Sepsis Early Prediction and Subphenotype Illumination Study (SEPSIS)

项目摘要 脓毒症是一种由感染引起的危及生命的器官功能障碍综合征，在住院患者和 导致严重的发病率、死亡率和成本。美国有170多万患者患上脓毒症 每年，这个数字都会随着人口老龄化而增加。脓毒症患者的捐款超过240亿美元 每年的医疗费用，最近的一项研究发现，败血症导致了高达一半的医院死亡。 此外，败血症的幸存者会遭受长期的认知障碍和身体残疾。因此， 改善脓毒症患者的护理将对社会产生巨大的好处。然而，有几个 该领域需要解决的关键差距：1)在识别受感染患者方面的延误很常见，而且 与死亡率增加有关；2)即将到来的危重疾病患者的风险分层错误和 脓毒症是常见且致命的；3)目前对感染患者的治疗策略是一刀切的。 方法，忽略了由于复杂的临床表现和潜在生物学的广泛范围 患者特征、感染有机体和宿主免疫反应之间的相互作用。 PI研究计划的总体愿景是通过利用详细的 多中心电子健康记录(EHR)、临床试验和生物标志物数据与机器学习相结合 改进脓毒症的识别、风险分层和发现重要的亚型的方法 减少可预防的感染死亡。在过去的五年里，PI成功地获得了独立 通过NIGMS R01和国防部奖励提供资金。该协会发表了80多份同行评议报告 在此期间，是几个国家和国际委员会的活跃成员，有 参加了多个NIH研究组，有40名学员，其中6人获得了NIH K-Level奖。 重要的是，PI还开发并实现了一个机器学习风险分层工具，称为Ecart， 在20多家医院，这降低了高危病房患者的死亡率。今后五年的奋斗目标是 在这些成功的基础上，通过三个未来方向解决该领域的主要差距：1)使用 语言处理和深度学习，以改进感染患者的识别和风险分层，2) 使用研究生物标记识别重要的亚表型，以及3)使用机器学习来开发 个性化治疗算法。这些项目具有创新性，因为它们将使用先进的机器 在结构化和非结构化的EHR和生物标志物数据的大型多中心集合中的学习方法 为脓毒症患者开发新的工具。未来，这些模型将在更早的时间内实现 识别、准确的风险分层，并在床边提供个性化护理。这是有可能的 对住院病人中最常见和最致命的疾病之一的护理进行革命。

PROJECT ABSTRACT Sepsis, a life-threatening organ dysfunction syndrome due to infection, is common in hospitalized patients and leads to significant morbidity, mortality, and costs. Over 1.7 million patients develop sepsis in the United States each year, a number that will increase as the population ages. Patients with sepsis contribute to over $24 billion in healthcare costs yearly, and a recent study found that sepsis contributed to up to half of hospital deaths. Furthermore, survivors of sepsis suffer long-term cognitive impairment and physical disability. Therefore, improving the care of patients with sepsis would be enormously beneficial to society. However, there are several critical gaps in the field that need to be addressed: 1) delays in identifying infected patients are common and associated with increased mortality; 2) errors in risk stratification of patients with impending critical illness and sepsis are common and deadly; 3) current treatment strategies for infected patients utilize a one-size-fits-all approach, which neglects the wide range of clinical presentations and underlying biology due to the complex interactions between patient characteristics, the infectious organism, and the host immune response. The overall vision of the PI’s research program is to address these knowledge gaps by utilizing detailed multicenter electronic health record (EHR), clinical trial, and biomarker data combined with machine learning approaches to improve the identification, risk stratification, and discover important subphenotypes of sepsis to decrease preventable death from infection. Over the past five years, the PI has successfully secured independent funding through an NIGMS R01 and Department of Defense award. The PI has published over 80 peer-reviewed publications during this time, is an active member on several national and international committees, has participated in several NIH study sections, and has 40 mentees, including six with NIH K-level awards. Importantly, the PI has also developed and implemented a machine learning risk stratification tool, called eCART, in over 20 hospitals, which has decreased mortality in high-risk ward patients. The goal of the next five years is to build upon these successes and address key gaps in the field through three future directions: 1) using natural language processing and deep learning to improve the identification and risk stratification of infected patients, 2) identifying important subphenotypes using research biomarkers, and 3) using machine learning to develop personalized treatment algorithms. These projects are innovative because they will utilize advanced machine learning methods in a large, multicenter collection of structured and unstructured EHR and biomarker data for developing novel tools in patients with sepsis. In the future, these models will be implemented for earlier identification, accurate risk stratification, and to deliver personalized care at the bedside. This has the potential to revolutionize the care of one of the most common and deadly conditions in hospitalized patients.