Abstract: Optimal management of pediatric central-line associated bloodstream infections (CLABSI) is often uncertain. We report a machine-learning (ML) approach to predicting two causes of unsuccessful antimicrobial therapy –central venous catheter (CVC) removal and infection recurrence. 969 CLABSI events were identified by retrospective chart review and used to train several ML models. Best ROC AUC were 0.82 and 0.77 for recurrence and CVC removal models respectively. Accurate predictions of outcomes could improve clinical management of CLABSI.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: The attendee will be able to describe the clinical decision of CVC salvage versus removal in pediatric CLABSI, and the manner in which machine learning models may improve this decision. This serves as an example of a machine learning approach that could be applied to other areas of clinical uncertainty.

Authors:

Lorne Walker (Presenter)
UPMC Children's Hospital of Pittsburgh

Andrew Nowalk, UPMC Children's Hospital of Pittsburgh
Shyam Visweswaran, University of Pittsburgh