模糊贝叶斯置信度递进神经网络法检测药品不良反应报告信号  

Detection of report signals of adverse drug reactions by fuzzy BCPNN

作　　者：刘靖 叶国菊[1] 王启明[1] 刘尉[1] 赵大方 孙骏[3] 李国亮 王新敏[3] 李明[3] LIU Jing;YE Guoju;WANG Qiming;LIU Wei;ZHAO Dafang;SUN Jun;LI Guoliang;WANG Xinmin;LI Ming(College of Science,Hohai University,Nanjing Jiangsu 211100,China;School of Mathematics and Statistics,Hubei Normal University,Huangshi Hubei 435002,China;Jiangsu Center for Adverse Drug Reactions Monitoring,Nanjing Jiangsu 210002,China)

机构地区：[1]河海大学理学院,江苏南京211100 [2]湖北师范大学数学与统计学院,湖北黄石435002 [3]江苏省药品不良反应监测中心,江苏南京210002

出　　处：《中国药物警戒》2022年第10期1113-1117,共5页Chinese Journal of Pharmacovigilance

基　　金：江苏省自然科学基金资助项目(BK20180500);江苏省药品不良反应信号检测及对药品生产安全风险的预警研究(KJ207559)。

摘　　要：目的充分挖掘药品不良反应报告,实现药品不良反应信号检测,为信号验证和临床用药工作提供参考。方法引入模糊数对药品不良反应报告中的模糊语义信息进行量化,构建模糊贝叶斯置信度递进神经网络(FBCPNN)法,与贝叶斯置信度递进神经网络(BCPNN)法进行对比分析一致性,并分析复方骨肽的信号检测结果。结果对江苏省药品不良反应监测中心提供的2014年1月1日至2019年12月31日药品不良反应报告进行信号检测,FBCPNN法检测到11454个信号,其中新的(说明书中未出现)信号共534个,BCPNN法检测到10915个信号,其中新的信号545个。FBCPNN与BCPNN法相比较,灵敏度为0.9103,特异度为0.9766,约登指数为0.8869。结论基于不确定信息的FBCPNN法可充分利用药品不良反应报告的不确定信息,有效地实现不良反应信号检测。Objective To fully tap the reports of adverse drug reactions and detect adverse drug reaction signals so as to provide reference for signal verification and clinical medications.Methods The fuzzy number was introduced to quantify the fuzzy semantic information in reports of adverse reactions.The fuzzy Bayesian confidence propagation neural network(FBCPNN)method was established to be compared with the Bayesian confidence propagation neural network(BCPNN)method in order to analyze the consistency.Finally,the results of signal detection of compound osteopeptide were analyzed.Results 11454 signals of ADR reports provided by Jiangsu ADR Monitoring Center were detected using the FBCPNN method between January 1,2014 and December 31,2019,including 534 new signals(not in the manual).10915 signals were detected using the BCPNN method,including 545 new signals.Compared with the BCPNN method algorithm,the sensitivity,specificity and Youden index of this algorithm were 0.9103,0.9766 and 0.8869 respectively.Conclusion The FBCPNN method based on uncertain information can make full use of the uncertain information of adverse reaction reports and bring about effective detection of adverse reactions.

关 键 词：药品不良反应 信号检测 模糊数 语言变量 贝叶斯置信度递进神经网络 

分 类 号：R911[医药卫生—药学]