基于BP算法的住院天数神经网络建模研究  

Research of Establishing Length of Stay Model Based on BP Neural Network

作　　者：严武[1] 

机构地区：[1]厦门大学附属中山医院信息中心,厦门市361004

出　　处：《中国病案》2009年第11期38-40,共3页Chinese Medical Record

摘　　要：目的建立基于BP神经网络的住院天数拟合模型,并在已建立的神经网络模型的基础上,进行住院天数的预测和影响因素的敏感度分析,利用本研究的建模结果,为BP神经网络建模的方法学提供一定的参考依据,并能帮助卫生管理决策者做出正确的决策和分析。方法利用SQL提取HIS数据,在Clementine 11.1中进行建模和预测,预测结果用SPSS16.0进行假设检验。结果BP神经网络的拟合度和预测准确度分别为96.678%和86.67%,术前住院天数对射频消融术患者的住院天数影响最大。结论BP神经网络相对其他传统统计方法而言,是比较适合于住院天数数据特征的建模方法。Objective Fitting model for length of stay （LOS） based on BP neural network was established. Moreover, the LOS was predicted and the sensitivity of impact factors was analyzed based on the above-mentioned BP neural network. The modeling results of our study couht be used to provide some reference for modeling methodology of the BP neutral network, and help the decision makers of health management make correct decisions and analysis. Methods HIS data was extracted by using SQI. software. Iu the Clementine 11. 1, we constructed the modeling and predicted the results, and the hypothesis test of the results was perfnrmed with the SPSS16.0 software. Results For the BP neutral network, the degree of fitting amt the degree of predicting accuracy were 96. 678 % and 86.67 %, respectively. Among all impact factors, the hospitalization days before operation had maximal influence on the LOS of patients receiving radiofrequency catheter ablation （RFCA）. Conclusions Compared with other traditional statistical methods, the BP neutral network is a modeling method suitable for the characteristics of the LOS data.

关 键 词：住院天数 BP神经网络 多元线形回归 影响因素 

分 类 号：R197.324[医药卫生—卫生事业管理]