Pharmacovigilance Data Mining With Methods Based on False Discovery Rates: A Comparative Simulation Study

Pharmacovigilance Data Mining With Methods Based on False Discovery Rates: A Comparative Simulation Study
复制标题

DOI:
10.1038/clpt.2010.111
复制
发表时间:
2010-10-01
影响因子:
6.7
通讯作者:
Tubert-Bitter, P.
Tubert-Bitter, P.
中科院分区:
医学2区
文献类型:
--
作者:
Ahmed, I.;Thiessard, F.;Tubert-Bitter, P.

文献摘要

被引文献

相似文献

早期发现已上市药物引起的不良反应是药物警戒工作的首要关注点;上市后监测所用的方法旨在根据卫生保健提供者的报告和各种数据库中提供的信息,检测表明潜在安全问题的信号。最近提出了基于错误发现率(FDR)估计的信号检测方法。它们解决了当前使用的自动方法的任意检测阈值的限制,包括美国食品和药物管理局和世界卫生组织乌普萨拉监测中心最近更新的方法。我们使用两个模拟程序来比较三种当前方法的假阳性性能:报告比值比 (ROR)、信息成分 (IC)、伽玛泊松收缩率 (GPS),以及从 GPS 模型和费舍尔检验导出的两种基于 FDR 的方法。 FDR 率的巨大差异与当前使用的信号检测方法有关。在仅限于至少三份报告的信号的分析中,这些差异范围为 0.01% 至 12%。生成的信号数量也变化很大。在固定大小的信号列表中,当使用基于 FDR 的方法时,FDR 会降低。总体而言,两项模拟研究的结果表明,使用基于 FDR 的 GPS 方法有望提高有效性。
The early detection of adverse reactions caused by drugs that are already on the market is the prime concern of pharmacovigilance efforts; the methods in use for postmarketing surveillance are aimed at detecting signals pointing to potential safety concerns, on the basis of reports from health-care providers and from information available in various databases. Signal detection methods based on the estimation of false discovery rate (FDR) have recently been proposed. They address the limitation of arbitrary detection thresholds of the automatic methods in current use, including those last updated by the US Food and Drug Administration and the World Health Organization's Uppsala Monitoring Centre. We used two simulation procedures to compare the false-positive performances for three current methods: the reporting odds ratio (ROR), the information component (IC), the gamma Poisson shrinkage (GPS), and also for two FDR-based methods derived from the GPS model and Fisher's test. Large differences in FDR rates were associated with the signal-detection methods currently in use. These differences ranged from 0.01 to 12% in an analysis that was restricted to signals with at least three reports. The numbers of signals generated were also highly variable. Among fixed-size lists of signals, the FDR was lowered when the FDR-based approaches were used. Overall, the outcomes in both simulation studies suggest that improvement in effectiveness can be expected from use of the FDR-based GPS method.