Effective pediatric gram-negative sepsis identification remains a significant challenge in emergency departments worldwide. While procalcitonin (PCT) serves as a common biomarker, its accuracy as a standalone tool is often insufficient. Consequently, a recent multi-site study investigated whether machine learning could improve its diagnostic utility by contextualizing it with other clinical features.

Researchers analyzed data from 431 pediatric encounters across four different sites. Specifically, they focused on cases where both blood cultures and PCT were co-ordered. The team used a priority-based NLP algorithm to classify culture results into specific categories. Interestingly, the primary outcome involved gram-negative bloodstream infection (BSI) with concurrent organ dysfunction. They used the Phoenix-8 criteria to ascertain this dysfunction accurately.

Optimizing Pediatric Gram-Negative Sepsis Identification with Multi-Feature Models

The study evaluated several models, including PCT alone and multi-feature combinations. Initially, the researchers tested a benchmarking model that included adult-aligned features like respiratory rate and systolic blood pressure. However, these factors contributed very little to the performance in children. Instead, systematic screening identified platelet count and creatinine as the most effective co-features to pair with PCT.

The results showed a marked improvement in accuracy. While PCT alone achieved an AUROC of 0.762, the three-feature machine learning model reached an AUROC of 0.884. Furthermore, this improved performance remained consistent across different algorithms, such as Random Forest and LASSO. Notably, the model's positive likelihood ratio was 12.56, suggesting high diagnostic reliability.

Therefore, embedding PCT within a targeted machine learning framework significantly enhances discrimination. This approach allows for much earlier pediatric gram-negative sepsis identification before culture results return. Consequently, clinicians can initiate appropriate antibiotic therapy faster, potentially improving outcomes for critically ill pediatric patients.

How does machine learning improve pediatric gram-negative sepsis identification?

Machine learning contextualizes procalcitonin results with other clinical data like platelet counts and creatinine. This multi-feature approach identifies subtle patterns that a single threshold cannot detect alone.

Why is early identification of gram-negative sepsis critical?

Gram-negative infections often cause rapid organ dysfunction and have high mortality rates. Identifying these pathogens before culture results allows for the timely administration of targeted antibiotics and better patient management.

Disclaimer: This content is for informational and educational purposes only. It does not constitute professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified healthcare provider with any questions you may have regarding a medical condition. Refer to the latest local and national guidelines for clinical practice.

References

Velez T et al. PCT-Anchored Machine Learning for Pre-Culture Identification of Gram-Negative Sepsis in Children: A Four-Site Study. Shock. 2026 Jun 22. doi: 10.1097/SHK.0000000000002895. PMID: 42320014.

Alpern ER et al. Derivation and Validation of Predictive Models for Early Pediatric Sepsis. JAMA Pediatr. 2025;179(12):1318–1325.

Beltrán et al. Microvascular phenotypes in pediatric sepsis identified by machine learning: prognostic implications for organ dysfunction and mortality. Critical Care. 2026; 30:61.