Prediction of central line-associated bloodstream infection: focus on time of insertion

• DOI: 10.1017/ice.2025.1
• Publication Date: 2025-03-10T13:02:57Z
• Authors: Ari Moskowitz; Melissa Fazzari; Luke Andrea; Jianwen Wu; Arup Gope; Thomas Butler; Amira Mohamed; Christine Shen; Fran Ganz-Lord; Inessa Gendlina; Michelle Ng Gong
• Keywords: Central line; Central venous catheter; Bloodstream infection

Abstract Objective: Central line-associated bloodstream infections (CLABSIs) result in morbidity and mortality among hospitalized patients. Hospital interventions to reduce the incidence of CLABSI are often broadly applied all patients with central venous access. Identifying lines at high risk for time insertion will allow a more focused delivery preventative interventions. Design: This was an observational cohort study conducted three hospitals including who received CLABSIs were identified using institutional database maintained by hospital epidemiology team. Logistic regression (LASSO) machine learning (random forest, XGboost) techniques prediction occurrence, adjusting selected patent insertion-level characteristics. Results: A total 40,008 catheters included, which 409 (1.02%) associated CLABSI. The random forest XGBoost models had highest discrimination (Area Under Received Operating Curve [AUC] 0.79) followed LASSO (0.73). High illness severity, receipt parenteral nutrition, hemodialysis, pre-insertion length-of-stay, low albumin levels predictive occurrence. Precision poor owing false-positive rate. Discussion: can be predicted based upon patient level factors electronic health record. In this study, gradient-boosted AUC. Prediction cut-offs identification adjusted acceptable rate false-positives given intervention.