文献速递呼吸衰竭与机械通气重症综合管理与镇静镇痛

机器学习预测80岁及以上ARDS患者ICU住院死亡率:一项多国多中心回顾性研究

Machine Learning Predicts ICU In-Hospital Mortality in Ards Patients Aged 80 and Above: A Multinational Multicenter Retrospective Study.
PubMed 文献 · Shock · 2026-03-17
一句话结论基于六家中国医疗机构与MIMIC-IV数据库构建的随机森林模型(AUC=0.835)可预测80岁以上ARDS患者ICU住院死亡率并进行风险分层,性能优于APACHE II评分和氧合指数风险分类。

本研究基于中国六家医疗机构与MIMIC-IV数据库的ICU临床数据,纳入80岁以上急性呼吸窘迫综合征(ARDS)患者,采用八种机器学习方法构建死亡预测模型,以受试者工作特征曲线下面积(AUC)等指标评价性能。随机森林(RF)模型整体性能最佳(AUC=0.835),被选为最终预测模型,用于风险预测与患者分层。RF模型的风险预测与分层能力显著优于急性生理与慢性健康评分II(APACHE II)和基于氧合指数的风险分类。结合SHapley Additive exPlanations(SHAP)分析可解释模型的决策机制,为这类患者的临床风险分层管理提供依据。

要点

  • 研究纳入中国六家医疗机构与MIMIC-IV数据库中80岁以上ARDS患者,构建八种机器学习死亡预测模型。
  • 随机森林模型整体性能最佳,AUC=0.835,被选为最终预测模型。
  • RF模型的风险预测与分层能力显著优于APACHE II评分和基于氧合指数的风险分类。
  • 应用SHAP分析在全局与局部层面解释模型决策机制,支持风险预测的精确解读。
  • 该模型为80岁以上ARDS患者ICU住院死亡率的临床风险分层管理提供依据。

适用人群与场景

  • 中国六家医疗机构与MIMIC-IV数据库中入住ICU的80岁以上急性呼吸窘迫综合征患者。

研究设计与关键结果

  • 研究类型为多国多中心回顾性研究,数据来自中国六家医疗机构与MIMIC-IV数据库,纳入80岁以上ARDS患者。
  • 采用八种机器学习方法构建预测模型,以AUC等指标评价性能。
  • 随机森林模型整体性能最佳,AUC=0.835,被选为最终预测模型。
  • RF模型用于风险预测与患者分层,其风险预测和分层能力显著优于APACHE II评分系统和基于氧合指数的风险分类。
  • 应用SHAP分析在全局与局部层面解释模型决策机制。

证据类型、等级与局限

  • 证据类型:多国多中心回顾性研究。
  • 局限:原文未报告局限性。

对床旁实践的改变

  • 原文结论认为,基于多中心队列开发的机器学习模型可准确预测80岁以上ARDS患者ICU住院死亡率,结合SHAP分析可精确解读风险预测,为这类患者的临床风险分层管理提供科学有效的依据。

提炼说明

  • 本条依据论文摘要整理;摘要未涉及「诊断标准与关键阈值」「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: The present study aims to develop and validate an interpretable machine learning (ML) model based on a multicenter cohort, which is intended for predicting the mortality of acute respiratory distress syndrome patients aged over 80 years admitted to the intensive care unit (ICU) and realizing risk stratification for this patient population.

METHODS: The research cohort drew from ICU clinical data from six medical institutions in China and the Medical Information Mart for Intensive Care (MIMIC-IV) database. In this study, eight distinct ML methods were employed to construct predictive models. The comprehensive performance of these models was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC). The best-performing model was utilized for risk prediction and patient stratification, and its results were compared with those of the traditional Acute Physiology and Chronic Health Evaluation II scoring system and the oxygenation index-based risk classification. Furthermore, SHapley Additive exPlanations analysis was applied to interpret the model's intrinsic decision-making mechanisms at both global and local levels.

RESULTS: Among the eight ML models evaluated, the random forest (RF) model demonstrated the highest overall performance (area under the receiver operating characteristic curve = 0.835) and was selected as the final predictive model. Utilizing the RF model, patients were stratified into risk categories. The results indicate that the model accurately predicted patient mortality and effectively stratified patients based on risk. Furthermore, the risk prediction and stratification capabilities of the RF model significantly outperformed those of the Acute Physiology and Chronic Health Evaluation II scoring system and the oxygenation index-based risk classification.

CONCLUSION: The ML model developed on the basis of a multicenter cohort demonstrated accurate prediction of mortality in ICU patients aged over 80 with acute respiratory distress syndrome. Integrated with SHapley Additive exPlanations analysis, the model enables precise interpretation of risk predictions and provides a scientific and effective basis for the clinical risk stratification management of these patients.

来源:Shock,查看原文。版权归原作者所有。

参考来源

  1. Liu X, Sun X, Zhou X, et al. Machine Learning Predicts ICU In-Hospital Mortality in Ards Patients Aged 80 and Above: A Multinational Multicenter Retrospective Study. Shock. 2026. PMID: 41949844. doi:10.1097/SHK.0000000000002834

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。