"Stopping Rules" for Drug Safety Monitoring Using Healthcare Databases

描述（由申请人提供）：目前FDA上市后药物安全性监测主要依赖于通过MedWatch报告至不良事件报告系统的被动监测。相比之下，使用基于人群的医疗利用数据库进行主动监测可以提高信号检测的有效性和及时性。FDA提出的Sentinel Initiative计划利用医疗保健数据库及其研究环境进行持续的药物安全性监测，并强调了发展主动安全监测能力以生成和确认信号的重要性。然而，在如何实施和利用这些数据方面，还需要学习很多东西。信号是药物和不良医学事件之间强于预期的关联。一种常见的方法是在药物暴露和医学事件数据可用时（例如每月）进行常规累积，并定期评价该累积队列研究。尽管正在进行的药物安全性监测具有巨大的潜在公共卫生效益，但人们对何时应该停止这种前瞻性累积监测活动以警告公众或得出药物安全的结论知之甚少。推迟这样的决定可能会不必要地使患者处于危险之中，但相反，错误地警告风险可能会减少重要药物的使用。由于决策不当，这两种情况都可能导致发病率和/或死亡率增加。基于25年的纵向索赔数据库经验以及与WellPoint/HealthCore的持续密切合作，这是一个收集了14个BlueCross健康计划的集合，记录了3000万投保成员的所有药物分配和医疗保健使用数据，我们建议根据经验评估已发布的停止规则，并开发和测试新的算法来解决这个紧迫的问题。具体而言，我们将：- 识别、表征和构建用于药物安全性监测的经验性示例研究，并开发一组模拟数据源，-针对上述数据场景开发并应用一组“停止规则”，该规则将终止基于流行病学和临床指标以及序贯检验统计的累积队列研究，- 使用决策分析框架，我们将结合联合收割机流行病学停止规则与假阳性和假阴性结果的社会成本信息，-所有测试方法的SAS宏将存放在网页上。这个为期2年的项目将大大增加我们对累积药物安全性监测的理解，这些监测基于经验和模拟数据，并结合决策分析框架，从而开发出增强的数据驱动的停止规则，用于新兴的药物流行病学数据库。

Description (provided by applicant): Current FDA post-marketing drug safety monitoring relies principally on passive surveillance via MedWatch reports to the Adverse Event Report System. By contrast, active surveillance using population-based healthcare utilization databases could improve the validity and timeliness of signal detection. FDA's proposed Sentinel Initiative plans to utilize healthcare databases and their research environments for ongoing drug safety monitoring and highlights the importance of developing the capability to conduct active safety surveillance for signal generation and confirmation. However, much needs to be learned about how to implement and utilize such data. Signals are stronger-than-expected associations between a drug and adverse medical events. One common approach is to routinely accumulate drug exposure and medical event data as they become available (e.g. monthly) and regularly evaluate this cumulative cohort study. Despite the enormous potential public health benefit of ongoing drug safety monitoring, there is little understanding of when such a prospective cumulative monitoring activity should be stopped to warn the public, or to conclude that a medication is safe. Delaying such a decision could unnecessarily put patients at risk, but, conversely, falsely warning of a risk may reduce use of important medications. Both scenarios may result in increased morbidity and/or mortality as a consequence of inadequate decision making. Based on 25 years of experience with longitudinal claims databases and an ongoing close collaboration with WellPoint/HealthCore, a collection of 14 BlueCross health plans that records data on all drug dispensing and healthcare utilization for 30 million insured members, we propose to empirically evaluate published stopping rules and develop and test new algorithms to address this pressing question. Specifically, we will: --- Identify, characterize, and build empirical example studies for drug safety monitoring and develop a set of simulated data sources, --- Develop and apply a set of "stopping rules" for the data scenarios identified above that would terminate a cumulative cohort study based on epidemiologic and clinical metrics and sequential test statistics, --- Using a decision analytic framework we will combine the epidemiologic stopping rules above with information on the societal cost of false positive and false negative findings, --- SAS macros for all tested approaches will be deposited on a web-page. This 2-year project will greatly increase our understanding of cumulative drug safety monitoring based on empirical and simulated data and in combination with a decision analytic framework, leading to the development of enhanced, data-driven stopping rules for use with emerging pharmacoepidemiologic databases.