AI-Driven Predictive Models for Early Detection of Pediatric Sepsis: A Systematic Review and Meta-Analysis

Abstract

Background Pediatric sepsis continues to pose a major challenge in healthcare, compounded by delayed diagnosis and treatment resulting in poor outcomes. Artificial intelligence (AI) and machine learning (ML) continue to develop predictive models that can support the early identification of pediatric sepsis and assist with better patient outcomes. Objective This systematic review and meta-analysis evaluated AI-driven predictive models for identifying early pediatric sepsis and evaluated diagnostic accuracy, performance metrics, and clinical readiness. Methods Following PRISMA guidelines and registered in PROSPERO (CRD420251244587), we searched PubMed, Google Scholar, Cochrane, and Scopus for studies on AI models predicting pediatric sepsis. AUROC (Area Under the Receiver Operating Characteristic Curve) was the major performance metric, along with sensitivity, specificity, and accuracy. For statistical analysis, AUROC values were converted into Cohen’s d to measure effect size, and upper and lower confidence intervals were determined. A forest plot was then generated, confirming the AI models’ strong predictive performance with statistically significant results. Results This review contains 14 studies with a total of 96,764,476 pediatric patients. AI models improved the accuracy and the ease of detecting sepsis [average AUROC = 0.868 (86.8%)]. ML models were accurate for predicting and detecting sepsis, especially when real-time vital signs, laboratory tests and waveform data were analyzed together, which increased specificity and reliability of early detection. Conclusion AI models are superior to traditional clinical scoring systems in early detection of pediatric sepsis. Nevertheless, the field needs to overcome challenges with data heterogeneity, model interpretability, and clinical adoption. Future work should prioritize validation outside of the original data set, federated learning, and explanations of AI to improve their usability in clinical practice. Keywords: Pediatric sepsis, artificial intelligence, machine learning, predictive analytics