Situation Awareness to Improve Infant Sepsis Recognition in the Presence of Clinical Uncertainty

项目总结 脓毒症在婴儿中的死亡率高于其他儿科年龄组，与严重的长时间感染有关。 30%-50%的幸存者患有足月症，并长期给医疗资源带来负担 住院和复杂的干预措施。脓毒症的快速识别和及时启动 抗菌治疗是改善婴儿结局的关键。然而，当前诊断的限制 治疗方法包括婴儿的不同的、微妙的临床表现和有限的准确性。 实验室检测。因此，迫切需要制定策略，改进脓毒症的早期发现。 以改善婴儿的预后。我们的目标是通过发展婴儿来提高对脓毒症的认识 败血症早期识别系统，将患者数据与预测模型输出相结合，以提供 向临床医生和护士提供及时、准确和相关的信息。我们的假设是， 临床数据的预测模型显示，提高对情况的认识将改善及时的脓毒症 认识和管理。我们将利用我们前期工作的坚实基础来预测 建模和我们新生儿败血症登记的现有数据，以产生识别婴儿的新方法 患新生儿败血症的风险最大。我们已经组建了一支由多个学科组成的调查小组， 数据科学、临床信息学、新生儿学和败血症/传染病研究的学科 提供专家一致的建议。在拟议的工作结束时，我们将拥有 开发的方法支持将临床数据和机器学习输出集成到 适用于临床工作流程的决策支持工具。我们预计这样的系统将配对先进的 以用户为中心的预测方法将在许多情况下具有广泛的适用性 和人口数量。这项工作将为未来的临床试验奠定基础，以评估其 确定败血症风险最高的婴儿，为临床医生和护士提供决策支持 需要改善他们的健康和安全。

PROJECT SUMMARY Sepsis has higher mortality in infants than other pediatric age groups, is associated with severe long- term morbidities in 30-50% of survivors, and burdens healthcare resources with prolonged hospitalization and complex interventions. Rapid identification of sepsis and timely initiation of antimicrobial therapy are critical to improve infant outcomes. However, limitations to current diagnostic approaches include the heterogeneous, subtle clinical presentation of infants and limited accuracy of laboratory tests. There is therefore an urgent need for strategies to improve the early detection of sepsis in infants to improve outcomes. Our objective is to improve sepsis recognition by developing an infant sepsis early recognition system that combines patient data with predictive model outputs to deliver timely, precise and relevant information to clinicians and nurses. Our hypothesis is that the integration of a predictive model with clinical data displays that improve situation awareness will improve timely sepsis recognition and management. We will utilize the strong foundation of our preliminary work in predictive modeling and existing data from our neonatal sepsis registry to produce novel methods to identify infants at greatest risk for neonatal sepsis. We have assembled a multi-disciplinary team of investigators from the disciplines of data science, clinical informatics, neonatology, and sepsis/infectious disease research to provide expert consensus recommendations. At the conclusion of the proposed work, we will have developed methods that support the integration of clinical data and machine learning outputs into decision support tools suited to clinical workflows. We anticipate such systems that pair advanced prediction methods with user-centered design processes will have broad applicability to many conditions and populations. This work will form the foundation for a future clinical trial to evaluate its ability to identify infants at highest risk of sepsis and provide clinicians and nurses with the decision support needed to improve their health and safety.