Dynamic risk prediction to reduce central line-associated bloodstream infections using electronic health records

Central line-associated blood stream infections (CLABSI) cause prolonged hospital stays, morbidity, and cost. We envisage an actionable trigger-based warning system that helps to reduce the incidence of CLABSI, and its negative impact on patients. We will develop a dynamic risk model that estimates CLABSI risk at any moment, using baseline characteristics at catheter placement and later observations during the patient stay (eg parenteral nutrition, admission to intensive care). We will compare regression and machine learning algorithms. Models will be developed on electronic health records (EHR) from 60,000 patient stays between 2014-2017 in Leuven. Validation will be performed on Leuven data from 2018-2020, as well as in hospitals from the Flemish Hospital Network and from the Netherlands. Through a decision analysis, we assess the model’s impact on clinical and financial outcomes. Finally, we aim to implement the model into the UZ Leuven EHR system.