Particle.news
Download on the App Store

AI Models Predict Pediatric Sepsis Within 48 Hours From Early ER Data

Researchers urge prospective testing before hospitals consider routine use.

Overview

  • The JAMA Pediatrics study across five PECARN health systems is the first to apply Phoenix Sepsis Criteria for pediatric sepsis prediction.
  • The models were trained on more than 1.6 million records from 2016–2020 and externally validated on nearly 720,000 ER visits from 2021–2022.
  • The approach uses routine EHR inputs from the first four hours of emergency care and excludes children already septic on arrival to target subsequent risk.
  • Investigators report balanced performance in identifying later sepsis but note limited positive predictive value given the rarity of the condition.
  • Bias evaluations were conducted, and authors recommend pairing predictions with clinician judgment in future prospective studies, with support from NICHD grant R01HD087363.