基于电子病历的急性冠脉综合征患者主要不良心血管事件预测

发布时间：2018-01-01 16:19

本文关键词：基于电子病历的急性冠脉综合征患者主要不良心血管事件预测　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：主要不良心血管事件预测与评估是研究急性冠脉综合征等心血管疾病致病危险因素与疾病发病率、死亡率之间数量依存关系及规律的技术,被普遍认为是进行疾病防治的核心环节。预测结果能够为医生提供临床决策支持,辅助医生制定合理的治疗及护理方案,从而减小患者发生不良事件的几率;更能规范医疗流程,减少医疗开支。传统队列研究通过入组标准控制患者质量,采用少量精选风险因子构建模型,使用简单并已得到广泛的临床认可。但其存在如入组标准导致入组患者与实际临床环境不同;少量风险因子限制模型性能;难以纳入新的风险因子等不足。随着电子病历等医疗信息系统的快速发展,大量研究开始采用电子病历数据构建预测模型。相对于队列研究,该类模型没有严格的入组标准,数据反应真实临床环境;数据丰富,可用患者信息多;可纳入新的风险因子。尽管克服了队列研究的不足,但依然存在如1)电子病历数据尚未充分利用2)数据不准确值及缺失值导致模型不确定性大、预测结果不准确等问题。因此,本论文针对上述基于电子病历数据预测方法的不足,提出了一种基于电子病历数据挖掘的主要不良心血管事件预测方法。该方法主要由四部分组成:第一,在处理检查检验数据同时,使用自然语言处理技术从入院记录中提取患者特征,充分使用获取到的电子病历数据。第二,使用四种常用的机器学习算法,即支持向量机、随机森林、朴素贝叶斯及范数一逻辑回归,构建独立不良事件预测模型。第三,使用粗糙集理论计算各独立不良事件预测模型的权重值,来确定其在集成模型中所应发挥的作用。第四,采用Dempster-Shafer证据理论,将多个独立预测模型的输出结果和已得到广泛临床认可的队列研究模型GRACE相融合,从而得到本轮文提出的集成主要不良心血管事件预测模型。通过使用从医院收集到的2,930份急性冠脉综合征电子病历数据对本论文所提出的集成主要不良心血管事件预测方法进行评估。评估结果表明:1)使用自然语言处理技术深度挖掘非结构化电子病历数据能有效提高不良事件预测精度;2)使用Dempster-Shafer·证据理论构建的集成预测模型在与独立预测模型和其他集成模型对比时,取得了最佳的综合预测性能,有效减少了电子病历数据中不准确值及缺失值对模型预测性能产生的影响。
[Abstract]:The prediction and evaluation of major adverse cardiovascular events is a technique to study the quantitative relationship and regularity between risk factors and morbidity and mortality of cardiovascular diseases such as acute coronary syndrome (ACS). It is generally considered as the core link of disease prevention and treatment. The predicted results can provide doctors with clinical decision support, assist doctors to formulate reasonable treatment and nursing programs, and thus reduce the probability of adverse events. Traditional cohort studies control patient quality through group standards and use a small selection of risk factors to build models. Use is simple and has been widely recognized in clinical practice. However, the presence of such criteria leads to a difference between the patients in the group and the actual clinical environment. A small number of risk factors limit the performance of the model; With the rapid development of medical information systems such as electronic medical records, a large number of studies began to use electronic medical records data to build prediction models. This kind of model has no strict entry standard, and the data reflect the real clinical environment. Abundant data, available patient information; Although it overcomes the shortage of cohort research, it still exists such as: 1) the electronic medical record data has not been fully utilized 2) the inaccurate value and missing value of the data lead to the uncertainty of the model. Therefore, this paper aims at the shortcomings of the above methods based on EMR data prediction. A main adverse cardiovascular event prediction method based on EMR data mining is proposed. The method consists of four parts: first, it processes the inspection data at the same time. Using natural language processing technology to extract patient features from hospital records, fully use the obtained electronic medical record data. Second, use four commonly used machine learning algorithms, namely support vector machine, random forest. Naive Bayes and norm-logical regression to build independent adverse event prediction model. Third, using rough set theory to calculate the weight of each independent adverse event prediction model. To determine its role in the integration model. 4th, using Dempster-Shafer evidence theory. The output results of multiple independent predictive models are fused with GRACE, a cohort study model that has been widely accepted in clinical practice. Thus the integrated major adverse cardiovascular events prediction model proposed in this paper is obtained. 2. 930 electronic medical records of acute coronary syndrome (ACS) were used to evaluate the integrated method for predicting major adverse cardiovascular events proposed in this paper. Deep mining of unstructured EMR data with natural language processing technology can effectively improve the prediction accuracy of adverse events. 2) compared with independent prediction model and other integrated models, the integrated prediction model constructed with Dempster-Shafer 路evidence theory achieves the best comprehensive prediction performance. It can effectively reduce the influence of inaccurate and missing values on the predictive ability of the model.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R541.4

【参考文献】