Create
Design Project Gallery
Deep Learning for Early Prediction of Multi-drug-Resistant Organisms in the Emergency Department
- Program: Biomedical Engineering
- Course: EN.580.480 Precision Care Medicine
- Year: 2025
Project Description:
The presence of multidrug resistant organisms (MDROs) and prescribing the appropriate antibiotics to treat them poses a significant challenge in Emergency Departments. Using room and healthcare worker interaction data from over 128,000 patients across six hospitals in the Johns Hopkins Medical system, we built contact networks to encompass MDRO infection spread. We then trained several edge-aware graph neural networks to predict patient MDRO status using the network topology, patient electronic health record data, and comorbidities to identify high-risk patients. Our approach extends on previous work by incorporating temporal patient movement patterns and heterogeneous clinical features (e.g., prior antibiotic use, renal function) to enhance risk stratification. The model aims to reduce inappropriate empirical therapy by flagging patients likely to harbor MDROs, enabling targeted broad-spectrum antibiotic use.
Project Photo:
Our logo features a ptarmigan formed by interconnected nodes and lines, representing patients connected by their locations and healthcare workers.
