Novel Sepsis Sub-phenotypes Based on Trajectories of Vital Signs

Overview

Sepsis is a dysregulated host response to infection resulting in organ dysfunction. Over the past three decades, more than 30 pharmacological therapies have been tested in \>100 clinical trials and have failed to show consistent benefit in the overall population of patients with sepsis. The one-size-fits-all approach has not worked. This has resulted in a shift in research towards identifying sepsis subphenotypes through unsupervised learning. The ultimate objective is to identify sepsis subphenotypes with different responses to therapies, which could provide a path towards the precision medicine approach to sepsis. The investigators have previously discovered sepsis subphenotypes in retrospective data using trajectories of vital signs in the first 8 hours of hospitalization. The team aims to prospectively classify adult hospitalized patients into these subphenotypes in a prospective, observational study. This will be done through the implementation of an electronic health record integrated application that will use vital signs from hospitalized patients to classify the patients into one of four subphenotypes. This study will continue until 1,200 patients with infection are classified into the sepsis subphenotypes. The classification of the patients is only performed to validate the association of the subphenotypes with clinical outcomes as was shown in retrospective studies. Physicians and providers treating the patients will not see the classification, and the algorithm classifying the patients will in no way affect the care of the patients. Further, all the data needed for the algorithm (vital signs from the first 8 hours) are standard of care, and enrollment in the prospective study does not require any additional data.

The primary goal of this study is to investigate the feasibility of implementing a prospective sepsis subphenotyping tool in the electronic health record and evaluating the characteristics and outcomes of the sepsis subphenotypes. During this study, clinicians will not see the results of the algorithm or have access to its predictions. Instead, the algorithm will run silently in the background and continuously compute the subphenotypes of patients who are presenting to the emergency department (ED). For each patient, the probability of subphenotype membership over the first 8 hours of presentation to the ED will be calculated using an algorithm previously validated on retrospective data. Differences in clinical characteristics and outcomes between the subphenotypes will be compared. Investigators will seek to classify 1,200 patients with suspected infections. Since it will not be apparent on ED presentation who has suspected infection, all patients will be classified into subphenotypes using the algorithm, but the primary subgroup who will be analyzed will be patients with suspected infection.