Safety Promotion through Early Event Detection in the Elderly (SPEEDe)

摘要 不良事件（AE）-医疗护理对患者造成的伤害-影响多达 13.5%的住院患者;这些AE中有一半是可以预防的，AE特别影响患者的健康。 老人众所周知，AE很难准确测量。各种纸张和电子 已经开发了触发工具来识别AE;但是，其阳性预测值（PPV） 低，需要后续的时间密集型手动病历审查，以准确测量AE。 在拟议的项目中，我们将使用创新的，最先进的机器交互学习 (IML)改进现有AE触发器的技术，大大提高其准确性。我们将 我还开发了一个新的AE浏览器，以加快审查可能的AE，以及创新的 一套预测分析工具和方法来测量和检测它们。我们的方法 结合并比较专家驱动的改进与最新的IML技术， 使触发器更准确，最终目标是创建准确的触发器 足以作为实际衡量危害的替代品。我们称之为安全 通过老年人早期事件检测促进，或SPEEDe。 我们的团队在机器学习、患者安全、风险管理、不良事件检测、 老年医学和触发工具专家将共同努力，以实现这一具体目标， 项目：（1）原型化并快速部署触发审查仪表板（不良事件 资源管理器）使用以用户为中心的设计过程，（2）开发和评估新的互动 采用机器学习方法进行更高效、更准确的不良事件图表审查， 触发器优化，以及（3）将交互式机器学习集成到不良事件中 探索并在临床环境中进行前瞻性评估。

ABSTRACT Adverse events (AEs) – harm to patients that results from medical care – affect as many as 13.5% of hospitalized patients; half of these AEs are preventable and AEs particularly affect the elderly. AEs are notoriously difficult to measure accurately. A variety of paper and electronic trigger tools have been developed to identify AEs; however, their positive predictive value (PPV) is low, requiting subsequent, time-intensive manual chart review to accurately measure AEs. In the proposed project, we will use innovative, state-of-the-art machine interactive learning (IML) techniques to refine existing AE triggers, improving their accuracy substantially. We will also develop a novel AE Explorer to speed review of possible AEs, as well as an innovative package of predictive analytics tools and methods to measure and detect them. Our approach combines and compares expert-driven improvement with the most recent IML techniques to make triggers more accurate, with the ultimate goal of creating triggers that are accurate enough to stand in as proxies for actual measurement of harm. We call our approach Safety Promotion through Early Event Detection in the Elderly, or SPEEDe. Our team of accomplished machine learning, patient safety, risk management, AE detection, geriatric medicine and trigger tool experts will work together to carry out the specific aims of this project: (1) prototype and rapidly iterate a trigger review dashboard (the Adverse Event Explorer) using a user-centered design process, (2) develop and evaluate novel Interactive Machine Learning approaches for more efficient and accurate adverse event chart review and trigger refinement, and (3) Integrate Interactive Machine Learning into the Adverse Event Explorer and evaluate it prospectively in a clinical setting.