Personalized Therapies of Pediatric Sepsis

University Of Pittsburgh At Pittsburgh, Pittsburgh PA

Investigators

Abstract

PROJECT ABSTRACT Sepsis is a life-threatening condition characterized by a dysregulated response to infection, leading to significant organ dysfunction and affecting 75,000 children and one million adults each year. The estimated yearly impact of this disease is 200,000 lives lost and $20 billion dollars in healthcare expenditures. Unfortunately, many initially promising sepsis therapeutics have failed to show wide benefit with tested in wider clinical trials. However, post-hoc analyses of some of these therapeutics demonstrate benefit to specific subgroups, raising concern that failure of sepsis therapeutics have failed due to the one-size-fits-all approach employed in clinical trials. It is now widely accepted that sepsis represents a heterogeneous group, likely with subtypes that are expected to respond differently to various treatments. Multiple efforts have been made to characterize this heterogeneity, though few of these have been in pediatric patients. Moreover, phenotyping efforts have typically relied on point measurements in time, including for vital signs. However, the ICU provides a wealth of continuously measured vital sign data, including high-frequency physiologic data, that may offer insights into underlying physiologic derangements and augment sepsis subtypes. Moreover, this type of data may provide evidence in real time of developing infections prior to the onset of more classic signs such as fever. Such insights have already been demonstrated in neonatal patients. The overarching goal of this study is to develop personalized sepsis therapeutics by developing predictive tools based on integrating high-frequency physiologic data with electronic health record data. Aim 1 focuses on augmenting sepsis phenotypes through the inclusion of high-frequency data. This data is expected to identify additional subtypes or aid in prediction of disease progression after day 1. Aim 2 aims to first retrospectively identify signs of bloodstream infection (BSI) prior to traditional diagnostic signs and develop predictive algorithms based on these signs. The algorithms will then be externally validated using the multicenter CHoRUS database. Aim 3 involves a real-time evaluation of BSI prediction algorithms by silently implementing (i.e. without alerts) the best performing model in the investigator’s hospital. This will permit fine-tuning of the model and real-world evaluation. This work will significantly enhance our ability to predict, treat, and potentially even prevent sepsis by early identification of BSI, improving outcomes for pediatric patients. This research will occur within the context of a career development plan under the guidance of a multidisciplinary team with expertise in critical care, infectious diseases, machine learning, and informatics. The coursework laid out will provide additional skills necessary to conduct the proposed research and future endeavors, and the proposed research will provide the foundation for future work developing and implementing predictive analytic tools to treat and prevention severe infections in children.