基于MIMIC-Ⅲ数据库对ICU患者结局预测的研究  

Study on Prediction of ICU Patient’s Outcome by Exploring Public MIMIC-Ⅲ Database

作　　者：黄菩臣 练作为 陶敏 杨永强 杜江[2] 赵蕴龙 HUANG Puchen;LIAN Zuowei;TAO Ming;YANG Yongqiang;DU Jiang;ZHAO Yunlong(Institute of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing Jiangsu 211106,China;Shanghai First People’s Hospital Affiliated to Shanghai Jiao Tong University,Shanghai 201620,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,江苏南京211106 [2]上海交通大学附属上海市第一人民医院,上海201620

出　　处：《中国医疗设备》2019年第12期92-96,共5页China Medical Devices

基　　金：南京航空航天大学计算机科学与技术学院大学生科技创新基金(NUAA-CS083)

摘　　要：目的通过真实数据预测ICU患者的结局及对关键样本特征进行可视化。方法基于MIMIC-Ⅲ数据库,通过数据清洗、特征选取等数据预处理方法从50000多例数据提取出研究所需要的原始数据,并通过机器学习算法(逻辑回归和线性SVM)进行ICU病房患者结局预测(存活/死亡)研究。同时基于ECharts开源可视化库对原始数据中关键样本特征进行数据可视化研究,分析出相应样本特征对于患者结局的关联性。结果逻辑回归算法的预测准确率最高,能达到70%,线性SVM能达到50%。可视化结果表明存活和死亡患者的乳酸、血肌酐、钾含量、钠含量分布有显著差异。结论针对MIIMC-Ⅲ数据库,逻辑回归算法所建立的模型对患者结局有更好的预测效果,可视化样本特征的结论对医生的诊断筛查具有重要的参考价值。Objective To predict outcomes in ICU patients and visualize key sample features using real-world data. Methods Based on the MIMIC-Ⅲ database, data preprocessing methods such as data cleaning and feature selection were used to extract the raw data needed by the research from more than 50,000 data, and the outcome prediction(survival/death) of ICU wards was performed by machine learning algorithm(logical regression and linear SVM). At the same time, based on the ECharts open source visualization library, the data visualization of the key sample features in the original data was carried out, and the correlation of the corresponding sample features to the patient outcome was analyzed. Results The prediction accuracy of logistic regression algorithm was up to 70% while the linear SVM was up to 50%. According to the visualization results, there was a significant difference in the distribution of lactic acid, serum creatinine, potassium, and sodium in patients with survival and death. Conclusion For the MIIMC-Ⅲ database, the model established by the logistic regression algorithm has a better predictive effect on the patient’s outcome, and the conclusion of visualizing the sample characteristics has important reference value for the doctor’s diagnostic screening.

关 键 词：MIMIC-Ⅲ 数据清洗 逻辑回归 SVM 数据可视化 

分 类 号：R449[医药卫生—诊断学]