WIP 145 - Evaluation of a digital phenotype for the early recognition of pediatric sepsis

Poster Number: WIP 145
Publication Number: WIP 145.809

Background: Sepsis results in significant morbidity and mortality among children worldwide. Many standardized guidelines exist to improve patient outcomes, emphasizing systematic screening methods and early initiation of antibiotic therapy, ideally within 1 hour of recognition. However, sepsis can be difficult to recognize in children due to varied presentations and nonspecific symptoms. We developed a digital phenotype to identify patients in real-time who are at high risk of meeting criteria for sepsis. The phenotype was developed using: (1) a previously described retrospective informatics-based definition of sepsis requiring 4 days of antibiotics (Full Weiss definition), which we modified to be usable in real-time (RT Weiss); and (2) the Duke Children’s Trigger Tool (DCTT), a local consensus-based phenotype to direct empiric antibiotic use which was developed by our local multi-specialty team. The Duke Pediatric Sepsis Phenotype (DPSP) represents the overlap of the RT Weiss and DCTT definitions and is designed to have better sensitivity and specificity than either definition alone.

Objective: We evaluate the performance of the RT Weiss, DCTT, and the DPSP phenotypes in identifying patients who meet the Full Weiss retrospective sepsis definition (gold standard) or receive ICD-10 codes for sepsis.

Design/Methods: Evaluation of the RT Weiss, DCTT and DPSP will occur using two methods:

1. A retrospective analysis of 28,399 Duke pediatric patient hospitalizations.
2. Real-time clinical adjudication of cases identified using RT Weiss, DCTT and DPSP from patients admitted to our hospital. Two physicians will evaluate each case to determine whether the patient meets clinical requirements for immediate, empiric treatment for sepsis at the time of recognition by each tool.

This study has been reviewed and approved by our local IRB. Both analysis of the retrospective data and real-time adjudications will occur within the next 3 months. Future goals for this project include publishing a description of our analysis and pursuing bedside implementation.