Intention-aware Recommender System for Improving Trauma Resuscitation Outcomes

项目摘要 危重病人因医疗差错死亡的风险比其他住院病人高四倍。 近一半的可预防死亡与初始复苏阶段的错误有关。虽然 协议，模拟和领导力培训提高团队绩效在这种情况下，多达12个协议 即使是有经验的团队，也观察到每次复苏的偏差。考虑到不利的结果， 可能导致的性能差距，有一个迫切需要建立新的方法，应用实时 关键护理环境中的决策支持。长期目标是实施创伤决策支持 复苏和其他快节奏，高风险的重症监护环境，提高性能，减少错误， 并防止不良后果。这次更新的总体目标是纵向推进 在第一个供资期间，通过设计、实施和测试一个意图感知 推荐系统，其（1）使用传感器数据、来自患者的输出和来自患者的反馈来识别和跟踪当前目标。 监视器和从数字设备捕获的数据，（2）得出支持遵守目标的建议， 基于协议，以及（3）在墙壁显示器上以真实的时间显示这些推荐。核心假设 决策支持与意图（“预期”或“当前”目标）一致将增强协议 顺应性，从而改善与创伤复苏相关的结果。更新的理由是， 与团队意图一致的支持方案合规性的建议更有可能 通过较少的分心和较低的认知负荷来采用。根据初步数据， 中心假设将通过追求两个具体目标进行测试：1）设计和实现一个自动化的真实的- 预测和监测创伤复苏的评估和治疗目标的时间方法;以及 2)生成并显示支持创伤期间当前目标追求的推荐活动计划 复苏术对于第一个目标，机器学习方法将用于使用数据识别目标 从传感器和其他数字数据源获得。在第二个目标下，机器学习策略将 执行和测试，生成响应团队意图的建议。拟议 研究是创新的，因为它侧重于开发实时方法，将目标作为输入 提出满足最新和最相关信息需求的建议。拟议 这项研究意义重大，因为它有望改善对严重受伤和其他危重病人的护理。 通过促进及时和适当地实现关键评估和治疗目标， 医疗差错的高风险环境。这一研究连续体的结果预计将有 通过解决复杂的决策与 以及人类在重症监护环境中对错误的脆弱性。

PROJECT SUMMARY Critically injured patients have a four-fold higher risk of death from medical errors than other hospitalized patients, with nearly half of preventable deaths related to errors during the initial resuscitation phase. Although protocols, simulation, and leadership training improve team performance in this setting, as many as 12 protocol deviations per resuscitation have been observed, even with experienced teams. Given adverse outcomes that can result from performance gaps, there is a critical need to establish novel approaches for applying real-time decision support in critical-care settings. The long-term goal is to implement decision support for trauma resuscitation and other fast-paced, high-risk critical care settings that improves performance, reduces errors, and prevents adverse outcomes. The overall objective for this renewal is to vertically advance what was achieved during the first funding period by designing, implementing and testing an intention-aware recommender system that (1) recognizes and tracks current goals using sensor data, the output from patient monitors, and data captured from digital devices, (2) derives recommendations that support adherence to goal- based protocols, and (3) displays these recommendations in real time on wall displays. The central hypothesis is that decision support aligning with intentions (“intended” or “current” goals) will enhance protocol compliance, leading to improved outcomes related to trauma resuscitation. The rationale for this renewal is that recommendations supporting protocol compliance that are aligned with team intentions are more likely to be adopted by being less distracting and associated with lower cognitive load. Guided by preliminary data, the central hypothesis will be tested by pursuing two specific aims: 1) design and implement an automated real- time approach for predicting and monitoring the assessment and treatment goals of trauma resuscitation; and 2) generate and display a recommended plan of activities that supports current goal pursuit during trauma resuscitation. For the first Aim, machine learning approaches will be applied for recognizing goals using data obtained from sensors and other digital data sources. Under the second Aim, a machine learning strategy will be implemented and tested that generates recommendations responsive to team intentions. The proposed research is innovative because it focuses on development of real-time methods that integrate goals as an input for making recommendations that meet the most current and relevant information needs. The proposed research is significant because it is expected to improve the care of severely injured and other critically ill patients by promoting timely and appropriate achievement of critical assessment and treatment goals in settings that remain at high-risk for medical errors. The results of this research continuum are expected to have an important positive impact on the outcome by addressing the mismatch between complex decision-making and human vulnerability to error that remain in critical care settings.