基于H省药品不良反应真实世界数据对信号检测方法的比较研究

Comparative Study on Signal Detection Methods Based on Real World Data of Adverse Drug Reactions in H Province

目的:比较3种信号检测方法对H省药品不良反应(adverse drug reaction,ADR)监测系统数据进行信号检测的灵敏度差异,以期探索适用于我国ADR数据库的方法。方法:对2016年1月至2020年12月H省ADR数据库20个季度的ADR数据按季度分组,并依次累加处理,分别使用报告比值比(reporting odds ratio,ROR)法、英国药品和保健产品管理局(Medicines and Healthcare Products Regulatory Agency,MHRA)综合标准法、贝叶斯置信传播神经网络(Bayesian confidence propagation neural network,BCPNN)法,对各个累加季度中西咪替丁注射制剂、阿米卡星注射剂等国家药监局于2021~2022年公示的10种修订说明书的药品进行信号检测,比较采用3种方法发现的20个累加季度后的ADR信号数量、各个信号最早被检测出来的时间,以及明确各个信号被最早检测出时所用的方法。结果:经数据处理后得到有效报告569843份。采用ROR法检测出的信号数量最多,其检测出的信号总数量在药品说明书中平均占比29.79%;采用MHRA法与BCPNN法检测出的信号总数量在药品说明书中平均占比均为24.24%。采用这3种方法检测出的新信号数量平均占比分别为21.92%、21.85%、18.70%。采用3种方法检测产生时间差异的信号数量平均占比59.15%,没有产生时间差异的信号数量平均占比40.85%。ROR法是最早检测出信号的方法,MHRA法次之,BCPNN法最晚。结论:在上述3种方法中,采用ROR法检测出的总信号数量与新信号数量最多,检测出信号的时间最早,且灵敏度最高,因此ROR法可能更适用于H省现阶段ADR信号检测工作。

Objective:This study aims to compare the sensitivity differences of three signal detection methods in detecting adverse drug reactions(ADR)data from the monitoring system in H Province.The goal is to explore a method suitable for China's adverse reaction database.Methods:ADR data of adverse reaction database of H Province from January 2016 to December 2020 in 20 quarters are grouped by quarters and sequentially accumulated.The reporting odds ratio(ROR)method,Medicines and Healthcare Products Regulatory Agency(MHRA)method,and Bayesian confidence propagation neural network(BCPNN)method are used to detect signals for ten drugs in each cumulative quarter,including Cimetidine injection and Amikacin injection,which have revised instructions published by the National Medical Products Administration in 2021~2022.The total number of detected signals,the earliest detected time for each signal,and the earliest detected method are compared after 20 cumulative quarters for each method.Results:After data processing,569843 effective reports are obtained.The ROR method detects the highest number of signals,with an average proportion of 29.79%of the total signals detected in the manual.The MHRA method and BCPNN method have an average proportion of 24.24%each.The average proportion of new signals detected by the three methods is 21.92%,21.85%and 18.70%,respectively.The average ratio of signals generated by the three methods with time difference is 59.15%,while signals without time differences accounts for 40.85%.The ROR method is the earliest in detecting signals,followed by the MHRA method and BCPNN method.Conclusion:Among the three methods,the ROR method has the largest number of detected signals and the new signals,the earliest detection time,and the highest sensitivity.Therefore,the ROR method may be more suitable for the current ADR detection work in H Province.