Automated Detection of Medical Errors

自动检测医疗错误

• 批准号: 7124706
• 负责人: PETER D STETSON
• 金额: $ 13.5万
• 依托单位: COLUMBIA UNIVERSITY HEALTH SCIENCES
• 依托单位国家: 美国
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-09-30 至 2008-09-29
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/7124706
• 关键词: automated medical record system; behavioral /social science research tag; biomedical automation; clinical research; health care quality; health services research tag; human data; information retrieval; medical records; method development; patient care management; patient safety /medical error

DESCRIPTION: The long-term goal of this proposal is to use the electronic medical record, including narrative text, to understand and encode the process of care for individual patients in order to improve patient safety. Achieving this goal has the potential to help detect adverse events, and to differentiate medical errors from appropriately tailored care. The specific aims for this proposal are as follows: 1) To understand and encode the process of care for individual patients using data in the electronic medical record, including narrative text. 2) To use a more detailed understanding of patients' processes of care to improve automated adverse event detection. 3) To match processes of care for individual patients against accepted care pathways in order to identify discrepancies. We will capitalize on three core technologies that are in active use by clinicians and researchers in our busy clinical setting: 1) a Web-based clinical information system and its associated clinical data repository (WebCIS), 2) a full medical language parser (MedLEE), and 3) a semi-structured, electronic physician documentation system built by the applicant specifically to support this project (eNote). Methods will include evaluating the performance (sensitivity, specificity and positive predictive value) of our system, DETER+MINE (DETecting ERrors Mining Narrative Electronically), to model the care process and detect adverse events and pathway deviations. We will utilize explicit process criteria and manual, retrospective chart review as a gold standard. This research is intended to provide proof of concept that combining natural language processing of clinical narrative with traditional sources of coded data is required for effective screening with automated defection systems. This approach has the potential to impact significantly on our ability to detect and investigate medical errors, adverse medical events, and pathway deviations by reducing reliance on costly and slow manual chart reviews.

描述： 该提案的长期目标是使用包括叙事文本在内的电子病历来理解和编码单个患者的护理过程，以提高患者的安全性。 实现这一目标有可能帮助检测不良事件，并将医疗错误与适当量身定制的护理区分开。该提案的具体目的如下：1）使用电子病历中的数据（包括叙事文本）理解和编码为个别患者的护理过程。 2）使用对患者护理过程的更详细的理解以改善自动不良事件检测。 3）匹配个别患者的护理过程，以与公认的护理途径相匹配，以确定差异。我们将利用三种在繁忙的临床环境中被临床医生和研究人员积极使用的核心技术：1）基于Web的临床信息系统及其相关的临床数据存储库（WebCIS），2）一个完整的医学语言Parser（Medlee）和3）半结构化的电子物理学文档系统，由应用程序专门支持该项目（启用）。 方法将包括评估系统的性能（敏感性，特异性和积极的预测价值），Deter+Mine（以电子方式检测误差叙事），以对护理过程进行建模并检测不良事件和途径偏差。我们将利用明确的过程标准和手册，回顾性图表审查作为黄金标准。 这项研究旨在提供概念证明，即将临床叙事的自然语言处理与传统的编码数据来源相结合才能通过自动损失系统进行有效筛选。这种方法有可能对我们检测和调查医疗错误，不良医疗事件和途径偏差的能力产生重大影响，从而减少了对昂贵和缓慢的手动图表审查的依赖。