CONTEXT The accurate identification of febrile infants who are at risk for invasive bacterial infections (IBIs), ie, bacteremia and meningitis, is essential to reduce morbidity and avoid unnecessary procedures. The role of machine learning (ML) in improving their diagnostic accuracy remains under evaluation. OBJECTIVE To systematically review the diagnostic performance of ML-based models for identifying febrile infants aged up to 90 days with IBIs. DATA SOURCES Comprehensive searches of MEDLINE, Scopus, Embase, CINAHL, CENTRAL, ClinicalTrials.gov, and World Health Organization International Clinical Trial Registry Platform databases through December 2024. STUDY SELECTION Eligible studies applied ML models to febrile infants aged up to 90 days to identify IBIs and reported diagnostic metrics and used culture-confirmed infections as the reference standard. Of 4756 screened records, 6 met inclusion criteria. DATA EXTRACTION Two reviewers independently extracted study characteristics, model types, outcomes, and accuracy measures. Risk of bias was assessed using a modified Quality Assessment of Diagnostic Accuracy Studies 2 tool. Marked methodological heterogeneity precluded meta-analysis. RESULTS Six studies evaluated various ML models, including logistic regression, random forests, neural networks, support vector machines, and ensemble. Sensitivity ranged from 57% to 100%, specificity ranged from 30% to 94%, and ROC-AUC ranged from 0.57 to 0.9. Several models outperformed traditional tools (eg, PECARN, Step-by-Step), particularly in specificity, for IBI prediction. LIMITATIONS Limitations included the predominance of retrospective data collection, limited reporting on data handling, and lack of external validation. CONCLUSIONS ML models show promising performance for identifying infants at risk for IBIs, improving specificity over traditional tools. Future work should systematically incorporate and transparently report explainability analyses to support clinical interpretability. Broader validation, methodological standardization, and careful clinical integration are essential before adoption.
Ferraris et al. (Thu,) studied this question.