Photo Credit: Mohammed Haneefa Nizamudeen

The following is a summary of “Hospital-wide, dynamic, individualized prediction of central line-associated bloodstream infections-development and temporal evaluation of six prediction models,” published in the April 2025 issue of BMC Infectious Diseases by Albu et al. 

Central line-associated bloodstream infections (CLABSI) were preventable hospital-acquired infections and predicting them helped improve early intervention strategies and enhance patient safety.  

Researchers conducted a retrospective study to develop and temporally assess dynamic prediction models for continuous monitoring of CLABSI risk.  

They used data from individuals who were hospitalized with central catheters admitted to University Hospitals Leuven between 2014 and 2017 to develop 5 dynamic models (a landmark cause-specific model, 2 random forest (RF) models, and 2 XGBoost models) for predicting 7-day CLABSI risk, accounting for competing events (death, discharge, and catheter removal). The predictions from these models were combined by a super learner model. All models were temporally analyzed on data from the same hospital from 2018 to 2020 via performance metrics for discrimination, calibration, and clinical utility.  

The results showed that among 61,629 catheter episodes in the training set, 1,930 (3.1%) resulted in CLABSI, while 1,059 (2.4%) of 44,544 catheter episodes in the test set experienced CLABSI. Of the individual models, 1 XGBoost model achieved the highest Area Under the Receiver Operating Characteristic curve (AUROC) of 0.748. Calibration was good for risks up to 5%, but the cause-specific and XGBoost models overestimated higher predicted risks. The super learner model showed a slight improvement in discrimination (AUROC 0.751) and better calibration than the cause-specific and XGBoost models but was outperformed by the RF models. The models had clinical utility for supporting standard care interventions (risk thresholds of 0.5–4%) but not for advanced interventions (thresholds of 15–25%).  

Investigators concluded that hospital-wide CLABSI prediction models offered clinical utility based on medium-risk thresholds, but their clinical utility was limited as the model performance deteriorated over time. 

Source: bmcinfectdis.biomedcentral.com/articles/10.1186/s12879-025-10854-1