药品不良反应监测中多重性检验和低频数膨胀问题的研究

Adverse drug reaction (ADR) monitoring is a key factor to ensure the safety use of drugs, and the surveillance is usually achieved by the process of ADR signal detection. We have conducted a series of studies about the methods for ADR signal detection over the past ten years. During the gradual further research process, we observed that the detection results were influenced greatly by two problems: Firstly, routine methods ignored the multiple hypotheses testing and led to high false positive rates of the detection results; Secondly, the spontaneous report database represented the characteristic of low frequency inflated, that was the proportion of reports whose report numbers were small (less than 3) was quite high in spontaneous report database, and routine methods detected no or few signals from these reports, and few of them were really true signals even if some potential signals were detected from them. To solve these issues, we will develop a procedure of Bayesian false discovery rate (FDR) based on prior information and characteristics of the national spontaneous report data, and on this basis to build exact unconditional test based on aforementioned FDR for signal detection. In addition, we will apply this model both in the simulation data and the actual data to validate and evaluate the method. In the end, this method will be combined with the ADR signal detection system and big data processing platform and applied to daily surveillance. This model is expected to solve the aforementioned two problems and improve the power to detect signals. The research will help to improve the drug risk management, so as to ensure the safety of drug use.

药品不良反应监测是保障用药安全的重要手段，其实施主要通过信号检测完成。课题组进行了十余年信号检测方法的研究，发现常规方法存在两个突出问题：①忽略多重性检验，导致检测结果中假阳性信号比例偏高；②现有数据呈现低频数组合膨胀特点，即报告例数少的组合所占比例很高，常规方法对该部分组合的信号检出率低且可靠性差。本研究拟以国家自发呈报系统为依托，针对不良反应复杂数据特点，充分融合先验信息，建立贝叶斯错误发现率（FDR）估计方法，进一步构建基于FDR控制的确切非条件检验模型用于信号检测。并以实际数据为背景产生模拟数据，对方法进行评价，再将方法应用于实际数据考核验证。最后研发程序实现所建方法，并与已建立检测系统及大数据平台对接提高计算效率，以便推广应用。实现在有效减少假阳性信号的同时，提高低频数组合的信号检出率和稳定性，切实提高信号检测的及时性、准确性和可靠性。从而提高药品风险管理水平、保障居民用药安全。

药品安全直接关系到人类的身体健康与生命安全，一直是公共卫生领域研究的重点问题。一系列触目惊心的药害事件和数字警示我们需要尽早发现药害事件、及时处理、减少危害。药品不良反应监测是保障用药安全的重要手段，其实施主要通过信号检测完成。但是常规信号检测方法存在两个突出问题：①忽略多重性检验，导致检测结果中假阳性信号比例偏高；②自发呈报系统呈现低频数组合膨胀特点，即报告例数少的组合所占比例很高，常规方法对该部分组合的信号检出率低且可靠性差。本研究以国内外自发呈报系统报告数据库为依托，通过对自发呈报系统报告数据中药品名称及不良反应名称的规范化整理，建立了标准的可用于不良反应记录查询及信号挖掘的不良反应数据库，并分析了近年来报告的基本特征。针对不良反应复杂数据特点，充分融合先验信息，建立了贝叶斯错误发现率（FDR）控制和估计方法，进一步构建基于FDR控制的确切非条件检验模型用于信号检测。并以实际数据为背景产生模拟数据，对方法进行评价，发现在自发呈报系统药品不良反应信号检测中，采用基于FDR控制的确切非条件检验信号检测方法在一定程度上降低了假阳性错误，提高了整体的信号检测效能，尤其在报告例数较少时的信号检出率和可靠性均有一定程度提高。此外，随着阈值的增加各种信号检测方法的结果均有不同的变化，若维持较优的总体效能，则较小的阈值范围如0.01-0.10是可供选择范围，其中0.01、0.025、0.05和0.10相对较为常见。最后研发程序实现所建方法。并紧密结合当前新冠肺炎用药的安全性问题将其应用到新冠肺炎用药尤其是磷酸氯喹的安全性研究中，可以有效提早发现磷酸氯喹导致的心脏毒性，为新冠肺炎用药的安全合理用药提供参考。.课题组已发表课题相关论文13篇，其中，第一标注论文9篇，第二标注论文4篇；核心及统计源期刊论文9篇，SCI 论文4 篇。完成博士论文2篇及硕士论文1篇。