Natural Language Processing to Study Epidemiology of Statin Side Effects

描述（由申请人提供）：本申请涉及广泛的挑战领域（10）处理医疗保健数据的信息技术和特定的挑战主题10-LM-101：上市后监督的信息学。本研究的总体目标是开发一个可推广的框架，用于研究叙述性医疗文件中记录的药物副作用。我们将在HMG-CoA还原酶抑制剂（又名HMG-CoA还原酶抑制剂）副作用的流行病学特征的示例中实施和测试该系统。他汀类药物）。他汀类药物是美国最常用的治疗高胆固醇血症的药物。在随机临床试验中，与安慰剂相比，他汀类药物仅与不良反应轻微增加相关，并且不增加治疗中断。然而，在临床实践中，副作用和停药的发生率明显更高，是关键的、可能挽救生命的治疗的主要障碍。例如，据报道，肌痛在临床试验中相对罕见，但被认为在临床实践中更常见。此外，一些其他他汀类药物相关的投诉报告轶事，但没有很好地阐明在临床试验中，包括抑郁症，易怒，记忆丧失等。其中大多数的流行病学特征不明确，其流行率和风险因素仍然未知。结构化电子病历（EMR）和管理数据已被用于研究药物副作用。然而，结构化数据有很大的局限性。它们可能不包含将特定问题与药物联系起来所需的时间或原因信息，并且可能不足以识别特定的不良反应。叙述性EMR数据，例如提供者注释，可以以高粒度级别提供药物和不良事件之间的因果关系的文档。自然语言处理（NLP）是一种新兴技术，可以从叙述性医疗文档中提取信息。在之前的工作中，我们已经成功地应用自然语言处理从叙述性提供者笔记中提取药物信息，包括药物强化，药物不依从和药物停用。我们将利用这些工具和Partners HealthCare广泛的EMR基础设施来开发和测试自然语言处理系统，以研究药物副作用。我们将以他汀类药物不良反应的流行病学研究为例验证该系统。该项目的研究结果将为一个开源系统奠定基础，该系统可用于使用叙述性EMR数据进行药物副作用的上市后监测。 公共卫生相关性（由申请人提供）：他汀类药物（用于治疗高胆固醇的药物）在日常医疗实践（而不是研究）中副作用的频率和风险因素尚不清楚。在本计画中，我们将设计一个系统来分析电子病历中有关他汀类药物副作用的资讯。如果成功，这种方法可以推广到研究许多其他药物的副作用。

DESCRIPTION (provided by applicant): This application addresses broad Challenge Area (10) Information Technology for Processing Health Care Data and specific Challenge Topic 10-LM-101: Informatics for post-marketing surveillance. The overall goal of this study is to develop a generalizable framework for studying medication side effects recorded in narrative medical documents. We will implement and test this system on the example of epidemiologic characterization of side effects of HMG-CoA reductase inhibitors (a.k.a. statins). Statins are the most commonly used class of medications for treatment of hypercholesterolemia in the U.S. In randomized clinical trials statins are associated only with a slight increase in adverse reactions and no increase in discontinuation of treatment compared to placebo. However, in clinical practice the rates of side effects and discontinuation appear significantly higher and represent a major barrier to a critical, potentially lifesaving therapy. For example, myalgias are reported to be relatively rare in clinical trials but are thought to be more common in clinical practice. Additionally, a number of other statin-associated complaints reported anecdotally but not well elucidated in clinical trials include depression, irritability, and memory loss among others. Most of these have been poorly epidemiologically characterized and their prevalence and risk factors remain unknown. Structured electronic medical record (EMR) and administrative data have been used to study medication side effects. However, structured data have important limitations. They may not contain temporal or causative information necessary to link particular problems to medications and may not be sufficiently granular to identify specific adverse reactions. Narrative EMR data, such as provider notes, can provide documentation of causative links between medication and adverse events at high levels of granularity. Natural language processing (NLP) is an emerging technology that enables computational abstraction of information from narrative medical documents. In prior work we have successfully applied natural language processing to abstract medication information from narrative provider notes, including medication intensification, medication non-adherence and medication discontinuation. We will leverage these tools and the extensive EMR infrastructure at Partners HealthCare to develop and test a natural language processing system to study medication side effects. We will validate this system on the example of studying epidemiology of adverse reactions to statins. The findings of this project will lay the foundation for an open-source system that can be used for post-marketing surveillance of medication side effects using narrative EMR data. PUBLIC HEALTH RELEVANCE (provided by applicant): Frequency and risk factors for side effects of statins (medications used to treat high cholesterol) in everyday medical practice (as opposed to research studies) are not known. In this project we will design a system for analyzing the information about statin side effects in the electronic medical records. If successful, this approach can be subsequently generalized to study side effects of many other medications.