基于电子病历数据挖掘的疾病危重度动态预测研究

Study of Dynamic Forecasting Model Based on EMR Data Mining for Disease Severity

本文基于覆盖多病种的电子病历大数据,构建了适用于各种常见严重疾病的危重度动态预测模型。具体依据朴素贝叶斯理论、相关性分析和信息增益法等数据挖掘方法,建立模型框架,通过真实电子病历大数据集MIMIC-Ⅲ上的挖掘实验,筛选疾病危重度预测的主要区分特征,验证模型的动态预测效果。针对本模型的大数据实验证明了其进行多病种危重度动态预测的有效性,并筛选出了对疾病危重度具有高分辨性的38项区分特征,揭示了模型短期预测准确度高的特性。

This article has built a dynamic forecast model suitable for all kinds of common serious diseases on the basis of the electronic medical record （EMR） big data. The dynamic forecast model was built according to the data mining methods, including naive Bayes theory, corrrelation analysis, and information gain. Based on the ex- periments in the database of EMR also known as MIMIC III, the effect of the dynamic forecast model was tested, and the main distinguishing characteristics of the severity of diease were screened. Data mining experiment results show that the 38 distinguishing characteristics discovered in this research is discernible to severity of dieases. Meanwhile, this dynamic forecast model is effective for forecasting the severity of various common severe dieases, and the short-term forecasting accuracy is especially high.