Risk prediction model for peripherally inserted central catheters related bloodstream infection: a systematic review.

backgroundPeripherally inserted central catheters (PICC) are widely used in patients requiring medium- to long-term intravenous therapy. Catheter-related bloodstream infection is a common complication that significantly impacts patient prognosis. Current research primarily focuses on identifying risk factors for infection, while studies integrating these factors into predictive models remain limited.

objectiveThis paper aims to assess the predictive model for PICC-related bloodstream infection to inform clinical practice.

methodsA systematic literature search was conducted across Medline, Embase, Web of science, CINAHL, Cochrane, CNKI, CBM, Wanfang, VIP databases for PICC-related bloodstream infection risk prediction models from the database built until January 10, 2024. Two researchers independently screened and extracted data using the CHARMS checklist and assessed bias using the PROBAST tool.2. This study was registered on PROSPERO (CRD42023333731).

resultsTen studies were included, yielding nine prediction models. Among these, four models underwent internal validation and three received external validation, with area under the ROC curve (AUC) values ranging from 0.608 to 0.950. The sample sizes varied substantially from 80 to 23,088 participants. Clinically significant predictors commonly incorporated in the models included days until catheter removal, diabetes history, and number of punctures. However, methodological reporting was generally incomplete, particularly regarding approaches to handling missing data and the rationale for predictor selection.

conclusionThis systematic review reveals that existing prediction models for PICC-related bloodstream infections, while demonstrating moderate discriminatory ability, are limited in clinical applicability due to a high risk of bias across studies. Key methodological weaknesses including retrospective designs, inconsistencies in predictor measurement and outcome assessment, and inadequate model validation significantly undermine the robustness and generalizability of current models. These limitations highlight an urgent need for methodologically rigorous, prospective, multicenter studies to develop reliable and clinically actionable tools. Adherence to TRIPOD reporting guidelines, integration of machine learning with conventional statistical approaches, and robust external validation are essential to advance this field. Such efforts will ultimately support personalized risk assessment and targeted interventions to improve patient outcomes.