Predicting nosocomial infections in critically Ill children: a comprehensive systematic review of risk assessment models.

Background: Nosocomial infections (NIs) pose a substantial global health challenge, affecting an estimated 7%-10% of hospitalized patients worldwide. Neonatal intensive care units (NICUs) are particularly vulnerable, with NIs representing a leading cause of infant morbidity and mortality. Similarly, pediatric intensive care units (PICUs) report that 28% of admitted children acquire NIs during hospitalization. Although prediction models offer a promising approach to identifying high-risk individuals, a systematic evaluation of existing models for ICU-ill children remains lacking.

Aim: This review systematically synthesizes and critically evaluates published prediction models for assessing NI risk in ill children in the ICU.

Methods: We conducted a comprehensive search of PubMed, Embase, Web of Science, CNKI, VIP, and Wanfang from inception through December 31, 2024. Study quality, risk of bias, and applicability were assessed using the PROBAST tool. Model performance metrics were extracted and summarized.

Results: Three studies involving 1,632 participants were included. Frequency analysis identified antibiotic use, birth weight, and indwelling catheters as the most consistently incorporated predictors. All models employed traditional logistic regression, with two undergoing external validation. However, critical limitations were observed across studies: inadequate sample sizes, omission of key methodological details, insufficient model specification, and a universally high risk of bias per PROBAST assessment.

Conclusion: Current NI prediction models for ill children in the ICU exhibit significant methodological shortcomings, limiting their clinical applicability. No existing model demonstrates sufficient rigor for routine implementation. High-performance predictive models can assist clinical nursing staff in the early identification of high-risk populations for NIs, enabling proactive interventions to reduce infection rates. Future research should prioritize (1) methodological robustness in model development, (2) external validation in diverse settings, and (3) exploration of advanced modeling techniques to optimize predictor selection. We strongly advocate adherence to TRIPOD guidelines to enhance predictive models' transparency, reproducibility, and clinical utility in this vulnerable population.

Systematic review registration: PROSPERO CRD420251019763.