Understanding and Addressing Disparities in Triage and Disposition Decisions in the Emergency Department

摘要 美国每年有近1.5亿急诊科(ED)就诊。病人 向急救人员介绍各种健康问题和敏锐性，从危及生命的紧急情况到 处于非卧床状态。此外，ED患者群体在人口统计方面具有高度多样性， 文化和社会经济因素。在数量大、时间有限和多样化的环境中提供公平的服务 ED设置已被证明是一项挑战，越来越多的证据表明临床决策和 为少数族裔和妇女提供保健服务。分类和处置决策涉及到一些 主观性强，特别容易产生偏见。电子健康档案(EHR)数据初步分析 从一个单一的学术ED中发现了在ED分类、房间优先顺序和医院方面存在差异的证据 录取决定。我们的长期研究目标是充分描述性别、种族和民族差异 在美国的ED分类和处置决策中，并应用统计和机器学习 开发和评估减少这些差距的创新解决方案的方法。我们的研究将是 嵌入到集成了科学证据、内部数据和利益相关者的学习健康系统中 参与改善教育署提供医疗服务的公平性。作为第一步，我们将获得并分析 回顾一个大型医疗系统中10个不同急诊室的电子病历数据。我们的目标是：(1)识别患者 急诊决策中的性别、种族和民族差异(分诊级别分配、房间优先顺序和医院 入院)，并确定ED的操作条件(例如，容量、等待时间)是否会加剧这些 差异；以及(2)开发一个原型机器学习模型，将患者和ED级别的数据集成到 预测教育署可能出现的不公平决策。在这项试验计划成功完成后，我们 将获得必要的初步证据来充分开发一种新的机器学习预测模型 并在多学习健康系统中对模型进行了验证。在未来的研究中，我们还打算调查 不公平判决的机器学习模型的潜在应用，例如看护点 提醒教育提供者的工具和为教育提供者提供的数据监测和报告反馈系统 管理人员和卫生系统领导人。这项研究的发现有可能导致创新 数据驱动的解决方案，为每年向急诊室提供的数百万人促进公平的以患者为中心的护理。

Abstract There are nearly 150 million emergency department (ED) visits in the United States each year. Patients present to EDs for a wide range of health problems and acuities from life-threatening emergencies to ambulatory conditions. Further, the ED patient population is highly diverse with respect to demographics, cultural, and socioeconomic factors. Providing equitable care in the high-volume, time-constrained, and diverse ED setting has proven to be a challenge with growing evidence of disparities in clinical decision making and health care delivery for racial and ethnic minorities and women. Triage and disposition decisions involve some subjectivity and are especially prone to bias. Our preliminary analyses of electronic health record (EHR) data from a single academic ED found evidence of disparities in ED triage, prioritization for rooming, and hospital admission decisions. Our long-term research goal is to fully characterize gender, racial, and ethnic disparities in ED triage and disposition decisions in the United States and to apply statistical and machine learning methods to develop and evaluate innovative solutions to mitigate these disparities. Our research will be embedded within a Learning Health System that integrates scientific evidence, internal data, and stakeholder engagement to improve equity of healthcare delivery in the ED. As an initial step, we will obtain and analyze retrospective EHR data from 10 diverse EDs across a large health system. Our aims are to: (1) identify patient gender, racial, and ethnic disparities in ED decisions (triage level assignment, rooming priority, and hospital admission) and determine whether ED operating conditions (e.g., volumes, wait times) exacerbate these disparities; and (2) develop a prototype machine learning model that integrates patient- and ED-level data to predict potentially inequitable decision making in the ED. Upon successful completion of this pilot project, we will have obtained essential preliminary evidence to fully develop a novel machine learning prediction model and validate the model in multiple Learning Health Systems. In future research, we also intend to investigate potential applications of the machine learning model of inequitable deicison making, such as a point-of-care tool to alert ED providers and a data monitoring and reporting feedback system for ED providers and administrators and health system leaders. Findings from this research has the potential to lead to innovative data-driven solutions to promote equitable patient-centered care for the millions who present to EDs each year.