Real-time Disambiguation of Abbreviations in Clinical Notes

描述(由申请人提供)：提供高质量医疗保健的一个关键前提是在医疗保健环境内和之间进行有效的沟通。然而，临床笔记中普遍使用缩略语可能会阻碍交流。临床医生在记录过程中使用缩写以节省时间。虽然缩略语对作者来说似乎毫不含糊，但它们经常会使其他读者感到困惑，包括医疗保健提供者、患者和试图从文本中提取临床术语的自然语言处理(NLP)系统。虽然缩写会导致错误的认识很普遍，但很少有人为这个重要问题部署实用的解决方案。拟议的项目将开发、评估和分享临床缩写识别和消除歧义的系统方法，这样做的主要目的是使现有的NLP系统受益，并通过实时减少电子记录中的歧义来改进基于计算机的文件系统。这项研究包括以下五个具体目标：1)开发从临床文本语料库中检测缩略语及其意义的自动化方法，并建立一个全面的临床缩略语知识库；2)开发和评估三个自动词义消歧(WSD)分类器，并建立将这些分类器组合在一起的方法，以最大化它们的性能和覆盖率；3)开发CARD系统，并通过将其与两个已建立的NLP系统(MedLEE和KnowledgeMap)集成来展示其有效性；4)将CARD与机构临床文档系统(Vanderbilt‘s StarNotes)集成，并评估其在临床医生生成记录时实时扩展缩略语的能力；5)将卡片知识库和软件分发给非商业用途。

DESCRIPTION (provided by applicant): A key prerequisite for high-quality healthcare delivery is effective communication within and across healthcare settings. However, communication can be hampered by the pervasive use of abbreviations in clinical notes. Clinicians use abbreviations to save time during documentation. While abbreviations may seem unambiguous to their authors, they often cause confusion to other readers, including healthcare providers, patients, and natural language processing (NLP) systems attempting to extract clinical terms from text. While the understanding that abbreviations can cause errors is widespread, few have deployed pragmatic solutions for this important problem. The proposed project will develop, evaluate, and share a systematic approach to Clinical Abbreviation Recognition and Disambiguation (CARD), and in doing so substantially aims to benefit existing NLP systems and to improve computer-based documentation systems by reducing ambiguities in electronic records in real-time. The study includes the following five Specific Aims: 1) Develop automated methods to detect abbreviations and their senses from clinical text corpora and build a comprehensive knowledge base of clinical abbreviations; 2) Develop and evaluate three automated word sense disambiguation (WSD) classifiers, and establish methods to combine those classifiers to maximize both their performance and coverage; 3) Develop the CARD system, and demonstrate its effectiveness by integrating it with two established NLP systems (MedLEE and KnowledgeMap); 4) Integrate CARD with an institutional clinical documentation system (Vanderbilt's StarNotes) and evaluate its ability to expand abbreviations in real-time as clinicians generate records; 5) Distribute the CARD knowledge base and software for non-commercial uses.