Patient Safety Event Surveillance Using Machine Learning and Free Text Clinical Notes

项目摘要/摘要 拟议的项目旨在通过收集以患者为中心的 以结果为输入数据支持学习健康的安全与改进模式 系统(LHS)。该项目将通过利用现有的机器学习来实现这些目标 方法对自由文本文档(如临床医生笔记)进行分类，以确定是否存在 感兴趣的特定事件。该项目有两个长期目标，共同关注这一点。 第一个广泛的项目目标是收集重要数据，以填补 5例危重儿科住院患者获得性保健的发生率及临床流行病学 条件(HAC)。这5种HAC是：外周静脉浸润性、静脉血栓栓塞症 (VTE)、压力损伤、病人跌倒，以及涉及伤害提供者的事件。 第二个目标是评估一种新的常规患者安全事件监测方法 可扩展、可转移、可适应其他条件和设置，且成本较低 可持续的持续运营。该项目有两个具体目标来实现这些目标： 目的1：加强对5家儿科HAC的监测。比较特征 以前和新发现的病例。描述高危人群。 目标2：评估现有系统的完备性。评估增强版的效果 监督质量改进活动；HAC的发生率；以及运营成本 系统，包括工作人员的时间和资源。 项目团队已经开发了一个以开放许可实现的机器学习界面 Windows软件。该团队有一个漫长的记录，使这些方法可以 研究、临床操作、质量改进和损伤方面的临床医生和非专业用户 预防设置。目前的项目提出了这些技术的创新应用 患者安全监护这一重要问题的技术、方法和工具。 这个项目的预期结果将是5个国家中的每一个的知识都有了很大的进步 建议加强监测的儿科HAC。结果将以现有的 数据完整性和临床流行病学。调查结果将直接解决人们对 现有数据来源的局限性，从而推动患者安全改进活动。 另一个预期结果将是对患者的新方法进行严格评估 安全监控。这将包括分析加强监测的成本和效益。 机器学习与当前方法的对比，以及该方法的成本效益 相比于对现有数据的依赖和合作伙伴社区医院的外部验证。

PROJECT SUMMARY/ABSTRACT The proposed project aims to make healthcare safer through collection of patient-centered outcomes as the input data to support a safety and improvement model of the Learning Health System (LHS). The project will accomplish these aims by leveraging existing machine learning methods to classify free text documents, such as clinician notes, for the presence or absence of specific events of interest. The project shares this focus with two long-term objectives. The first broad project goal is to collect important data to address knowledge gaps in the incidence and clinical epidemiology of 5 serious pediatric inpatient healthcare acquired conditions (HACs). These 5 HACs are: peripheral IV infiltrates, venous thromboembolisms (VTEs), pressure injuries, patient falls, and incidents involving harm to providers. The second goal is to evaluate a novel approach to routine patient safety event surveillance that is scalable, transferrable, adaptable to other conditions and settings, and with low cost of sustainable ongoing operation. The project has two specific aims to achieve these goals: Aim 1: Implement enhanced surveillance for 5 pediatric HACs. Compare characteristics of previously and newly identified cases. Describe high-risk populations. Aim 2: Estimate completeness of existing systems. Evaluate effects of enhanced surveillance on quality improvement activities; incidence of HACs; and cost to operate system, including staff time and resources. The project team has developed a machine learning interface implemented in open license Windows software. The team has a lengthy track record making these methods accessible to clinicians and lay users in research, clinical operations, quality improvement, and injury prevention settings. The current project proposes an innovative application of these technologies, methods, and tools to the important problem of patient safety surveillance. An expected outcome of this project will be substantial advance in knowledge for each of the 5 pediatric HACs proposed for enhanced surveillance. Results will be reported in terms of existing data completeness and clinical epidemiology. Findings will directly address concerns over limitations of existing data sources and thereby drive patient safety improvement activities. An additional expected outcome will be the rigorous evaluation of a novel approach to patient safety surveillance. This will include analysis of the costs and benefits of enhanced surveillance with machine learning versus current approaches, and the cost-effectiveness of the approach compared to reliance on existing data, and external validation at a partner community hospital.