文献速递创伤与烧伤复苏、休克与脓毒症

基于机器学习的凝血风险指数可预测出血创伤患者的急性创伤性凝血病

A machine learning-based Coagulation Risk Index predicts acute traumatic coagulopathy in bleeding trauma patients.
PubMed 文献 · J Trauma Acute Care Surg · 2024-09-27
一句话结论在出血创伤患者中,基于入院后15分钟连续生命体征的机器学习凝血风险指数(CRI)能准确预测急性创伤性凝血病(ATC)。

本研究在单中心4年数据中,利用入院后15分钟连续生命体征(光电容积描记与心电波形、血氧、血压趋势)开发机器学习凝血风险指数(CRI),以2年数据训练、2年数据测试。共纳入17,567例有连续生命体征数据的患者,52.8%为钝性伤,30.2%为女性,平均年龄44.6岁。在达到大量输血启动阈值(critical administration threshold)的患者中,CRI预测国际标准化比值(INR)>1.2和>1.5的ATC真阳性率分别为95.6%和94.9%,真阴性率分别为88.3%和89.2%。在接受大量输血(MT)的患者中,预测INR>1.2和>1.5的真阳性率分别为92.8%和100%,真阴性率分别为91.3%和88.1%。作者认为CRI可准确预测早期ATC,临床应用可能指导早期止血复苏,推广至院前可能更早治疗ATC。

要点

  • CRI基于入院后15分钟连续生命体征(光电容积描记与心电波形、血氧、血压趋势)预测ATC,训练与测试各用2年数据。
  • 在达到大量输血启动阈值的患者中,CRI预测INR>1.2的ATC真阳性率95.6%、真阴性率88.3%;预测INR>1.5的真阳性率94.9%、真阴性率89.2%。
  • 在接受大量输血的患者中,CRI预测INR>1.2的ATC真阳性率92.8%、真阴性率91.3%;预测INR>1.5的真阳性率100%、真阴性率88.1%。
  • ATC定义分别为到达时INR>1.2和INR>1.5,两个阈值分别建模评估。
  • 作者认为CRI预测准确,临床应用可能指导早期止血复苏,推广至院前可能更早治疗ATC。

适用人群与场景

  • 出血创伤患者,来自单中心4年数据,纳入有连续生命体征数据者共17,567例,其中52.8%为钝性伤、30.2%为女性、平均年龄44.6岁。
  • 评估的两组高危人群为达到大量输血启动阈值(critical administration threshold)的患者和接受大量输血(MT)的患者。

诊断标准与关键阈值

  • 急性创伤性凝血病(ATC)分别定义为到达时国际标准化比值(INR)>1.2和INR>1.5,并在两个独立模型中评估。
  • CRI基于入院后15分钟内的连续生命体征计算,数据来自光电容积描记与心电波形、血氧值和血压趋势的连续特征。

研究设计与关键结果

  • 研究类型为单中心大数据机器学习模型开发与验证,使用4年数据,其中2年用于训练、2年用于测试,共17,567例有连续生命体征数据的患者可供分析。
  • 在达到大量输血启动阈值的患者中,CRI预测ATC的真阳性率与真阴性率:INR>1.2为95.6%与88.3%,INR>1.5为94.9%与89.2%。
  • 在接受大量输血的患者中,CRI预测ATC的真阳性率与真阴性率:INR>1.2为92.8%与91.3%,INR>1.5为100%与88.1%。
  • 模型准确度以受试者工作特征曲线下面积(AUC)及95%置信区间描述。

证据类型、等级与局限

  • 证据类型:诊断试验/标准研究,证据等级为Level III。
  • 局限:原文未报告局限性。

对床旁实践的改变

  • 作者认为,CRI可准确预测出血患者的早期ATC,临床应用可能指导早期止血复苏。
  • 作者认为,将该技术推广至院前环境可能为ATC提供更早的治疗。

提炼说明

  • 本条依据论文摘要整理;摘要未涉及「处置流程与药物剂量」「红旗征与不能漏的情况」「指南或共识出处」,故未列出这些栏目;全文请点「阅读原文」。
英文原文(摘要)· 点击展开

BACKGROUND: Acute traumatic coagulopathy (ATC) is a well-described phenomenon known to begin shortly after injury. This has profound implications for resuscitation from hemorrhagic shock, as ATC is associated with increased risk for massive transfusion (MT) and mortality. We describe a large-data machine learning-based Coagulation Risk Index (CRI) to test the early prediction of ATC in bleeding trauma patients.

METHODS: Coagulation Risk Index was developed using continuous vital signs (VSs) available during the first 15 minutes after admission at a single trauma center over 4 years. Data to compute the CRI were derived from continuous features of photoplethymographic and electrocardiographic waveforms, oximetry values, and blood pressure trends. Two groups of patients at risk for ATC were evaluated: critical administration threshold and patients who received an MT. Acute traumatic coagulopathy was evaluated in separate models and defined as an international normalized ratio (INR) >1.2 and >1.5 upon arrival. The CRI was developed using 2 years of cases for training and 2 years for testing. The accuracy of the models is described by area under the receiver operator curve with 95% confidence intervals.

RESULTS: A total of 17,567 patients were available for analysis with continuous VS data, 52.8% sustained blunt injury, 30.2% were female, and the mean age was 44.6 years. The ability of CRI to predict ATC in critical administration threshold patients was excellent. The true positive and true negative rates were 95.6% and 88.3%, and 94.9% and 89.2% for INR >1.2 and INR >1.5, respectively. The CRI also demonstrated excellent accuracy in patients receiving MT; true positive and true negative rates were 92.8% and 91.3%, and 100% and 88.1% for INR >1.2 and INR >1.5, respectively.

CONCLUSION: Using continuous VSs and large-data machine learning capabilities, the CRI accurately predicts early ATC in bleeding patients. Clinical application may guide early hemostatic resuscitation. Extension of this technology into the prehospital setting could provide earlier treatment of ATC.

LEVEL OF EVIDENCE: Diagnostic Tests/Criteria; Level III.

来源:J Trauma Acute Care Surg,查看原文。版权归原作者所有。

参考来源

  1. Richards JE, Yang S, Kozar RA, et al. A machine learning-based Coagulation Risk Index predicts acute traumatic coagulopathy in bleeding trauma patients. J Trauma Acute Care Surg. 2024. PMID: 39330762. doi:10.1097/TA.0000000000004463

供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。