Learning-Enabled Autonomous Decision-Support for Blood Pressure Management in Hemorrhage Resuscitation via Population-Informed Statistical Inference

项目摘要/摘要 全世界约40%的创伤死亡是由出血造成的，还有 导致1-46岁美国人死亡的主要原因。由于失血致死率很高 在得到明确护理之前发生，为出血患者提供即时救生干预措施是 非常重要。血压管理是出血的一个非常重要的组成部分 复苏，因为它在(一)减少出血引起的死亡率以及(二)发展中的中心作用 临床试验中的新出血复苏方案。但是，临床医生在维持血压方面并不是很有效 在目标范围内，BP管理方案故障在临床试验中很常见。不管怎么说，还是有 没有成熟的技术可供临床使用，以支持临床医生进行BP管理。 通过使用自主血管加压剂给药指导技术来扩展其持续的成功 目前正在FDA集成开发环境下进行临床试验，调查小组建议开发一种学习- 启用自主决策支持(LEAD)系统，用于在出血复苏期间进行BP管理， 它可以预测患者未来的血压，并建议复苏液体的注射时间和剂量 为了将患者的血压保持在临床医生指定的目标范围内，同时不断优化其 通过了解患者对输液的反应来提高准确性。引线系统将适用于 临床应用于ICU、急诊室，甚至是院前环境。领导系统将在以下情况下发挥最大作用 临床医生是新手、心烦意乱或疲惫不堪。此外，通过将临床医生保持在环路中，将有很多 降低监管风险，允许快速过渡到临床试验和传播。这样一来，领头羊 系统有可能在出血复苏期间通过增强 临床医生对患者的动态治疗轨迹的认识。 与领导制度有关的主要创新是：(一)一种新的人口知情的、递归的、集体的 基于物理学基础的生理学方法预测患者未来血压的统计推断方法 模型和由调查小组开发的集体推理，以及(2)其在 一个可供临床使用的计算用户界面平台。为了实现和验证Lead系统， 我们将(I)开发基于人口信息的递归集合的Lead系统BP预测算法 推理(SA1)；(Ii)使用临床数据集评估领先BP预测算法(SA2)；以及(Iii)实现 主导系统采用计算用户界面平台，并进行模拟实时测试(SA3)。 如果该项目成功，调查小组将继续进行技术商业化和 通过跟进R01提案进行翻译，以优化Lead系统算法和用户界面 平台，并在FDA集成开发环境下进行临床试验。

PROJECT SUMMARY/ABSTRACT Hemorrhage is accountable for approximately 40% of deaths due to traumatic injuries worldwide as well as the leading cause of mortality in Americans 1-46 years of age. Since high rate of hemorrhage-induced deaths occur before reaching definitive care, providing immediate life-saving interventions to hemorrhaging patients is of paramount importance. Blood pressure (BP) management is a very important component of hemorrhage resuscitation due to its central role in (i) reducing the hemorrhage-induced mortality as well as in (ii) developing novel hemorrhage resuscitation protocols in clinical trials. But, clinicians are not effective at maintaining BP within a goal range, and BP management protocol failures are common in clinical trials. Regardless, there is no mature technology ready for clinical use to support clinicians with BP management. By extending its ongoing success with an autonomous vasopressor administration guidance technology currently undergoing a clinical trial under an FDA IDE, the investigative team proposes to develop a learning- enabled autonomous decision-support (LEAD) system for BP management during hemorrhage resuscitation, which can predict future BP in a patient and recommend timings and doses of resuscitation fluid administration in order to maintain the patient’s BP within a clinician-specified goal range, while continuously optimizing its accuracy by learning the patient’s response to administration of fluids. The LEAD system will be suitable for clinical use in ICUs, EDs, and even pre-hospital environments. The LEAD system will be most impactful when a clinician is novice, distracted, or tired. In addition, by maintaining clinicians in the loop, there will be much reduced regulatory risk, allowing for rapid transition to a clinical trial and dissemination. In this way, the LEAD system has the potential to enable tight BP management during hemorrhage resuscitation by enhancing the awareness of clinicians on a patient’s dynamic treatment trajectory. Key innovations pertaining to the LEAD system are (i) a novel population-informed, recursive, collective statistical inference approach to prediction of future BP in a patient based on a physics-based physiological model and a collective inference developed by the investigative team and (ii) its real-world implementation into a computational user interface platform being ready for clinical use. To realize and validate the LEAD system, we will (i) develop a BP prediction algorithm for the LEAD system via population-informed recursive collective inference (SA1); (ii) evaluate the LEAD BP prediction algorithm using clinical datasets (SA2); and (iii) realize the LEAD system using a computational user interface platform and conduct simulated real-time testing (SA3). If this project is successful, the investigative team will proceed to technology commercialization and translation by pursuing a follow-up R01 proposal to optimize the LEAD system algorithm and user interface platform, and conduct a clinical trial under an FDA IDE.