摘要
ICU中患者的死亡风险预测被认为是重要临床预测任务之一,准确地预测ICU患者的死亡风险可提供有关患者的病情信息,便于及时采取措施来干预,同时有助于有限医疗资源的有效分配.ICU患者病情不稳定,需要密切的监护,大量临床数据被相关监测设备采集、记录和保存,为ICU患者的相关临床决策提供重要参考.以ICU患者30天死亡风险预测为研究目标,基于重症监护医疗信息市场Ⅲ数据库,构建ICU患者死亡风险预测模型,分析相关影响因素,以支持医疗实践中的临床决策.首先提取相关患者数据并进行预处理,采用多种统计量对时间序列特征进行表示;随后选取基分类器,基于融合遗传算法和模拟退火算法的混合优化算法对相应基分类器进行特征选择,基于信息融合思想采用集成学习方法将分类器以装袋方式进行集成,采用真实数据对模型进行验证,并基于准确率、F评分和AUC三个评价指标与经典死亡风险预测模型进行比较,展现出较好的性能;最后基于信息融合对ICU患者死亡风险重要影响因素进行分析,发现集中趋势类统计量更为重要,为临床决策提供参考.
Predicting the mortality risk of patients in the intensive care unit(ICU)is considered to be one of the most important clinical prediction tasks.Accurately predicting the mortality risk of ICU patients can provide information about the patient's condition and facilitate timely measures to intervene.At the same time,it contributes to the effective allocation of limited medical resources.The condition of ICU patients is unstable and requires close monitoring.A large amount of clinical data is collected,recorded and saved by relevant monitoring equipment,which provides an important reference for relevant clinical decision-making of ICU patients.This paper takes the 30-day mortality risk prediction of ICU patients as the research objective,based on the medical information mart for intensive care Ⅲ database,constructs a mortality risk prediction model for ICU patients,and analyzes related influencing factors to support medical care clinical decision in practice.First extract relevant patient data and perform preprocessing,and use multiple statistics to represent time series features.The hybrid optimization algorithm SAGA based on the fusion of genetic algorithm and simulated annealing algorithm is used for feature selection of the corresponding base classifier,and finally uses the ensemble learning method to integrate the classifiers in bagging based on the idea of information fusion,and uses real data to verify the model.And based on the accuracy rate,F_(1) score and AUC three evaluation indicators,the model proposed in this article is compared with the classic mortality risk prediction model and showing better performance.Finally,based on information fusion,the important factors affecting the mortality risk of ICU patients are analyzed,and the central tendency statistics are more important,which provide reference for clinical decision-making.
作者
徐良辰
郭崇慧
XU Liangchen;GUO Chonghui(Institute of Systems Engineering,Dalian University of Technology,Dalian 116024,China)
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2022年第10期2815-2828,共14页
Systems Engineering-Theory & Practice
基金
国家自然科学基金(71771034)
中央高校基本科研业务费(DUT21YG108)。
关键词
患者表示
死亡风险预测
遗传算法
信息融合
可解释性
patient representation
mortality risk prediction
genetic algorithm
information fusion
interpretability
