基于SRS的苦碟子注射液(碟脉灵)ADR信号分析-决策树和神经网络模型比较研究  

ADR signal analysis of Diemailing  Kudiezi injection based on SRS-a comparative study of the decision tree and neural network model

作　　者：阎博华[1,2] 彭成[3] 谢雁鸣[1] 王巧灵[2] 

机构地区：[1]中国中医科学院中医临床基础医学研究所,北京100700 [2]成都中医药大学附属医院/四川省中医院,成都610072 [3]成都中医药大学,成都610075

出　　处：《时珍国医国药》2016年第2期455-458,共4页Lishizhen Medicine and Materia Medica Research

基　　金：国家"重大新药创制"科技重大专项(No.2009ZX09502-030);四川省科技支撑计划(No.2014SZ0071)

摘　　要：目的比较决策树和神经网络模型在苦碟子注射液(碟脉灵)自发呈报系统(SRS)数据库信号预警分析中的应用准确性和价值,明确碟脉灵ADR信号的发生特征。方法选取2007年1月至2012年12月国家不良反应监测中心SRS数据库中的碟脉灵ADR信息,以此为基础构建了碟脉灵不良反应数据库,分别采用决策树和神经网络两种方法对碟脉灵ADR信号进行预测模型建构和提升度检验。结果 (1)碟脉灵ADR发生的可能相关因素,决策树模型显示按重要性排序依次为发生年度、剂量、原发疾病影响、触发时间和年龄,神经网络模型显示重要性排序依次为发生季节、发生年度、年龄、用药次数和触发时间;(2)预测准确度方面,决策树模型为53.93%,神经网络模型50.4%;(3)模型提升度方面,以最常见的不良反应"皮疹"为例进行检测,决策树模型比神经网络模型提升度阈值下降更平缓且表现相对稳定。结论基于现有数据,苦碟子注射液ADR的发生很可能与年龄和触发时间相关;在预测准确度和提升度方面,决策树模型均优于神经网络模型;两者比较决策树可能更适合用于对于碟脉灵现有SRS系统数据的分析。Objective To compare the decision tree and neural network model in Diemailing  Kudiezi injection spontaneous reporting system（ SRS） analysis database application and accuracy of early warning signal value,clear Diemailing injection ADR signal characteristics. Methods Diemailing ADR information from 2007 January- 2012 year in December the National Center for ADR monitoring in the SRS database,which is the base of Diemailing adverse reaction database,using decision tree and neural network two methods for forecasting model construction and improve the degree of inspection of Diemailing ADR signal. Results（ 1） possible related factors of Diemailing ADR,decision tree model shows that according to the importance of order are annual,dose,primary disease effect,triggering time and age,the neural network model shows the importance is in the order of occurrence season,year,age,medication frequency and trigger time;（ 2） forecasting accuracy,decision tree model is 53. 93%,the neural network model 50. 4%;（ 3） model to improve the degree of adverse reactions,with ＂ the most common rash case detection,decision tree model is better than the neural network model of lifting threshold decreased more slowly and relatively stable. Conclusion Based on the existing data,the occurrence of Kudiezi Injection ADR probably correlated with age and time triggered; accuracy and enhance the degree in terms of prediction,decision tree model was better than the neural network model; comparison decision tree may be more suitable for the analysis of the existing Diemailing injection SRS system data.

关 键 词：苦碟子注射液 决策树 神经网络 自发呈报系统 ADR信号 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R283[自动化与计算机技术—控制科学与工程]