Discovering and Applying Knowledge in Clinical Databases

描述(由申请人提供)：我们的项目正在进行的目标是“在临床数据库中发现和应用知识”，是开发和应用方法来开发电子病历数据用于决策支持，重点是叙述性数据。自1994年作为R29项目开始以来，我们一直在开发方法来准备原始电子病历数据，应用和评估自然语言处理，开发包括机器学习在内的数据挖掘技术，并将结果用于临床护理和研究。 在这一竞争的继续中，我们建议解决电子病历中的时间信息，并与纽约市卫生和精神卫生部门(NYC DOHMH)合作，将自然语言处理和时间处理应用于综合征监测任务。 我们已经开始了时间处理系统的工作。它提取叙事报告中陈述的时态断言，使用MedLEE自然语言处理器解析非时态信息，基于知识库推断隐含的时态断言，并以简单的时间约束满足问题的形式产生信息。后者可以用来回答有关事件的时间和事件对之间的时间关系的问题。我们建议完善系统，扩展知识库，提高计算速度，解决时态断言的不确定性，从结构化数据中整合时态信息，并对系统进行评估。 纽约市DOHMH拥有一个成熟的综合症监测系统，可以监测近800万人，它拥有尚未开发的叙述性和结构化电子医疗记录形式的数据源。我们建议应用自然语言处理和我们提出的时间处理来将数据转换为适合监视的形式。我们将评估结构化数据、叙事数据和经过时间处理的叙事数据的增量收益。

DESCRIPTION (provided by applicant): The ongoing goal of our project, "Discovering and applying knowledge in clinical databases," is to develop and apply methods to exploit electronic medical record data for decision support, with an emphasis on narrative data. Since the inception of our project as an R29 in 1994, we have been developing methods for preparing raw electronic medical record data, applying and evaluating natural language processing, developing data mining techniques including machine learning, and putting the results to use for clinical care and research. In this competing continuation, we propose to address the temporal information in the electronic medical record and to apply natural language processing and temporal processing to the task of syndromic surveillance in collaboration with the New York City Department of Health and Mental Hygiene (NYC DOHMH). We have begun work on a temporal processing system. It extracts temporal assertions stated in narrative reports, uses the MedLEE natural language processor to parse the non-temporal information, infers implicit temporal assertions based on a knowledge base, and produces the information in the form of a simple temporal constraint satisfaction problem. The latter can be used to answer questions about the time of events and the temporal relation between pairs of events. We propose to complete the system, expand the knowledge base, speed computation, address the uncertainty of temporal assertions, incorporate temporal information from structured data, and evaluate the system. NYC DOHMH has a mature syndromic surveillance system that watches over almost eight million persons, and it has as-yet unexploited data sources in the form of narrative and structured electronic medical records. We propose to apply natural language processing and our proposed temporal processing to convert the data to a form appropriate for surveillance. We will evaluate the incremental benefit of structured data, narrative data, and temporally processed narrative data.