Enhancing information retrieval in electronic health records through collaborative filtering

项目摘要 当我们考虑在亚马逊网站上购买一本书时，我们通常会受益于一个名为 “顾客们也看了。”这些推荐是由一种称为协作过滤(CF)的方法生成的， 根据其他客户查看和购买的内容推荐可能感兴趣的项目。然而， 当临床医生搜索关于特定患者问题的电子健康记录(EHR)时，EHR 不会为潜在有用的信息提供建议。相反，它需要临床医生通过 相同的手工、繁琐和费力的搜索和检索相似的信息的过程 每一次都是病人/问题。这一限制在高风险情况下被放大，例如管理胸腔 急诊科疼痛(ED)。本项目的目标是实施和评估功能配置文件作为一种方法 改善EHR的信息检索，减少认知超负荷。我们的核心假设是 建议是，CF将(1)帮助临床医生更有效地检索和审查正确的患者信息，并 比目前的方法更有效；以及(2)在有用性和易用性方面比目前的EHR得分更高。我们 我将在CareWeb Plus中实现我们的CF算法，这是我们目前正在构建的一款智能FHIR应用程序 整合印第安纳州患者护理网络(INPC)的相关信息，印第安纳州的主要健康 信息交流，与CERNER/EPIC的ED工作流程。我们的目标是(1)将CareWeb Plus扩展到 支持协同过滤；(2)协同过滤算法的设计与实现；(3)实现 并在两个成人急诊科评估CareWeb Plus。超过190名临床医生将使用和评估 CareWeb Plus在印第安纳波利斯最繁忙的两个急诊科(&gt；每年200,000人次就诊) 合计)，其中13,000多起与胸痛有关。我们将评估(1)流程措施，如 随着CareWeb Plus的使用，信息检索和查看模式，以及关键决策的时间(第一顺序， (2)结果变量，如实验室/程序利用率、急诊室住院时间和 录取率；以及(3)用户对有用性和可用性的看法和态度。我们的项目是 意义重大，因为它解决了目前EHR在临床实践中的两个主要限制。(1)临床医生有 难以有效地查看大量的患者特定信息，特别是来自多个来源的信息 相关事实，特别是在时间敏感的情况下。(2)EHR用户很少或根本没有能力更改静态 以及用于信息检索的电子病历接口的不灵活性质。我们的方案是创新的，因为它使用了CF， 一种用于定制信息检索的方法，该方法在除医疗保健之外的许多领域都很成熟，以帮助解决 这两个问题。协作过滤将提供持续适应、动态的 自然地跟随临床实践的演变的信息检索和呈现。此外，我们的 建议是由医生和患者共同产生的，因此超越了传统的方案 仅查看用户或项目关系来生成推荐的CF算法的一部分。

Project Abstract When we consider buying a book on Amazon's Website, we often benefit from items listed in a section called "Customers also viewed." These recommendations, generated by a method called collaborative filtering (CF), suggest items of possible interest based on what other customers have viewed and purchased. However, when clinicians search the electronic health record (EHR) with regard to a particular patient problem, the EHR does not make suggestions for potentially useful information. Instead, it requires clinicians to go through the same manual, cumbersome and laborious process of searching for and retrieving information for similar patients/problems every single time. This limitation is magnified in high-risk situations, such as managing chest pain in the emergency department (ED). The goal of this project is to implement and evaluate CF as a method to improve information retrieval from EHRs and reduce cognitive overload. The central hypothesis of our proposal is that CF will (1) help clinicians retrieve and review the right patient information more efficiently and effectively than current methods; and (2) score higher on usefulness and ease-of-use than current EHRs. We will implement our CF algorithms in CareWeb Plus, a SMART-on-FHIR app we are currently building to integrate relevant information from the Indiana Network for Patient Care (INPC), Indiana's major health information exchange, with the ED workflow in Cerner/Epic. Our aims are to (1) extend CareWeb Plus to support collaborative filtering; (2) design and implement collaborative filtering algorithms; (3) and implement and evaluate CareWeb Plus in two adult emergency departments. Over 190 clinicians will use and evaluate CareWeb Plus in the two busiest emergency departments in Indianapolis (> 200,000 patient visits/year collectively), with more than 13,000 of them related to chest pain. We will evaluate (1) process measures, such as CareWeb Plus use, information retrieval and viewing patterns, and time to key decisions (first order, admission, discharge); (2) outcomes variables, such as lab/procedure utilization, ED length of stay and admission rate; and (3) user perceptions and attitudes regarding usefulness and usability. Our project is significant because it addresses two current, major limitations of EHRs in clinical practice. (1) Clinicians have difficulty reviewing voluminous patient-specific information, especially from multiple sources, efficiently to find relevant facts, especially in time-sensitive situations. (2) EHR users have little to no ability to change the static and inflexible nature of EHR interfaces for information retrieval. Our proposal is innovative because it uses CF, a method for tailoring information retrieval well-established in many fields except healthcare, to help solve these two problems. Collaborative filtering will provide a continually adapting, dynamic paradigm of informational retrieval and presentation that naturally follows the evolution of clinical practice. In addition, our recommendations are generated from both physicians and patients, and thus go beyond the traditional scheme of CF algorithms that only look at user or item relations to generate recommendations.