R01 Administrative Supplement for AI Prediction of Trauma Resuscitation Responsiveness

PADAPOR R 01项目摘要 创伤患者的最初复苏通常被描述为混乱， 护理必须创造平静，同时作出生与死的决定，往往与不足， 信息.尽管对出血生物学的理解取得了一些进展，但损伤仍然是一个复杂的问题。 每年有500多万人死于疟疾，占全球死亡人数的十分之一， 仍然是美国45岁以下人群的主要死因。而超过90%的创伤患者 对可预防死亡的最大贡献仍然是那些患有出血的人 及其相关并发症。创伤有一个已知的时间零点，这使得它成为一个理想的 模型研究与出血相关的直接病理生理变化， 不同的结果。迄今为止，治疗途径被认为是初步的， 根据平均治疗效果优化结果，而不是适应于 独特的患者表型。母R 01提案的重点是探索更深层次的 理解患者对损伤的初始反应的机制建模（目标1）， 将其与改进的实时护理点床边决策支持技术相结合（目标2） 以尽早识别那些有不良结局风险的患者。母方案的净产品是定义 新的患者表型，可能需要精确的复苏方法， 出血后的结局。母公司R 01的目标是开发数字生物标志物， 精确的创伤复苏，重点是了解 作为对创伤性损伤的全身反应而发生的炎症和凝血特征。的 父R 01项目的总体目标保持不变，旨在解决我们的 通过改善对护理点患者结局轨迹的预测， 那些遭受创伤的人。父R 01项目通过两个方面解决了这些差距 相互关联的目标也保持不变： AIM 1.开发一种知识网络（神经网络）方法， 出血后的患者轨迹。我们假设（1A）预测轨迹为 死亡率和并发症可以通过深度学习方法确定;（1B）增加 生物学数据（炎症和凝血标志物）将进一步改善对患者状态的预测 或独特的表型患者概况（也称为数字生物标志物）， 结果，和（1C）这些表型可用于提高早期识别患者关闭他们的 预测轨迹，从而优化分诊和治疗途径。 AIM 2.通过以下方法开发隐匿性出血检测的先导性预测模型： 集成高保真度、集成的床旁数据和床旁成像。我们 我假设，通过开发一种新的诊断方法， (2A)使用先进的机器学习自动计算机算法来检测 通过整合多模态数据源，在床旁超声图像（2B）上显示腹腔 使用2A，我们可以预测临床上显著的隐匿性出血，以及（2C）结合成像 目标1中的知识网络预测模型，我们可以开发试点增强数字 生物标志物，以识别患有临床隐匿性出血的患者， 并发症，和那些最有可能死于他们的伤害。 迄今为止，我们对亲本R 01的研究表明，血栓形成的风险不同， 基于初始凝血特征的出血、呼吸和死亡率结果3 -4。我们的研究结果 生物学在区分两种看似不同的生物中起着重要的作用。 相同的患者在严重损伤后有不同的结果3 -19。这包括 表明严重受伤的患者最初由于纤维蛋白溶解亢进而遭受出血， 血小板功能障碍和凝血障碍。那些幸存下来的人迅速转变成一种状态 高凝状态其机制仍然是难以捉摸的，但紊乱的炎症 细胞因子的上调已经涉及，我们继续在亲本R 01中探索这一点。

PARENT R01 PROJECT ABSTRACT The initial resuscitation of a trauma patient is often described as chaos and the clinician directing the care must create calm while making life and death decisions often with inadequate information. Despite some advances in understanding the biology of hemorrhage, injury still accounts for over 5 million deaths per year, represents 1 out of every 10 deaths worldwide, and remains the leading U.S. cause of death for those under 45. While > 90% of trauma patients do well, the largest contribution to preventable death remains for those suffering from hemorrhage and its related complications. Trauma has a known time zero of onset which makes it an ideal model to study the immediate pathophysiologic changes associate with hemorrhage that lead to differential outcome. To date, treatment pathways are considered rudimentary reflecting attempts to optimize outcome based upon the average treatment effect, rather than being adaptable for unique patient phenotypes. The parent R01 proposal is focused on exploring a deeper understanding of the mechanistic modeling of initial patient response to injury (Aim 1) and coupling this with improved real-time point of care bedside decision support technology (Aim 2) to identify early those at risk of poor outcome. The net product of the parent proposal is to define novel patient phenotypes that may require precision resuscitation approaches to maximize outcome following hemorrhage. The goal of the parent R01 is to develop digital biomarkers for precision trauma resuscitation with a focus on understanding the cross talk between the inflammatory and coagulation profiles that occur as a systemic response to traumatic injury. The overall goal of the parent R01 project remains unchanged and are to address limitations of our current knowledge by improving forecasts of patient outcome trajectory at the point of care for those suffering from traumatic injury. The parent R01 project addresses these gaps through two interrelated aims which also remain unchanged: AIM 1. To develop a knowledge network (neural net) approach for characterizing early patient trajectory following hemorrhage. We hypothesize that (1A) predictive trajectories for mortality and complications can be ascertained through deep learning approaches; (1B) the addition of biologic data (inflammatory and coagulation markers) will further improve the prediction of patient states or unique phenotypic patient profiles (also known as digital biomarkers) attributable to differential outcome, and (1C) these phenotypes could be utilized to improve earlier recognition of patients off their predicted trajectory and thus, optimize triage and treatment pathways. AIM 2. To develop pilot prediction models for the detection of occult hemorrhage through the integration of high fidelity, integrated point-of-care data and bedside imaging. We hypothesize that the poor sensitivity for prediction of occult hemorrhage can be improved by developing (2A) an automated computer algorithm using advanced machine learning to detect any free fluid in the abdominal cavity on point- of-care sonographic images (2B) by integrating multimodality data sources with 2A, we can predict clinically significant occult hemorrhage, and (2C) combining the imaging prediction models with the knowledge network in Aim 1, we can develop pilot enhanced digital biomarkers to identify patients suffering from clinically occult hemorrhage, those most likely to develop complications, and those most likely to succumb to their injuries. Our work to date on the Parent R01 has shown that there is differential risk of thrombotic, bleeding, respiratory, and mortality outcomes based on initial coagulation profiles3-4. Our findings have suggested that biology plays a major factor in distinguishing between two seemingly identical patients who have disparate outcomes following critically injury3-19. This includes demonstrating that severely injured patients initially suffer from bleeding due to hyperfibrinolysis, platelet dysfunction, and disordered coagulation. Those who survive rapidly transition to a state of hypercoagulability. The mechanisms for this remains elusive but disordered inflammatory cytokine upregulation has been implicated and we continue to explore this in the parent R01.