Comprehensive Summary
This study by Alpern et al. investigated the use of machine learning to predict early-onset sepsis in children presenting to the emergency department (ED). Researchers derived and validated predictive models using data from more than 1.6 million ED visits across five U.S. health systems in the Pediatric Emergency Care Applied Research Network (2016–2022). Using electronic health record (EHR) data from the first four hours of ED care, the team developed logistic regression and gradient tree boosting models to predict sepsis within 48 hours according to the Phoenix Sepsis Criteria (PSC). The gradient tree boosting model demonstrated excellent performance, achieving an AUROC of 0.94 for sepsis and 0.93 for septic shock. Key predictors included the emergency severity index, age-adjusted vital signs, oxygen saturation, and indicators of medical complexity. Model performance was consistent across demographic groups, with slightly higher accuracy among Medicaid-insured patients.
Outcomes and Implications
Early recognition of pediatric sepsis is essential for improving outcomes, yet current diagnostic methods remain constrained by variability in presentation and criteria. The validated models demonstrated high accuracy in identifying children at risk for sepsis using data available within the initial hours of ED evaluation. These models may enable earlier recognition of sepsis and facilitate prompt intervention. The authors emphasize that prospective validation and real-world implementation studies are needed to confirm effectiveness and ensure equitable performance across populations.