SBIR Phase II: DATA-DRIVEN DECISION SUPPORT FOR EFFICIENT PATIENT PROGRESSION

小企业创新研究（SBIR）第二阶段的更广泛的影响/商业潜力是减少因急诊科（ED）拥挤问题加剧而造成的患者伤害和经济负担。急诊科拥挤对高敏度患者的安全构成威胁，并且在许多情况下与可避免的发病率和死亡率有关。同时，急诊科面临着有效管理大量非急诊低视力患者的挑战。拟议的电子健康记录（EHR）集成技术部署了新的机器学习算法，可以预测患者护理路径中可操作时间点的临床事件。将急诊科定性为流动系统，这些决策支持工具将集中于急诊科流入和医院流出的根本原因。这一新的前瞻性预计将实现创新的医院运营模式，从而加快患者的进展（最大限度地减少患者等待），提高患者安全性，并直接转化为可衡量的成本节约和/或创收。SBIR第二期项目将带来更高的产量和可扩展的决策支持技术，同时促进数据支持科学和工程在医疗保健领域的价值。拟议的项目目标是通过推进数据科学、决策科学和运营研究来大大减少ED拥挤，这些研究是我们决策支持技术的基础。这项新技术基于数据归一化、特征选择和监督机器学习方法的新颖组合，以预测临床事件，从而驱动明确的行动来优化医院资源。这包括一个决策支持工具，在急诊科分流（流入）中发挥作用，以预测关键事件的风险，从而使急症患者和非紧急患者的服务流能够安全分离。一个辅助工具在医院出院（流出）附近发挥作用，以预测预期出院时间，从而在医院范围内优先考虑加快出院所需的资源。这将解锁下游产能，并消除造成长时间等待的主要ED流出瓶颈。拟议的技术在设计上具有创新性，可扩展，但可适应医院的个体患者群体、运营目标和风险承受能力。该技术有望进一步促进与供应商的学习，了解如何为决策支持使用新的预测和解释信息。具体的II期目标是通过以下方式继续发展该技术：(1)将反馈整合到预测输出中；(2)开发绩效监测能力；(3)推动基于系统的患者进展管理。

The broader impact/commercial potential of the Small Business Innovation Research (SBIR) Phase II is to reduce the patient harm and financial burden created by the intensifying problem of emergency department (ED) crowding. ED crowding is a threat to patient safety for high acuity patients and has been associated with avoidable morbidity and mortality across many conditions. Concurrently, EDs are challenged to efficiently manage large volumes of low-acuity patients visiting with non-emergency conditions. The proposed electronic health record (EHR) integrated technology deploys novel machine learning algorithms that predict clinical events at actionable time-points in patients care pathways. Characterizing the ED as a flow system, these decision support tools will concentrate on the root causes that exist at ED inflow and hospital outflow. This new foresight is expected to enable innovative hospital operational models that expedite patient progression (minimize patient waiting), improve patient safety, and directly translate to measurable cost-savings and/or revenue generation. The SBIR Phase II project should result in higher yield and scalable decision support technology while promoting the value of data-enabled science and engineering in healthcare. The proposed project objective is to greatly reduce ED crowding by advancing the data-science, decision-science, and operations research that underpins our decision support technology. This new technology is based upon a novel combination of data normalization, feature selection, and supervised machine learning methods to predict clinical events that drives clear action to optimize hospital resources. This includes a decision support tool that functions at ED triage (inflow) to predict risk of critical events to empower safe separation of service streams for acutely ill and non-urgent patients. A complementary tool functions near hospital discharge (outflow) to predict expected discharge time enabling hospital-wide prioritization of resources required to expedite discharge. This unlocks downstream capacity and removes a major ED outflow bottleneck that creates prolonged waiting. The proposed technology is innovative by design to be scaled, yet adaptive to hospitals individual patient populations, operational objectives, and risk tolerances. The technology is expected to further advance learning with providers about how to consume new predictive and explanatory information for decision support. The specific Phase II objectives are to continue the development of the technology by: (1) integrating feedback into prediction output, (2) developing performance monitoring capabilities, and (3) driving systems-based management of patient progression.