This study evaluates the effectiveness of clinical decision support systems powered by biomedical language models in enhancing antibiotic stewardship using electronic health records (EHRs). We focus on predicting the effectiveness of various antibiotics for individual patients, with a strong emphasis on model interpretability to understand decision-making processes.
- Biomedical Language Models: Utilization of state-of-the-art language models tailored for biomedical applications.
- Antibiotic Effectiveness Prediction: Targeted prediction for eight different antibiotics, assessing their suitability for patient treatments based on clinical data.
- Interpretability: Detailed analysis of model decision-making processes to identify strengths and limitations, aiding clinical understanding and application.
The study uses the MIMIC-IV-ED dataset, which includes comprehensive records from emergency department (ED) visits, providing a rich source for model training and evaluation.
The models demonstrate potential in accurately predicting antibiotic effectiveness, with insights into their operational strengths and limitations detailed in the analysis. Results suggest significant promise for AI in supporting clinical decisions, with necessary improvements for deployment in practical settings.
Contributions to the project are welcome. Please refer to CONTRIBUTING.md for guidelines on how to make contributions.
This project is licensed under the MIT License - see the LICENSE file for details.