Utility of Predictive Systems to identify Inpatient Diagnostic Errors: The UPSIDE Study

项目总结/摘要 虽然自医学研究所发表《为了 在2000年的“错误是人的”中，对医院诊断错误的研究相对缺乏。 拟议研究的广泛，长期目标是更好地了解发病率，原因， 住院患者诊断错误的危险因素。这项工作将提供基础研究， 制定干预措施，以减少这些错误，包括预测工具，干预目标， 未来干预试验的结果评估方法。为了实现这一总体目标，我们将 具体目标如下：1）确定死亡患者中诊断错误的发生率， 住院或在入院后两天或更长时间内通过 患者记录的结构化、标准化裁定流程，2）将联合收割机裁定数据与数据 确定哪些特定因素会导致诊断错误的风险，并使用风险 估计以计算导致这些错误的因素的发生率和影响，以及3）创建机器- 学习模型可用于回顾性识别可能出现诊断错误的患者 已经发生了。这项研究将对2000名普通科住院的患者进行回顾性评估。 20家美国医院的医学部门参与了国家研究合作，并做出了贡献 数据提供给基准和采购组织（Vizient）。使用更安全的诊断（Safer-Dx）和 诊断错误评估和研究（DEER）分类工具，都适用于住院设置， 裁决者将审查电子病历数据，并确定是否存在诊断性 使用严格的培训和持续审查流程，确保各研究中心、裁定者 和时间将使用标准建模技术来了解以下各项的人群归因风险： DEER过程失败指向诊断错误以及几个患者、提供者和 系统级风险因素。最后，先进的机器学习方法将用于创建模型， 识别发生诊断错误的患者，其性能上级标准方法，例如 逻辑回归总之，这些方法将提供一个广泛的和有代表性的情况， 在遭受伤害的住院患者中诊断错误，建立患者模型， 基于系统的因素，使诊断错误或多或少的可能性，并建立先进的，高效的， 需要可扩展的工具来支持各种机构的未来监督和改进计划。 这项研究将建立一个基础，从医疗保健系统可以评估和实现卓越的， 诊断在住院设置。

PROJECT SUMMARY/ABSTRACT While much research has been conducted on patient safety since the Institute of Medicine published “To Err is Human” in 2000, there is a comparative dearth of research on diagnostic errors in the hospital setting. The broad, long-term objectives of the proposed research is to better understand the incidence, causes, and risk factors for diagnostic errors in the inpatient setting. This work will provide foundational research for the development of interventions to reduce these errors, including predictive tools, targets for intervention, and a methodology for outcome assessment in future trials of interventions. To achieve this overall goal, we will carry out the following specific aims: 1) To determine the incidence of diagnostic errors among patients who die in hospital or are transferred to the ICU two days or more after admission to a general medicine service through a structured, standardized adjudication process of patient records, 2) To combine adjudication data with data from Vizient to determine which specific factors contribute to risks for diagnostic errors, and to use risk estimates to calculate incidence and impact of factors contributing to those errors, and 3)To create machine- learning models that can be used to retrospectively identify patients in whom a diagnostic error was likely to have taken place. The research will involve a retrospective evaluation of 2000 patients admitted to general medicine units at 20 US hospitals participating in a national research collaborative and which also contribute data to a benchmarking and purchasing organization (Vizient). Using the Safer-Diagnosis (Safer-Dx) and Diagnostic Error Evaluation and Research (DEER) taxonomy tools, both adapted for the inpatient setting, adjudicators will review electronic medical record data and determine the presence or absence of diagnostic errors using a rigorous training and continuous review process to ensure reliability across sites, adjudicators, and time. Standard modelling techniques will be used to understand the population-attributable risk of each of the DEER process failure points to diagnostic error as well as the contributions of several patient, provider, and system-level risk factors. Lastly, advanced machine-learning methods will be used to create models that can identify patients in whom diagnostic error occurred, with superior performance to standard approaches such as logistic regression. Together, these approaches will provide a broad and representative picture of the incidence of diagnostic errors among hospitalized patients who have suffered harm, develop models of patient and system-based factors that make a diagnostic error more or less likely, and build advanced, efficient, and scalable tools needed to support future surveillance and improvement programs for a variety of institutions. This research will establish a foundation from which healthcare systems can assess and achieve excellence in diagnosis in the inpatient setting.