Central line-associated bloodstream infections (CLABSI) are a significant concern within the healthcare sector. These preventable hospital-acquired infections can lead to prolonged hospital stays, increased healthcare costs, and heightened morbidity and mortality rates. In a crucial step towards improving patient safety and enabling early intervention strategies, innovative dynamic prediction models are being developed for continuous CLABSI risk monitoring.
Recent research conducted at the University Hospitals Leuven utilized the increasing availability of continuously updated electronic health records (EHR) datasets. Examining patient data dating from 2014 to 2020, five dynamic models were developed to predict the 7-day CLABSI risk. Each model accounted for competing events such as death, discharge, and catheter removal. The models varied in design and included a cause-specific model, two random forest models, and two XGBoost models. After development, the outcomes were combined using a superlearner model for optimal results.
The training set for model development consisted of 55,910 admissions, 61,628 catheter episodes while the test for temporal evaluation included 40,994 admissions and 44,544 catheter episodes. When considering individual models, one XGBoost model achieved the loftiest Area Under the Receiver Operating Characteristic curve (AUROC) of 0.748. However, the superlearner model proved most effective overall, providing improved discrimination and better calibration than individual models.
Despite the progress made through these models, their clinical utility currently appears to be limited. The performances of the models were seen to diminish over time. However, the development and application of these prediction models constitute a vital step forward in managing and reducing the incidence of hospital-acquired infections. This shows promise in improving patient safety and offers an example of how data-driven approaches can enhance healthcare interventions.
Source: https://bmcinfectdis.biomedcentral.com/articles/10.1186/s12879-025-10854-1