Pharmacovigilance Methods: Leveraging Heterogeneous Adverse Drug Reaction Data

药物不良反应（ADR）是患者和医疗保健的主要负担， 住院和死亡，并承担巨大的成本。这项建议的长期目标是促进 通过使用自动化方法发现新的严重ADR， 联合收割机从大量不同的患者人群以及从文献中获得的信息。有 在药物警戒方面取得了相当大的进展，但还需要做更多的工作。例如，万络，一种常用的 最近发现，该药导致至少88，000例心肌梗死， 现有方法的不足。迄今为止，各种方法主要依赖于使用单一数据来源， 主要来自联邦食品药品监督管理局不良事件报告系统（FAERS）和 电子健康记录（EHRS）。虽然重要，但每个来源都有不同的局限性， 因此，将它们之间的数据结合起来应该会带来更有效的药物安全性 通过增加统计能力，以及允许每个数据源相互补充， 源我们已经开发了与每个单一来源相关的方法，因此， 是一个很好的机会，以建立在我们的研究成果，以推进国家的艺术， 药物警戒 更具体地说，我们将a）从电子健康中心获取并联合收割机综合临床数据， 两个不同的医疗保健网站的记录（EHR），利用自然语言为不同的人群提供服务 处理（NLP），以获得大量的细粒度数据，然后通过开发数据挖掘 基于临床数据检测新ADR信号的方法，B）分析治疗相关风险的差异 两个EHR人群之间的因素，如种族和民族差异，c）检测ADR信号， 使用已建立的方法建立FAERS数据库，d）开发获得ADR信号的改进方法 基于文献中的信息，以及e）开发利用上述来源的结果的方法， 最大化效率。我们将重点关注八个严重的ADR，并收集高质量的参考标准， 这样我们就能够评估和比较不同检测方法的性能 以及组合这些源的联合收割机。 这一建议很好地克服了与现有自动化方法相关的问题， 主要是基于对个人数据来源的使用。我们相信这些方法会有效 因为我们有一个强大的基础设施来建设。最重要的是， 该提案提供了一个极好的机会，可以利用异构数据源来显著提高 患者安全，降低成本。

Adverse drug reactions (ADRs) are a major burden for patients and healthcare, causing preventable hospitalizations and deaths, and incurring a huge cost. The long-term objective of this proposal is to advance patient safety and reduce costs by discovering novel serious ADRs through use of automated methods that combine information from large and varied patient populations as well as from the literature. There have been considerable advances in pharmacovigilance, but more work is needed. For example, Vioxx, a commonly used drug, was recently found to cause at least 88,000 occurrences of myocardial infarction, highlighting the insufficiency of current methods. To date, methods have mainly depended on the use of single sources of data, primarily from the Federal Food and Drug Administration Adverse Event Reporting System (FAERS) and from electronic health records (EHRS). Although important, each of the sources has different limitations and advantages, and therefore, combining the data across them should lead to more effective drug safety surveillance by increasing the statistical power, and also by allowing each data source to complement the other sources. We already have developed methods associated with each of the single sources, and therefore, this is an excellent opportunity to build upon our research accomplishments to advance the state of the art in pharmacovigilance. More specifically, we will a) acquire and combine comprehensive clinical data from the electronic health records (EHRs) of two different health care sites serving diverse populations by utilizing natural language processing (NLP) to obtain vast quantities of fine-grained data, and then by developing data mining methodologies on the clinical data to detect novel ADR signals, b) analyze differences in therapy-related risk factors between the two EHR populations, such as racial and ethnic differences, c) detect ADR signals in the FAERS database using an established methodology, d) develop improved methods to acquire ADR signals based on information in the literature, and e) develop methods that utilize the results from the above sources to maximize effectiveness. We will focus on eight serious ADRs, and collect a high-quality reference standard for those ADRs so that we will be able to evaluate and compare performance of the different detection methods individually as well as the methods that combine the sources. This proposal is well positioned to overcome problems associated with existing automated methods, which are primarily based on use of individual sources of data. We are confident the methods will be effective because a strong infrastructure is in place for us to build upon. Most importantly, the methodology developed in this proposal presents an excellent chance to leverage heterogeneous data sources to dramatically improve patient safety and reduce costs.