Transforming Kidney Care in the Emergency Department using Artificial Intelligence Driven Clinical Decision Support

项目摘要/摘要 这项提案的目标是确定纳入人工智能(AI)的最佳实践方法- 获得了对急救护理的见解。本次调查将使用迭代开发和评估 人工智能驱动的预防或减轻急性肾损伤(AKI)的临床决策支持(CDS)系统作为参考 用例。我们正在回应AHRQ的特别强调通知：卫生服务研究的优先事项 实现高价值医疗体系(NOT-HS-18-015)，呼唤疾病预防研究 通过将人工智能纳入医疗保健和预防肾脏疾病进展的干预措施。 急诊科(ED)在危险的决策环境中提供大容量的患者护理 充满了过度的认知负担和时间压力。人工智能具有支持急诊临床医生决策的潜力 通过利用大规模电子健康记录(EHR)数据来辅助预测，从噪声中提取信号，以及 在实践中减少令人不快的变异性。尽管人工智能热衷于推广，但翻译成实践是罕见的和手段 将值得信赖、透明和可解释的人工智能纳入教育部门是未知的。 Aki是人工智能驱动的预测建模的一个重要目标。它很普遍，并与不良反应密切相关 结果，包括透析和死亡，但仍未得到充分认识，因此未得到充分治疗。此外，还有许多 ED治疗无意中促进了AKI和肾脏疾病的进展。预防Aki是可以实现的 以循证CDS为基础的护理要点。我们将使用我们的人工智能驱动的模式，并在早期使用经过验证的能力 和可靠的AKI风险评估，以实现以下具体目标： 目的1：开发一种人工智能驱动的算法，以促进急诊科以AKI为重点的临床决策。 我们将利用之前开发的AKI监视和预测工具来生成基于EHR的统一 使ED临床医生能够预防肾脏疾病进展的算法。 目的2：将人工智能算法转化为AKI-CDS系统，以实现对临床医生-AI的深入研究 ED中的相互作用。我们将在创建数据收集工具的同时确定最终用户要求，以 检查ED中的人工智能。这两项努力都将支持AKI-CDS系统的开发，以进行试点和调查 ED临床医生对人工智能可信度和可解释性的看法，为多地点实施做准备。 目标3：对急诊室AKI-CDS系统进行多点有效性实施评估。 我们将在三个ED研究站点实施人工智能驱动的CDS系统，使用务实的调查 建立一个框架，同时进行有效性评价和执行情况评价。 建议的研究将产生新的知识和工具，以推动教育署对人工智能的研究，并将 带来可扩展的CDS产品，能够将肾脏护理的质量提高到超过 美国每年有100万名患者受到AKI的影响。

PROJECT SUMMARY/ABSTRACT The objective of this proposal is to determine best-practice methods for incorporating artificial intelligence (AI)- derived insights into emergency care. This investigation will use the iterative development and evaluation of an AI-driven clinical decision support (CDS) system to prevent or mitigate acute kidney injury (AKI) as a reference use-case. We are responding to the AHRQ Special Emphasis Notice: Health Services Research Priorities for Achieving a High Value Healthcare System (NOT-HS-18-015), calling for research on prevention of disease through incorporation of AI into healthcare and on interventions to prevent kidney disease progression. Emergency departments (EDs) deliver high-volume patient care in hazardous decision-making environments fraught with excessive cognitive loading and time-pressure. AI has potential to support ED clinician decisions by exploiting large-scale electronic health record (EHR) data to aid prognosis, extract signal from noise, and reduce untoward variability in practice. Despite AI’s fervent promotion, translation to practice is rare and means to incorporate AI that is trustworthy, transparent, and explainable in the ED are unknown. AKI is an important target for AI-driven predictive modeling. It is prevalent and strongly associated with adverse outcomes including dialysis and death, yet is under-recognized and therefore under-treated. In addition, many ED treatments inadvertently promote the progression of AKI and kidney disease. AKI prevention is achievable with evidence-based CDS at the point of care. We will use our AI-driven model, with proven capacity for early and reliable AKI risk estimation, to achieve the following Specific Aims: Aim 1: Develop an AI-driven algorithm for promotion of AKI-focused clinical decision-making in the ED. We will leverage previously developed AKI surveillance and prediction tools to generate a unified EHR-based algorithm that empowers ED clinician prevention of kidney disease progression. Aim 2: Translate the AI algorithm to an AKI-CDS system to enable in-depth study of Clinician-AI interactions in the ED. We will establish end-user requirements while creating data collection instruments to examine AI in the ED. Both efforts will support the development of the AKI-CDS system to pilot and investigate ED clinician perceptions of AI trustworthiness and explainability in preparation for multi-site implementation. Aim 3: Perform a multi-site effectiveness-implementation evaluation of the AKI-CDS system in the ED. We will implement the AI-driven CDS system across three ED study sites using a pragmatic investigational framework to perform effectiveness and implementation evaluations in parallel. The research proposed will generate new knowledge and tools to advance the study of AI in the ED, and will result in a scalable CDS product with the capacity to improve the quality of kidney care delivered to more than 1 million patients affected by AKI in the US each year.