摘要
体外心肺复苏(extracorporeal cardiopulmonary resuscitation,ECPR)在提高心搏骤停患者生存率和神经功能恢复方面具有重要的临床价值,但其启动决策高度依赖医生经验,且常受到资源紧张和信息不足的制约。现有风险评分工具多基于静态变量,预测目标单一,难以动态反映患者病情变化,限制了其在ECPR场景中的临床适用性。机器学习为复杂临床决策提供了新的技术路径。静态集成方法虽在性能上优于传统回归模型,但在应对ECPR患者高度异质性及病情动态演变方面仍显不足。动态集成选择与界标法可实现因人而异、随时更新的预测,契合ECPR“启动—管理—撤机”全过程的动态决策需求。与此同时,可解释人工智能通过揭示模型判断依据,增强医护人员对模型输出的理解与信任。当前我国在高质量临床数据积累、跨机构数据协作与标准化体系建设方面尚存在不足,限制了ECPR智能模型的开发与实际部署。未来应依托多中心真实世界数据和前瞻性研究,建立统一数据标准,强化模型可解释性与临床信任机制,推动智能预测工具与临床流程深度融合,构建以人机协作为核心的ECPR智能决策体系,助力实现从经验驱动向数据驱动的科学决策转型。
Extracorporeal cardiopulmonary resuscitation(ECPR)has shown considerable clinical value in improving survival and neurological outcomes in patients with cardiac arrest.However,decisions regarding ECPR initiation are still largely based on clinician judgment,often made under conditions of time pressure,limited data,and resource constraints.Existing risk scoring systems rely heavily on static variables and single-point outcomes,making them insufficient for capturing the real-time,individualized progression of patient conditions in ECPR contexts.Machine learning offers promising methodologies for enhancing complex clinical decision-making.While static ensemble methods have demonstrated improved performance over traditional regression models,they fall short in addressing the heterogeneity of ECPR patients and the dynamic nature of critical illness.In contrast,dynamic ensemble selection and landmarking techniques enable continuously updated,personalized risk prediction,aligning well with the full spectrum of ECPR decision-making—from initiation and management to weaning.Furthermore,explainable artificial intelligence improves model transparency by clarifying the rationale behind predictions,thereby fostering clinician trust and interpretability in high-stakes scenarios.Currently,the development and practical deployment of intelligent ECPR models in China remain constrained by the lack of high-quality clinical data,limited cross-institutional data collaboration,and the absence of standardized data frameworks.Looking ahead,efforts should focus on leveraging multicenter real-world data and prospective studies to establish unified data standards,enhance model interpretability,and build clinical trust mechanisms.By deeply integrating predictive tools into clinical workflows and fostering human-artificial intelligence collaboration,an intelligent decision-making system for ECPR can be established—facilitating a strategic shift from experience-based decisions to data-driven,evidence-based clinical practice.
作者
吴娟
季学丽
陈旭锋
黄夕华
WU Juan;JI Xueli;CHEN Xufeng;HUANG Xihua(Department of Emergency Medicine,the First Affiliated Hospital with Nanjing Medical University,Nanjing,210029,China)
出处
《临床急诊杂志》
2025年第11期655-660,共6页
Journal of Clinical Emergency
基金
江苏省人民医院临床能力提升工程项目(No:JSPH-NB-2022-3)。
关键词
体外心肺复苏
机器学习
动态风险预测
可解释人工智能
智能临床决策
extracorporeal cardiopulmonary resuscitation
machine learning
dynamic risk prediction
explainable artificial intelligence
intelligent clinical decision-making
