Leveraging artificial intelligence methods and electronic health records for pediatric pharmacovigilance

项目摘要 美国食品和药物管理局的一个首要目标是有效实施上市后 已批准药物的药物警戒能力。实现儿科人群的这一目标是 特别具有挑战性。例如，目前对药物的安全性、风险、药物相互作用和 怀孕期间使用的许多药物的致畸作用，由于对 孕妇参与药物开发试验。此外，许多药物的安全性和有效性 由于缺乏对儿童的临床试验，儿科使用很少。因此，儿科实践经常 涉及“标签外”使用具有未知副作用的药物。这可能会造成不可预测的悲剧性影响， 儿科患者，包括可能影响其发育的严重药物不良反应和毒性， 未来的生殖能力。大量真实世界医疗保健数据的可用性，例如电子 健康档案（EHR）提供了一个机会，以满足有效调查的影响， 大规模的儿科人群的药物暴露。我们的目标是进行药物和表型范围的 在一个大型的EHR母子二人数据库上进行的关联研究，这将使我们能够研究不良的儿科疾病， 与1）母亲在怀孕期间和怀孕前的药物和物质使用暴露相关的结果;以及 2)儿童在其所有发育阶段的药物暴露。次要分析将包括 母亲的药物使用暴露与儿科结局之间的关联，广泛的药物相互作用 关联研究和药物-物质使用相互作用广泛关联研究。此外，我们将利用 人工智能方法，如自然语言处理（NLP）和机器学习， 暴露错误分类，并改善拟议研究的儿科结局识别。我们的项目 目的是：1）进行高通量药物流行病学研究，以确定不良儿科结局， 和2）使用利益相关者评估实时儿科药物警戒系统的临床效用 参与战略。这一建议的预期成果是提供一个知情的监测工具， 实时监测儿童药物不良反应。这将为部署临床 用于儿科人群中药物不良反应早期检测和实时监测的决策支持系统 识别有这种负面结果风险的患者。

PROJECT SUMMARY One overarching goal of the US Food and Drug Administration is to effectively implement post-market pharmacovigilance capabilities of already approved medications. Achieving this goal for pediatric population is particularly challenging. For example, little is currently known about the safety, risks, drug-interactions, and teratogenic effects of many drugs used during pregnancy due to the strict regulations imposed for the participation of pregnant women in drug development trials. Further, safety and efficacy of many drugs for pediatric use is scarce due to the lack of clinical trials on children. For this reason, pediatric practice often involves “off-label” use of drugs with unknown side effects. This may cause unpredictable and tragic effects in pediatric patients including severe adverse drug reactions and toxicity that can affect their development and future reproductive capacity. The availability of large volumes of real-world healthcare data such as electronic health records (EHRs) provides an opportunity to meet the critical need of effectively investigating the effect of drug exposures on pediatric populations at large scale. Our goal is to conduct drug- and phenome-wide association studies on a large EHR database of mother-child dyads that will allow us to study adverse pediatric outcomes associated with 1) drug and substance use exposures of mothers during and before pregnancy; and 2) drug exposures of children during all their developmental milestones. Secondary analyses will include associations between substance use exposure of mothers and pediatric outcomes, drug-drug interaction wide association studies, and drug-substance use interaction wide association studies. Further, we will leverage artificial intelligence methods such as natural language processing (NLP) and machine learning to address exposure misclassification and improve pediatric outcome identification for the proposed studies. Our project aims are to: 1) conduct high-throughput pharmacoepidemiologic studies to identify adverse pediatric outcomes, and 2) evaluate the clinical utility of a real-time pediatric pharmacovigilance system using stakeholder engagement strategies. The expected outcome of this proposal is a stakeholder-informed tool to monitor adverse drug reactions of children in real-time. This will pave the way towards the deployment of a clinical decision support system for early detection of adverse drug reactions in pediatric populations and for real-time identification of patients who are at risk of such negative outcomes.