基于处方数据的医院药品需求量的关联性预测方法研究

发布时间：2018-04-29 20:10

本文选题：药品需求量预测 + 数据挖掘　；参考：《东北大学》2014年硕士论文

【摘要】：随着疾病种类的增多,药品的供求关系及流通环节越来越复杂,药品用量骤增或骤减的情况也越来越频繁,致使医院药品库存频繁出现供不应求,或库存积压和浪费等现象,因此如何准确地掌握医院药品的需求规律,从而准确预测药品的使用量是医院管理者迫切需要解决的现实问题。另一方面,随着医院管理信息化的深入发展,积累了大量的医院药品出库和使用数据,这些数据以病理为依据,以处方的形式反映了病人的病况、医生的用药习惯、开具的药品品种、数量及药品之间的相互关系。如何充分利用这些历史数据,采用数据挖掘技术,发现药品配伍及用量之间的知识、规律和模式,基于这些模式和知识进行药品需求量的预测,特别是基于用药关联性的医院药品需求量预测,已经成为医院管理者面临的难题,也是管理者普遍关注的研究热点。本课题正是在这一背景下提出的,它是是将需求量分析从单纯的时序分析迈向结合了药品内在关联分析的因果预测的大胆尝试。本文应用数据挖掘技术,从药品出库数据的时序性和药品药理的关联性两个角度出发,通过对大连某三级甲等医院的实地调研所获得的数据进行分析和建模来预测药品的需求量,具体研究内容包括以下几个方面：首先,在熟悉药品需求量预测问题研究方法的文献综述基础上,从提取样本潜在时序性信息的角度,以求和自回归移动平均(ARIMA)模型为基础,同时考虑到ARIMA模型在非线性特征上的不足以及BP神经网络模型优秀的非线性关系学习能力,构建了引入组合思想的ARIMA-BP神经网络预测模型,并通过实例测试验证模型的实用性；然后,从提取样本潜在关联性信息的角度,考虑影响目标药品出库规律的因素,如医生用药习惯、药品的药理性质的影响,构建因果关系预测模型。为完成这一工作,首先从药品的处方数据着手,应用数据挖掘技术Apriori算法分析药品出库规律之间的关联性,旨在为后续的因果关系预测模型提供基础数据；然后以BP神经网络模型为基础,引入遗传算法对BP神经网络模型的初始权值和阀值进行优化,构建GA-BP神经网络预测模型,并通过实例测试验证模型的实用性；最后,从同时提取样本潜在时序性信息和关联性信息的角度,为了进一步提高药品需求量预测模型的准确度,充分利用样本数据的信息,引入组合预测方法的思想,以ARIMA-BP模型和GA-BP模型的预测数据为基础,将两个模型的预测结果进行组合,构建基于GA-BP神经网络算法的智能非线性组合预测模型,并通过实例测试与模型比较验证模型的实用性。
[Abstract]:With the increase of disease types, the relationship between supply and demand of drugs and their circulation are becoming more and more complicated, and the situation of sudden increase or sharp decrease in drug consumption is becoming more and more frequent. As a result, the supply of drugs in hospitals frequently exceeds the supply, or the overstocking and wasting of the stocks occur frequently. Therefore, how to accurately grasp the law of hospital drug demand and accurately predict the use of drugs is a practical problem that hospital administrators urgently need to solve. On the other hand, with the further development of hospital management information, a large number of hospital drug delivery and use data have been accumulated. These data are based on pathology and reflect the patient's condition and doctors' drug use habits in the form of prescriptions. The variety, quantity and interrelation of prescribed drugs. How to make full use of these historical data and adopt data mining technology to find the knowledge, rules and patterns between drug compatibility and dosage, and based on these patterns and knowledge to predict the demand for drugs, Especially, the prediction of hospital drug demand based on drug use relevance has become a difficult problem for hospital administrators, and it is also a research hotspot that managers generally pay attention to. It is a bold attempt to change demand analysis from simple time series analysis to causality prediction combining drug intrinsic correlation analysis. In this paper, the data mining technology is used to analyze the timing of the data and the correlation of pharmaceutical pharmacology. Through the analysis and modeling of the data obtained from the field investigation of a certain Grade 3A hospital in Dalian to predict the demand for drugs, the specific research contents include the following aspects: first, Based on the literature review of the methods of drug demand prediction, and from the point of view of extracting samples' potential temporal information, the sum autoregressive moving average (ARIMA) model is used as the basis. At the same time, considering the lack of nonlinear characteristics of ARIMA model and the excellent learning ability of BP neural network model, the prediction model of ARIMA-BP neural network with combination idea is constructed, and the practicability of the model is verified by an example. Then, from the angle of extracting samples' potential relevance information, considering the factors that affect the law of target drugs' exit, such as doctors' drug usage and pharmacological properties, a causality prediction model is constructed. In order to complete this work, the correlation between drug exiting rules is analyzed by using the data mining technique Apriori algorithm, which aims at providing basic data for the subsequent causality prediction model. Then on the basis of BP neural network model, genetic algorithm is introduced to optimize the initial weight and threshold value of BP neural network model, and the prediction model of GA-BP neural network is constructed, and the practicability of the model is verified by an example. In order to further improve the accuracy of drug demand forecasting model and make full use of the information of sample data, the idea of combination forecasting method is introduced from the angle of simultaneously extracting samples' potential time series information and correlation information, in order to further improve the accuracy of drug demand forecasting model. Based on the prediction data of ARIMA-BP model and GA-BP model, the prediction results of the two models are combined to construct the intelligent nonlinear combination prediction model based on the GA-BP neural network algorithm. The practicability of the model is verified by an example test and a comparison of the model.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R95;TP183

【参考文献】