基于集成学习与深度特征及人工特征的临床电子健康记录早期脓毒症预测
本研究利用PhysioNet/Computing in Cardiology Challenge 2019发布的ICU患者临床电子健康记录数据,其中两个数据集共40,336名受试者的部分数据公开可用,其余作为隐藏测试集。研究采用监督机器学习,在集成学习框架下整合基于脓毒症临床先验知识的人工特征与长短期记忆神经网络自动提取的深度特征,构建6小时前脓毒症早期预测模型。通过30折交叉验证评估,公开数据集上敏感度0.641±0.022、特异度0.844±0.007、归一化效用评分0.401±0.019;团队(UCAS_DataMiner)在完整隐藏测试集上最终归一化效用评分为0.313(测试集A、B、C分别为0.406、0.373、-0.215)。结果表明该模型在脓毒症早期预测中表现良好,且集成学习较任一单一模型有显著改善(p<0.01)。
要点
- 模型可在脓毒症发生前6小时进行早期预测,基于ICU患者每小时记录的40个临床变量(8项生命体征、26项实验室值、6项人口学特征)。
- 公开数据集30折交叉验证:敏感度0.641±0.022,特异度0.844±0.007,归一化效用评分0.401±0.019。
- 隐藏测试集归一化效用评分:完整测试集0.313,测试集A 0.406、B 0.373、C -0.215。
- 集成学习较任一单一模型在性能上有显著改善(p<0.01)。
- 数据来自三个独立医院系统的ICU患者,部分数据公开可用,其余为隐藏测试集。
适用人群与场景
- ICU患者,数据来自三个独立医院系统。
- 使用PhysioNet/Computing in Cardiology Challenge 2019发布的临床电子健康记录数据,其中两个数据集共40,336名受试者的部分数据公开可用,其余为隐藏测试集。
研究设计与关键结果
- 研究类型为监督机器学习模型开发,采用集成学习框架整合基于临床先验知识的人工特征与长短期记忆神经网络自动提取的深度特征,进行6小时前脓毒症早期预测。
- 数据来源为PhysioNet/Computing in Cardiology Challenge 2019发布的ICU患者临床电子健康记录,来自三个独立医院系统;两个数据集共40,336名受试者的部分数据公开可用,其余为隐藏测试集。
- 每个个体每小时测量并记录40个临床变量(8项生命体征、26项实验室值、6项人口学特征),同时提供二分类标签(0或1)。
- 模型通过30折交叉验证评估,公开数据集上敏感度0.641±0.022、特异度0.844±0.007、归一化效用评分0.401±0.019。
- 团队(UCAS_DataMiner)在完整隐藏测试集上最终归一化效用评分为0.313(测试集A 0.406、测试集B 0.373、测试集C -0.215)。
- 集成学习较任一单一模型在性能上有显著改善(p<0.01)。
证据类型、等级与局限
- 证据类型:监督机器学习模型开发研究,使用PhysioNet/Computing in Cardiology Challenge 2019数据。
- 局限:原文未报告局限性。
对床旁实践的改变
- 原文未据此给出推荐。
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
OBJECTIVES: Sepsis is caused by infection and subsequent overreaction of immune system and will severely threaten human life. The early prediction is important for the treatment of sepsis. This report aims to develop an early prediction method for sepsis 6 hours ahead on the basis of clinical electronic health records.
DATA SOURCES: Challenge data are released by PhysioNet/Computing in Cardiology Challenge 2019 and obtained from ICU patients in three separate hospital systems. Part of the data from two datasets, including 40,336 subjects, are publicly available, and the remaining are used as hidden test set. A normalized utility score defined by the organizing committee is used for model performance evaluation.
STUDY SELECTION: The supervised machine learning is applied to tackle this challenge. Specifically, we establish the prediction model under the framework of ensemble learning by integrating the artificial features based on clinical prior knowledge of sepsis with deep features automatically extracted by long short-term memory neural network.
DATA EXTRACTION: Forty clinical variables, including eight vital signs, 26 laboratory values, and six demographics, were measured and recorded once an hour for each individual, and the binary label (0 or 1) was simultaneously provided for each item.
DATA SYNTHESIS: The proposed model was evaluated by 30-fold cross-validation. The sensitivity, specificity, and normalized utility score were 0.641 ± 0.022, 0.844 ± 0.007, and 0.401 ± 0.019 on publicly available datasets, respectively. The final normalized utility score our team (UCAS_DataMiner) has obtained was 0.313 on full hidden test set (0.406, 0.373, and -0.215 on test set A, B, and C, respectively).
CONCLUSIONS: We realized a 6-hour ahead early-onset prediction of sepsis on the basis of clinical electronic health record by ensemble learning. The results indicated the proposed model functioned well in the early prediction of sepsis. In particular, ensemble learning had a significant (p < 0.01) improvement than any single model in performance.
参考来源
- He Z, Du L, Zhang P, et al. Early Sepsis Prediction Using Ensemble Learning With Deep Features and Artificial Features Extracted From Clinical Electronic Health Records. Crit Care Med. 2020. PMID: 33044286. doi:10.1097/CCM.0000000000004644
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。