用于心衰患者重入院预测的LSTM模型

LSTM-BASED READMISSION PREDICTION OF PATIENTS WITH HEART FAILURE

电子健康记录(EHR)中蕴含着丰富的语义信息,目前对EHR的数据挖掘主要基于传统机器学习方法,涉及大量人工特征工程。但由于EHR数据存在维度大、时间跨度长等特点,传统机器学习方法较难有效地捕捉EHR中的深层语义信息。使用LSTM模型对心力衰竭患者的EHR数据进行建模和训练,以30天内重入院作为预测目标开展研究。实验结果表明,LSTM模型能够有效地捕捉EHR时序数据中的语义信息,与传统机器学习方法相比,在ROC-AUC指标上提升了10.48百分点。

Electronic Health Records(EHR)contain rich semantic information.Currently,the mining of EHR is mainly based on traditional machine learning methods and involves lots of manual feature engineering.However,due to the high dimensionality and long span characteristic of EHR,it is difficult for traditional machine learning methods to effectively capture the deep semantic information contained in EHR.LSTM was used to model and train the EHR data of patients with heart failure.The research was carried out on 30-day readmission prediction.The experimental results show that LSTM model can effectively capture the semantic information in the time series of EHR.Compared with the traditional machine learning methods,the ROC-AUC of the proposed method is improved by at least 10.48%.