛋白2(nucleolar protein 2,NOP2)、假尿苷合成酶1(pseudouridine synthase 1,PUS1)、假尿苷合成酶3(pseudouridine synthase 3,PUS3)和脂肪量和肥胖相关蛋白(fat mass and obesity-associated protein,FTO)有望作为脓毒症的临床诊断生物标志物,为从表观转录组学角度解析脓毒症内型分类提供新见解[48]。可见,整合多组学数据构建个体化分层模型已成为提升脓毒症精准诊疗的一个重要组成部分。

2.4 AI辅助强化脓毒症器官功能的精准维护

ML能更好地预测严重脓毒症或脓毒性休克患者液体反应性[49]。埃默里医疗保健系统使用经过验证的ML算法实施意向性脓毒症精准复苏治疗,有望从根本上重新定义静脉液体复苏的国际标准[50]。以重症医学数据库(medical information mart for intensive care,MIMIC)-Ⅲ为背景构建的XGBoost模型能帮助临床医生识别脓毒症相关急性肾损伤(acute kidney injury,AKI)高危患者,并实施早期干预以降低死亡率[51]。采用多变量回归分析与eICU外部验证,则将脓毒症危重症患者AKI分为8类,并按早期肌酐轨迹进行分层管理[52]。Wen CL等[53]也联合MIMIC-Ⅳ、MIMIC-Ⅲ构建了对脓毒症相关肝损伤(sepsis-associated liver injury,SALI)28 d死亡率具有良好预测能力的模型。可见,AI辅助能够识别脓毒症相关器官功能障碍,并通过优化个体化治疗决策减轻器官损伤。

3 AI辅助脓毒症诊疗面临的挑战

3.1 ICU实时数据解析能力与质量挑战

ICU每天都在产生海量的数据,AI辅助脓毒症诊疗在数据整合与标准化、模型性能与临床适用性、临床接受度与决策协同、伦理与安全性等方面面临挑战。脓毒症早期预测AI模型(如:DeepAise)就易受临床数据集中缺失的影响。有研究对28项临床医生使用AI管理脓毒症的研究进行了评价,认为早期脓毒症预测模型性能因临床环境而异,在大量实时临床数据背景中可能获得最佳结果[54]。且脓毒症预测准确性在ICU(AUC=0.68~0.99)、院内(AUC=0.96~0.98)、急诊科(AUC=0.87~0.97)等不同临床场景之间也存在差异[55]。但鉴于脓毒症异质性导致样本极度不均衡,尤其是ICU内数据更加复杂多样,AUC可能因负样本过多而虚高,掩盖模型对正样本的识别能力。此外,即使在同一ICU内,不同时间段的数据也可因医生治疗行为或ICU环境因素变化产生分布偏移,导致AUC波动。例如:纳入脓毒症发病后3 h内的Epic脓毒症预测模型(Epic sepsis model,ESM)时,住院患者的AUC可提高至0.80(95%CI=0.79~0.81)[56]

3.2 模型可解释性与临床实践挑战

敏感性和特异性间的不平衡影响AI辅助脓毒症诊疗效能。Philadelphia三级教学医院的一种基于ML算法的脓毒症预测模型特异性为98%,但敏感性仅为26%[57]。ESM在外部验证中也暴露出一些局限性,其敏感性低于当前临床实践,导致67%的脓毒症患者未被识别,7%的患者未能及时接受抗生素治疗,进而引发了警报疲劳等问题[56]。过高的假阳性亦会导致不必要的治疗,过度消耗医疗资源[58]。通过对比模型预测结果与ML里的真值(Ground Truth,GT),可调整分类阈值或模型参数,实现灵敏度与特异度的平衡。GT有助于缩短临床医生排除脓毒症并制定重大治疗决策的时间窗,是体现ML临床价值潜力的关键[59]。而当数据标注不充分或缺乏有效整合时,模型易发生本地化适应失败(localization failure)和跨模态失效(cross-modal failure)[60]。因而,鉴于脓毒症复杂异质性,构建AI模型的数据(临床电子病历资料、多组学数据等)呈指数级增长。在儿科脓毒症生物标志物风险模型(pediatric sepsis biomarker risk model,PERSEVERE)系列研究中,随着研究者纳入血液学指标、生物标志物等后,PERSEVERE-Ⅱ[61]、PERSEVERE-XP[62]、PERSEVEREnce[63]在脓毒性休克和血小板减少症相关的多器官衰竭评估中性能逐渐提升。然而,随着纳入数据指数级增长,过拟合风险也会相应增加,从而限制模型在不同患者群体中的适用性[464]。当前AI在脓毒症诊疗的模型结构与功能设计上仍存在一定问题,循环神经网络(recurrent neural networks,RNNs)、XGBoost等普遍存在可解释性较低,常被批评为“黑箱”模型[65],且难以满足动态、复杂临床环境下的推理需求[66],缺乏临床医护人员信任,难以在ICU场景中普遍推广应用。始终应该清晰地认识到,AI模型输出需遵循临床推理逻辑[67],不要简单地将预测效能低下归因于变量或模型本身,而是要回到指标计算逻辑与临床场景,深入分析背后的原因,单一指标也只是反映了脓毒症复杂异质性与病理过程中某一个节点的情况。

3.3 对多样化专业融合需求的挑战

脓毒症AI诊疗的设计、开发以及效能提升还受研究者对疾病认知等专业素养影响。本文以(脓毒症)AND(“人工智能”OR“AI”OR“机器学习”OR“深度学习”OR“神经网络”)为检索词,筛选出近10年来相关文献5 712篇(包含651篇综述/系统综述),医学类专业作者4 662篇(81.6%),计算机科学专业作者171篇(2.9%),2个专业合作完成889篇(15.5%)。目前,已逐渐有更多的计算机科学专业领域研究人员参与到脓毒症AI研究工作中,近5年(2021至2025年)计算机科学专业领域作者占比为4.78%(127/2 659),较前5年的1.44%(44/3 063)显著升高;但以多学科合作形式开展的研究占比为13.43%(357/2 659),较前5年的17.37%(532/3 063)有所下降。从技术角度来看,脓毒症生物标志物从发现平台成功转化为应用平台需要更优质的ML算法,以有效解决性能损失问题。因此,将AI预测转化为切实的脓毒症临床诊疗需求,在准确性、可解释性和临床实用性之间取得“适当”的平衡是关键,这也凸显了技术创新与临床实践深度融合的必要性。

4 AI赋能脓毒症诊疗实践方向与展望

4.1 多组学融合下AI生物标志物筛选与转化

AI与多模态技术融合被视为揭示脓毒症分子异质性,促进表型分类与分型诊断标准化,制定靶向治疗策略的重要发展方向。以标志性生物标志物(组)为突破口提升早期诊断效能是未来AI赋能脓毒症诊疗最具潜力的探索方向,最有可能筛选出治疗中的获益群体[68]。通过AI技术各类信息(包括免疫学、临床、微生物学和高通量组学数据)整合在所谓“组合型”中,指导临床决策,是实现脓毒症精准医学的直观步骤[2969-70]。分析脓毒症临床样本的转录组学[71]、蛋白质组学[72]、代谢组学[73]等有助于快速诊断和不同亚类(“内型”)鉴定,也为个体化诊疗提供了坚实数据支撑与理论基础[6774]。高通量组学技术能用于筛选特异性免疫生物标志物,解析“免疫失调”新范式——“sepsis systems immunology”[75]。单细胞测序分析则对脓毒症细胞组成和功能特征分析更加精准[76],也更能识别与捕捉脓毒症患者靶细胞群中的亚群特性[77]。未来可基于多组学数据构建AI神经网络分类器(neural network classifier,NNET),用于区分脓毒症的病原微生物类型,从而在早期优化抗感染治疗方案。未来应积极推动多组学生物标志物的组合在特殊人群(如老年人、儿童及免疫功能低下者)中的应用,将其整合至可行的算法或决策树中,以支持临床医生在不同场景下的精准决策(例如PERSEVERE模型系列研究[61-63])。通过AI实现脓毒症诊疗数据的集成、模型的持续优化与转化研究的临床落地,有望显著提升脓毒症的诊断精度。

4.2 自主化AI脓毒症风险预测平台构建与搭载

构建AI脓毒症临床决策支持工具并实现部署是智慧重症发展重要工作。当前应积极对MIMIC、eICU、儿科重症监护(paediatric intensive care,PIC)、AmsterdamUMCdb、高时间分辨率重症监护数据集(high time resolution ICU data set,HiRID)等临床重症医学数据库进行深度挖掘,探索性地构建与使用可解释性ML,推进脓毒症不同表型聚类分析[78],优化脓毒症的集束化治疗。进一步对体外膜肺氧合(extracorporeal membrane oxygenation,ECMO)、连续肾脏替代治疗(continuous renal replacement therapy,CRRT)等高级生命支持远程与体系化管理提供更优质的支持。新开发的AI工具SepsisLab基于对大量临床数据分析,能够在4 h内预测脓毒症风险,并主动识别数据缺失问题,明显提升预测准确性(数据量增加8%可使准确率提高11%)[79],已逐步形成可部署平台。更为重要的是需要构建具有自主性的多中心临床队列,反映我国脓毒症诊疗的实际状态,开发AI脓毒症风险在线预测平台[33]。同时,这些工作需要全面了解临床使用者对AI工具的态度、经验、熟悉程度和使用舒适度,以提升AI采用率[31]。另一种新兴未来数据驱动的方法被称为“数字孪生(digital twins)”[80],即物理实体或流程的虚拟复制体,通过传感器数据实时映射实体状态,实现全生命周期仿真与动态优化。可以通过“数字孪生”创建危重患者病理生理的虚拟复制品,开展模拟/虚拟诊疗,以及研发治疗新技术、新方法。此外,还能通过“人工智能临床医生”从大量患者数据中提取隐含知识,通过分析海量(大多数次优)治疗决策来学习脓毒症最佳治疗方案[81]

综上所述,脓毒症的高度异质性使其诊疗面临复杂多变的临床挑战,推动其精准医疗升级已刻不容缓。借助AI工具实施富集策略以破解脓毒症的生物学异质性,是开展精准医疗的重要前提[1]。AI能够整合多组学、多维度数据,构建动态风险评估模型,实现对脓毒症病情恶化风险的早期预警与及时干预,是推动精准医疗落地的关键举措。为确保AI系统可解释性与临床部署,临床医学与生物学、数据与计算机科学、伦理学等多个领域需加强合作[66],构建集技术实现、价值导向与临床落地于一体的AI系统(概念性图谱见图1),使其真正融入脓毒症临床诊疗流程。未来应研发出契合我国重症医学诊疗特征的脓毒症预测“通用”模型并加以应用,获得突破性解决方案,以提升脓毒症早期诊疗与精准治疗效能。

参考文献

[1]

Wong HR. Intensive care medicine in 2050:precision medicine[J]. Intensive Care Med201743(10):1507-1509.

[2]

Pickkers PKox M. Towards precision medicine for sepsis patients[J]. Crit Care201721(1):11.

[3]

Cecconi MEvans LLevy Met al. Sepsis and septic shock[J]. Lancet2018392(10141):75-87.

[4]

Llitjos JFCarrol EDOsuchowski MFet al. Enhancing sepsis biomarker development: key considerations from public and private perspectives[J]. Crit Care202428(1):238.

[5]

Wang WLiu CF. Sepsis heterogeneity[J]. World J Pediatr202319(10):919-927.

[6]

Guo FZhu XSWu ZHet al. Clinical applications of machine learning in the survival prediction and classification of sepsis: coagulation and heparin usage matter[J]. J Transl Med202220(1):265.

[7]

Seymour CWKennedy JNWang Set al. Derivation,validation,and potential treatment implications of novel clinical phenotypes for sepsis[J]. JAMA2019321(20):2003-2017.

[8]

Postelnicu RPastores SMChong DHet al. Sepsis early warning scoring systems:the ideal tool remains elusive![J]. J Crit Care201952:251-253.

[9]

Martin-Loeches ISinger MLeone M. Sepsis: key insights,future directions,and immediate goals. A review and expert opinion[J]. Intensive Care Med202450(12):2043-2049.

[10]

Kim HJKo RELim SYet al. Sepsis alert systems,mortality,and adherence in emergency departments: a systematic review and meta-analysis[J]. JAMA Netw Open20247(7):e2422823.

[11]

Póvoa PCoelho LDal-Pizzol Fet al. How to use biomarkers of infection or sepsis at the bedside: guide to clinicians[J]. Intensive Care Med202349(2):142-153.

[12]

Kumar SGupta EKaushik Set al. Evaluation of oxidative stress and antioxidant status:correlation with the severity of sepsis[J]. Scand J Immunol201887(4):e12653.

[13]

Liu WGeng FYu L. Long non-coding RNA MALAT1/microRNA 125a axis presents excellent value in discriminating sepsis patients and exhibits positive association with general disease severity,organ injury,inflammation level,and mortality in sepsis patients[J]. J Clin Lab Anal202034(6):e23222.

[14]

Cao JXu FLin SHet al. IL-35 is elevated in clinical and experimental sepsis and mediates inflammation[J]. Clin Immunol2015161(2):89-95.

[15]

Xu FLin SHYan XXet al. Interleukin 38 protects against lethal sepsis[J]. J Infect Dis2018218(7):1175-1184.

[16]

Cao JXu FLin SHet al. IL-27 controls sepsis-induced impairment of lung antibacterial host defence[J]. Thorax201469(10):926-937.

[17]

You YHLi YYYe Let al. Interleukins-27 aggravates liver injury by impairing the antimicrobial response of macrophages via the promotion of mitochondrial dysfunction in the context of sepsis[J]. Mediators Inflamm20252025:6608718.

[18]

Sinha PMeyer NJCalfee CS. Biological phenotyping in sepsis and acute respiratory distress syndrome[J]. Annu Rev Med202374:457-471.

[19]

Ye QLai XFLiu YLet al. Single-cell multi-omic landscape reveals anatomical-specific immune features in adult and pediatric sepsis[J]. Nat Immunol202627(1):150-165.

[20]

Hotchkiss RSMoldawer LLOpal SMet al. Sepsis and septic shock[J]. Nat Rev Dis Primers20162:16045.

[21]

Vincent JL. The clinical challenge of sepsis identification and monitoring[J]. PLoS Med201613(5):e1002022.

[22]

Kofoed KAndersen OKronborg Get al. Use of plasma C-reactive protein,procalcitonin,neutrophils,macrophage migration inhibitory factor,soluble urokinase-type plasminogen activator receptor,and soluble triggering receptor expressed on myeloid cells-1 in combination to diagnose infections:a prospective study[J]. Crit Care200711(2):R38.

[23]

Zhang CZhang XLSun ZDet al. MetaSepsisBase: a biomarker database for systems biological analysis and personalized diagnosis of heterogeneous human sepsis[J]. Intensive Care Med202349(8):1015-1017.

[24]

Tidswell RParker TBrealey Det al. Sepsis-the broken code how accurately is sepsis being diagnosed?[J]. J Infect202081(6):31-32.

[25]

Seymour CWLiu VXIwashyna TJet al. Assessment of clinical criteria for sepsis: for the third international consensus definitions for sepsis and septic shock(sepsis-3)[J]. JAMA2016315(8):762-774.

[26]

Smith GBRedfern OCPimentel MAet al. The national early warning score 2 (NEWS2)[J]. Clin Med201919(3):260.

[27]

Churpek MMSnyder AHan Xet al. Quick sepsis-related organ failure assessment,systemic inflammatory response syndrome,and early warning scores for detecting clinical deterioration in infected patients outside the intensive care unit[J]. Am J Respir Crit Care Med2017195(7):906-911.

[28]

Yan MYGustad LTNytrø Ø. Sepsis prediction,early detection,and identification using clinical text for machine learning: a systematic review[J]. J Am Med Inform Assoc202229(3):559-575.

[29]

Goh KHWang LYeow AYKet al. Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare[J]. Nat Commun202112(1):711.

[30]

Perlin JBJackson EHall Cet al. 2019 john M. eisenberg patient safety and quality awards: spotting sepsis to save lives: a nationwide computer algorithm for early detection of sepsis: innovation in patient safety and quality at the national level(eisenberg award)[J]. Jt Comm J Qual Patient Saf202046(7):381-391.

[31]

Henry KEAdams RParent Cet al. Factors driving provider adoption of the TREWS machine learning-based early warning system and its effects on sepsis treatment timing[J]. Nat Med202228(7):1447-1454.

[32]

Zhao HYKennedy JNWang Set al. Revising host phenotypes of sepsis using microbiology[J]. Front Med20218:775511.

[33]

Sun BSLei MXWang Let al. Prediction of sepsis among patients with major trauma using artificial intelligence: a multicenter validated cohort study[J]. Int J Surg2025111(1):467-480.

[34]

Peiffer-Smadja NDellière SRodriguez Cet al. Machine learning in the clinical microbiology laboratory: has the time come for routine practice?[J]. Clin Microbiol Infect202026(10):1300-1309.

[35]

Weis CCuénod ARieck Bet al. Direct antimicrobial resistance prediction from clinical MALDI-TOF mass spectra using machine learning[J]. Nat Med202228(1):164-174.

[36]

Kim JIMaguire FTsang KKet al. Machine learning for antimicrobial resistance prediction: current practice,limitations,and clinical perspective[J]. Clin Microbiol Rev202235(3):e0017921.

[37]

Scicluna BPvan Vught LAZwinderman AHet al. Classification of patients with sepsis according to blood genomic endotype: a prospective cohort study[J]. Lancet Respir Med20175(10):816-826.

[38]

Davenport EEBurnham KLRadhakrishnan Jet al. Genomic landscape of the individual host response and outcomes in sepsis: a prospective cohort study[J]. Lancet Respir Med20164(4):259-271.

[39]

Sweeney TEShidham AWong HRet al. A comprehensive time-course-based multicohort analysis of sepsis and sterile inflammation reveals a robust diagnostic gene set[J]. Sci Transl Med20157(287):287ra71.

[40]

Sweeney TEWong HRKhatri P. Robust classification of bacterial and viral infections via integrated host gene expression diagnostics[J]. Sci Transl Med20168(346):346ra91.

[41]

Sweeney TEPerumal TMHenao Ret al. A community approach to mortality prediction in sepsis via gene expression analysis[J]. Nat Commun20189(1):694.

[42]

Mayhew MBButurovic LLuethy Ret al. A generalizable 29-mRNA neural-network classifier for acute bacterial and viral infections[J]. Nat Commun202011(1):1177.

[43]

You GHZhao XGLiu JRet al. Machine learning-based identification of CYBB and FCAR as potential neutrophil extracellular trap-related treatment targets in sepsis[J]. Front Immunol202314:1253833.

[44]

Xu PTao ZZhang C. Integrated multi-omics and artificial intelligence to explore new neutrophils clusters and potential biomarkers in sepsis with experimental validation[J]. Front Immunol202415:1377817.

[45]

Li XLJiang SJWang BYet al. Integrated multi-omics analysis and machine learning developed diagnostic markers and prognostic model based on efferocytosis-associated signatures for septic cardiomyopathy[J]. Clin Immunol2024265:110301.

[46]

Gong TLiu YBTian ZYet al. Identification of immune-related endoplasmic reticulum stress genes in sepsis using bioinformatics and machine learning[J]. Front Immunol202213:995974.

[47]

Li ZZhang CLiu YQet al. Diagnostic and predictive values of ferroptosis-related genes in child sepsis[J]. Front Immunol202213:881914.

[48]

Zhang QQBao XWCui MTet al. Identification and validation of key biomarkers based on RNA methylation genes in sepsis[J]. Front Immunol202314:1231898.

[49]

Bataille Bde Selle JMoussot PEet al. Machine learning methods to improve bedside fluid responsiveness prediction in severe sepsis or septic shock: an observational study[J]. Br J Anaesth2021126(4):826-834.

[50]

Oostenbrug GSMensink RPHardeman MRet al. Exercise performance,red blood cell deformability,and lipid peroxidation: effects of fish oil and vitamin E[J]. J Appl Physiol199783(3):746-752.

[51]

Yue SRLi SSHuang XYet al. Machine learning for the prediction of acute kidney injury in patients with sepsis[J]. J Transl Med202220(1):215.

[52]

Takkavatakarn KOh WChan LLet al. Machine learning derived serum creatinine trajectories in acute kidney injury in critically ill patients with sepsis[J]. Crit Care202428(1):156.

[53]

Wen CLZhang XLi Yet al. An interpretable machine learning model for predicting 28-day mortality in patients with sepsis-associated liver injury[J]. PLoS One202419(5):e0303469.

[54]

Bignami EGBerdini MPanizzi Met al. Artificial intelligence in sepsis management: an overview for clinicians[J]. J Clin Med202514(1):286.

[55]

Fleuren LMKlausch TLTZwager CLet al. Machine learning for the prediction of sepsis: a systematic review and meta-analysis of diagnostic test accuracy[J]. Intensive Care Med202046(3):383-400.

[56]

Wong AOtles EDonnelly JPet al. External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients[J]. JAMA Intern Med2021181(8):1065-1070.

[57]

Giannini HMGinestra JCChivers Cet al. A machine learning algorithm to predict severe sepsis and septic shock: development,implementation,and impact on clinical practice[J]. Crit Care Med201947(11):1485-1492.

[58]

Schinkel Mvan der Poll TWiersinga WJ. Artificial intelligence for early sepsis detection: a word of caution[J]. Am J Respir Crit Care Med2023207(7):853-854.

[59]

Lindner HAThiel MSchneider-Lindner V. Clinical ground truth in machine learning for early sepsis diagnosis[J]. Lancet Digit Health20235(6):338-339.

[60]

Cesaro AHoffman SCDas Pet al. Challenges and applications of artificial intelligence in infectious diseases and antimicrobial resistance[J]. NPJ Antimicrob Resist20253(1):2.

[61]

Wong HRCvijanovich NZAnas Net al. Pediatric sepsis biomarker risk model-Ⅱ: redefining the pediatric sepsis biomarker risk model with septic shock phenotype[J]. Crit Care Med201644(11):2010-2017.

[62]

Wong HRCvijanovich NZAnas Net al. Improved risk stratification in pediatric septic shock using both protein and mRNA biomarkers. PERSEVERE-XP[J]. Am J Respir Crit Care Med2017196(4):494-501.

[63]

Atreya MRCvijanovich NZFitzgerald JCet al. Integrated PERSEVERE and endothelial biomarker risk model predicts death and persistent MODS in pediatric septic shock: a secondary analysis of a prospective observational study[J]. Crit Care202226(1):210.

[64]

Sarantopoulos AMastori Kourmpani CYokarasa ALet al. Artificial intelligence in infectious disease clinical practice:an overview of gaps,opportunities,and limitations[J]. Trop Med Infect Dis20249(10):228.

[65]

Gorecki GPTomescu DRPleș Let al. Implications of using artificial intelligence in the diagnosis of sepsis/sepsis shock[J]. Germs202414(1):77-84.

[66]

O’Reilly DMcGrath JMartin-Loeches I. Optimizing artificial intelligence in sepsis management:opportunities in the present and looking closely to the future[J]. J Intensive Med20234(1):34-45.

[67]

Nicolaou AStylianides CSulaiman WAet al. An overview of explainable AI studies in the prediction of sepsis onset and sepsis mortality[J]. Stud Health Technol Inform2024316:808-812.

[68]

Cavaillon JMSinger MSkirecki T. Sepsis therapies: learning from 30 years of failure of translational research to propose new leads[J]. EMBO Mol Med202012(4):e10128.

[69]

Cajander SKox MScicluna BPet al. Profiling the dysregulated immune response in sepsis:overcoming challenges to achieve the goal of precision medicine[J]. Lancet Respir Med202412(4):305-322.

[70]

Topol EJ. High-performance medicine: the convergence of human and artificial intelligence[J]. Nat Med201925(1):44-56.

[71]

Baghela APena OMLee AHet al. Predicting sepsis severity at first clinical presentation:the role of endotypes and mechanistic signatures[J]. EBioMedicine202275:103776.

[72]

Sharma NKSalomao R. Sepsis through the eyes of proteomics: the progress in the last decade[J]. Shock2017,47(1S ):17-25.

[73]

Chang YJYoo HJKim SJet al. A targeted metabolomics approach for sepsis-induced ARDS and its subphenotypes[J]. Crit Care202327(1):263.

[74]

Stylianides CNicolaou ASulaiman WAet al. AI advances in ICU with an emphasis on sepsis prediction: an overview[J]. Mach Learn Knowl Extr20257(1):6.

[75]

Bonaguro LSchulte-Schrepping JUlas Tet al. A guide to systems-level immunomics[J]. Nat Immunol202223(10):1412-1423.

[76]

Sun XJWu JRLiu Let al. Transcriptional switch of hepatocytes initiates macrophage recruitment and T-cell suppression in endotoxemia[J]. J Hepatol202277(2):436-452.

[77]

Reshef YARumker LKang JBet al. Co-varying neighborhood analysis identifies cell populations associated with phenotypes of interest from single-cell transcriptomics[J]. Nat Biotechnol202240(3):355-363.

[78]

Jiang ZYBo LLWang Let al. Interpretable machine-learning model for real-time,clustered risk factor analysis of sepsis and septic death in critical care[J]. Comput Methods Programs Biomed2023241:107772.

[79]

Yin CCChen PYYao BSet al. SepsisLab: early sepsis prediction with uncertainty quantification and active sensing[J]. KDD20242024:6158-6168.

[80]

Vodovotz Y. Towards systems immunology of critical illness at scale: from single cell'omics to digital twins[J]. Trends Immunol202344(5):345-355.

[81]

Komorowski MCeli LABadawi Oet al. The artificial intelligence clinician learns optimal treatment strategies for sepsis in intensive care[J]. Nat Med201824(11):1716-1720.

基金资助

国家自然科学基金资助项目(82172161)

AI Summary AI Mindmap
PDF (1032KB)

201

访问

0

被引

详细

导航
相关文章

AI思维导图

/