Comparison of 2005 and 2024 diagnostic criteria for early identification of pediatric sepsis and septic shock in PICU patients: a prospective cohort study

Background: In 2024, new international consensus criteria for pediatric sepsis and septic shock (2024 criteria) were introduced, replacing the 2005 criteria. The 2024 criteria use the Phoenix Sepsis Score (PSS) to define sepsis (score ≥2) and septic shock (cardiovascular PSS ≥1) in children with suspected infection, moving away from the 2005 reliance on systemic inflammatory response syndrome (SIRS). This study compares the two criteria in terms of diagnostic consistency, disease severity, prognosis, and early identification. Methods: Pediatric patients with infection admitted to the PICU at the Capital Institute of Pediatrics from May 2023 to May 2025 were prospectively enrolled. Those diagnosed with sepsis within 0-6 h of admission were included. Data on demographics, infection sites, pathogens, laboratory markers (platelets, albumin, creatinine, lactate), organ dysfunction scores (PCIS), and clinical outcomes (mechanical ventilation, CRRT, MODS, DIC, mortality) were collected. Diagnostic agreement was assessed using Kappa statistics, and performance was compared using McNemar's test. The 2005 criteria served as the reference for calculating sensitivity, specificity, and predictive values of the 2024 criteria. Results: < 0.001), with significant differences in severity markers (lactate, PCIS, MODS) and outcomes. The 2024 criteria better reflected sepsis severity but were associated with potential underdiagnosis of early septic shock. For septic shock, 20 cases met vasoactive criteria only, risking missed early diagnosis. Cardiovascular thresholds in the 2024 PSS may be overly strict, delaying recognition. No significant difference in predicted mortality was observed between criteria. Conclusions: The 2024 sepsis criteria improve specificity but may overlook early septic shock. The 2024 septic shock criteria are stricter, potentially delaying diagnosis and treatment. Prospective studies and AI-supported early warning models are needed for better early identification and outcomes.