A clinical prediction model for central venous catheter-associated bloodstream infections: derivation and validation from the MIMIC-IV cohort

BACKGROUND: Central venous catheter-associated bloodstream infections (CVC-BSIs) significantly affect clinical outcomes in critically ill patients. Existing risk prediction models lack comprehensive integration of early clinical indicators. This study aimed to develop a predictive model using demographic, laboratory, and catheter-related variables. METHODS: We analyzed observations on 21,667 adults with central venous catheters from the MIMIC-IV database (2008-2019). The CVC-BSI was identified using ICD code. After data preprocessing and imputation, a sequential modeling approach combined LASSO regression for variable selection and Firth regression to address class imbalance. Calibration was performed using isotonic regression, and performance was assessed via ROC analysis and bootstrap validation. RESULTS: Among 21,667 patients, 396 (1.8%) developed CVC-BSIs. Coagulase-negative Staphylococci (30.9%) and Staphylococcus aureus (12.1%) were the most common pathogens. The final model incorporated 15 predictors. Significant risk factors included renal replacement therapy (odds ratio [OR] = 2.01) and malignant cancer (OR = 2.08, 95% CI 1.58-2.71), while higher minimum hemoglobin level and metastatic solid tumor were associated with lower risk. The model demonstrated moderate discrimination with an AUC of 0.723 (95% CI 0.697-0.749) and good calibration. Patients with CVC-BSIs had significantly longer durations of catheterization, ICU stay, and hospital stay. CONCLUSIONS: This parsimonious model, calibrated with isotonic regression, effectively stratified CVC-BSI risk using early clinical indicators. Implementation can guide targeted infection prevention strategies in high-risk populations.