北京生物医学工程

基于时间序列相似性的患者结局预测模型

Predictive model for patient outcomes based on time series similarity

作者：王牧雨王妮周阳陈卉

单位：首都医科大学生物医学工程学院（北京  100069） 临床生物力学应用基础研究北京市重点实验室（北京 100069） 通信作者：陈卉。E-mail：chenhui@ccmu.edu.cn

分类号：R318.04 

出版年·卷·期（页码）：2022·41·3（249-254）

摘要：

目的评估患者时间序列相似性，验证融合时间序列相似性的K近邻（K-nearest neighbor，KNN）模型是否可以有效提高患者结局预测的效果。方法整合Medical Information Mart for Intensive Care（MIMIC-III）数据库中急性心肌梗死患者的人口学信息、药物使用情况、疾病诊断、影像学报告、实验室指标以及手术操作数据，使用Jaccard系数、欧氏距离、编辑距离以及动态时间规整计算患者相似性。分别以入院基线数据和住院全程数据计算患者相似性，进而对患者死亡、长时住院和长时重症监护（intensive care unit，ICU）进行预测。使用接受者操作特征曲线下面积（area under curve，AUC）评估预测效果，与基于静态数据的支持向量机（support vector machine，SVM）模型、基于时间序列的长短时记忆（long short-term memory，LSTM）模型进行对比。结果输入数据为住院全程数据时，KNN模型在死亡和长时住院预测中AUC值为0.877和0.946，高于SVM模型（0.825，0.930）和LSTM模型（0.853，0.928）；输入数据为入院基线数据时，KNN模型在三个结局预测中AUC值为0.680、0.738、0.728，与SVM模型（0.719，0.715，0.708）相比各有高低。结论时间序列患者相似性与机器学习方法相结合可以有效提高信息利用率和模型的预测效果。

Objective To demonstrate the effectiveness and superiority of the proposed patient similarity measurement via building the K-nearest neighbor (KNN) for predicting patient outcomes in intensive care unit. Methods We firstly extracted the demographic information, drugs, disease diagnoses, radiological reports, laboratory tests and procedures of patients diagnosed with acute myocardial infarction from the medical information mart for intensive care (MIMIC-III) database. We then used the Jaccard distance, Euclidean distance, edit distance, and dynamic time warping algorithm to calculate patient similarity. The patient similarity was also measured using information at admission and during the whole hospitalization separately. The KNN model was built to predict patient mortality, prolonged length of hospital stay and intensive care unit（ICU） duration, whose performance was compared with the support vector machine (SVM) based on static data and the long short-term memory (LSTM) model based on time series. The area under receiver operating characteristic curve (AUC) was used to evaluate the predictive performance. Results When using all available information from admission to discharge, the KNN model outperformed the SVM and LSTM model for mortality and prolonged length of hospital stay prediction with the AUC of 0.877 and 0.946, separately. When using information at admission, the proposed KNN model also outperformed the SVM model when predicting prolonged length of hospital stay and ICU duration, but lost to the SVM in mortality prediction. Conclusions The proposed patient similarity measurement can effectively improve the information utilization and the performance of the similarity-based models for patient outcomes prediction.

参考文献：

[1]?? Cowie M, Blomster J, Curtis L, et al. Electronic health records to facilitate clinical research[J]. Clinical Research in Cardiology, 2017,106(1):1-9.

?[2] Brown S. Patient similarity: emerging concepts in systems and precision medicine[J]. Frontiers in Physiology, 2016,7:251.

?[3] 李惠萍, 胡安民. 机器学习算法对危重患者死亡率预测效能的比较[J]. 广东医学, 2020,41(16):1642-1646.

Li HP, Hu AM.Artificial intelligent learning for prediction of all-cause mortality in critically ill patients[J]. Guangdong Medical Journal, 2020,41(16):1642-1646.

?[4] Sharafoddini A, Dubin JA, Lee J. Patient similarity in prediction models based on health data: a scoping review[J]. JMIR Medical Informatics, 2017,5(1):e7.

?[5] Beaulieu-Jones BK, Orzechowski P, Moore JH. Mapping patient trajectories using longitudinal extraction and deep learning in the MIMIC-III critical care database[J]. Pacific Symposium on Biocomputing, 2018,23:123-132.

[6]?? Johnson AEW, Pollard TJ, Shen L, et al. MIMIC-III, a freely accessible critical care database[J]. Scientific Data, 2016,3:160035.

?[7] Guo Q, Wu M, Li H, et al. Development and validation of a prognostic nomogram for myocardial infarction patients in intensive care units: a retrospective cohort study[J]. BMJ Open, 2020,10(12):e40291.

?[8] Rajkomar A, Oren E, Chen K, et al. Scalable and accurate deep learning with electronic health records[J]. NPJ Digit Medicine, 2018,1:18.

?[9] Vilares M, Ribadas FJ, Vilares J.Phrase similarity through the edit distance[J]. Lecture Notes in Computer Science, 2004, 3180: 306-317.

[10] Oregi I, Pérez A, Del Ser J, et al. On-line elastic similarity measures for time series[J]. Pattern Recognition, 2019,88:506-517.

[11] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997,9(8):1735-1780.

[12] Kurniati AP, Rojas E, Hogg D, et al. The assessment of data quality issues for process mining in healthcare using Medical Information Mart for Intensive Care III, a freely available e-health record database[J]. Health Informatics Journal, 2018,25(4):1878-1893.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录