Review Article

Published: Nov 05, 2026 | DOI: 10.24911/amem.15-3130

Artificial intelligence and machine learning for real-time late-onset neonatal sepsis prediction using continuous physiologic monitoring: asystematic review and meta-analysis


Authors: Layla Abdullah Alajlan ORCID logo , Aseel M Alsamaani ORCID logo , Rola H Alseghair ORCID logo , Sarah A Alomar ORCID logo , Hind H Almazyad ORCID logo , Razan A Aljameeli ORCID logo , Abdulrahman Obaid Alharbi


Abstract

Background: Late-onset neonatal sepsis (LOS) causes significant mortality and morbidity in the NICU. Early detection is challenging due to non-specific symptoms and low sensitivity of current diagnostic methods. Artificial intelligence (AI) and machine learning (ML) may enable earlier detection through continuous physiologic monitoring. This systematic review evaluates the performance and clinical utility of AI/ML models for early LOS prediction in NICU patients. 

Methods: Following PRISMA 2020 guidelines, PubMed, Google Scholar, and Web of Science were searched for studies using AI/ML to predict LOS from continuous vital signs in the NICUs. Eligible studies reported AUC-ROC as the primary outcome, pooled using random-effect models. Subgroup analyses were conducted by model type.  Quality was assessed using PROBAST. 

Results:  Ten studies encompassing 8,882 patients were included. The pooled AUC-ROC was 0.84 (95% CI: 0.77-0.92), with high heterogeneity (I2= 99.2%). Random Forest and Support Vector Machine were top performance  (AUC = 0.86). Neural Network showed lower predictive ability (AUC = 0.78) but no significant heterogeneity (I2= 0%). Combining multiple vital signs enabled the prediction of sepsis 6 to 48 hours prior to the onset of sepsis, with sensitivity reaching 96%. Multilevel analysis indicated that heterogeneity was driven by intra-study effect size rather than inter-study variance. Half of the studies carried low risk of bias.

Conclusion: AI and ML models based on continuous physiologic monitoring demonstrate clinically meaningful accuracy for early LOS prediction and may enable timely intervention without requiring laboratory data, offering particular value in resource-limited NICU settings.


Keywords: Late-onset neonatal sepsis, artificial intelligence, machine learning, neonatal intensive care unit, physiologic monitoring.



Pubmed Style

Layla Abdullah Alajlan, Aseel M Alsamaani, Rola H Alseghair, Sarah A Alomar, Hind H Almazyad, Razan A Aljameeli, Abdulrahman Obaid Alharbi. Artificial intelligence and machine learning for real-time late-onset neonatal sepsis prediction using continuous physiologic monitoring: asystematic review and meta-analysis. AMEM. 2026; 05 (November 2026): -. doi:10.24911/amem.15-3130

Publication History

Received: October 09, 2026

Accepted: October 19, 2026

Published: November 05, 2026


Authors

Layla Abdullah Alajlan

College of Medicine, Qassim University, Buraydah, Qassim, Saudi Arabia.

ORCID logo ORCID

Aseel M Alsamaani

College of Medicine, Qassim University, Buraydah, Qassim, Saudi Arabia.

ORCID logo ORCID

Rola H Alseghair

College of Medicine, Qassim University, Buraydah, Qassim, Saudi Arabia.

ORCID logo ORCID

Sarah A Alomar

College of Medicine, Qassim University, Buraydah, Qassim, Saudi Arabia.

ORCID logo ORCID

Hind H Almazyad

College of Medicine, Qassim University, Buraydah, Qassim, Saudi Arabia.

ORCID logo ORCID

Razan A Aljameeli

College of Medicine, Qassim University, Buraydah, Qassim, Saudi Arabia.

ORCID logo ORCID

Abdulrahman Obaid Alharbi

Neonatologist and Head of the Neonatal Intensive Care Service Line, Qassim Health Cluster, Buraydah, Qassim, Saudi Arabia.