一种ADR信号检测的改进算法及其在上海市ADR自发呈报数据库上的应用

An improved algorithm for ADR signal detection and its application to Shanghai spontaneous ADR reporting system

目的建立改进的信号检测算法并评估其实际检测效果。方法建立基于新ADR报告(NAR)的信号检测算法;通过理论分析和实际演算比较基于报告数(REP)和基于药物-不良事件组合(DEC)两种算法的ADR检测效果,评估数据库中已知ADR报告对信号检测的影响;并将NAR算法应用于上海市ADR自发呈报数据库。结果①对上海市ADR自发呈报数据库部分数据(2006-2007年)进行了信号检测计算,结果发现基于DEC的算法会产生不合逻辑的虚假信号,而基于REP的算法能避免产生这类信号并能发现被DEC算法忽略的有价值信号;筛除已知ADR报告能提高信号检出率57%。②建立了包含2650种药物,27230种DEC的已知ADR数据库,可自动筛选NAR。③采用基于NAR的信号检测算法对上海市2004-2007年ADR自发呈报数据库进行检测,产生信号380条,包括加替沙星-低血糖反应、头孢他啶-过敏性休克,茵栀黄-荨麻疹等值得关注的信号。结论基于NAR的算法能显著提高ADR信号的检测效率。

AIM To examine the efficiency of current algorithms for adverse drug reaction （ADR） signal detection and put forward an improved algorithm. METHODS An algorithm based on new ADR reports （termed as NAR algorithm） was worked out and applied to the ADR report data （during year 2004- 2007） of Shanghai spontaneous reporting system （SRS）. Two algorithms based on drug-event combination （DEC） and ADR reports [ termed as DEC algorithm and report （REP） algorithm, respectively] was compared by simulation and calculation with real SRS data. The influence of reports carrying only known ADRs on signal detection was evaluated. RESULTS ① The DEC algorithm, but not the REP algorithm, produced obviously illogical signals in the calculation with ADR report data （during year 2006 - 2007） from Shanghai SRS. Moreover, the REP algorithm detected some noteworthy new signals that were missed by the DEC algorithm. Excluding the reports carrying only known ADRs from Shanghai SRS database above greatly improved the signal hit rote by 57%. ②An ADR database containing 2 650 drugs and 27 230 DECs was established, which can be used to pick out new ADR reports from SRS automatically. ③ The NAR algorithm identified 380 signals from Shanghai SRS database （during year 2004 - 2007）, including gatifloxacin-induced hypoglycemia, ceftazidime-induced allergic shock and Yinzhihuang （traditional Chinese medicine） -induced urticaria. CONCLUSION The NAR algorithm can significantly improve the efficiency of ADR signal detection.