Manual chart abstraction of unstructured data has created big bottlenecks in medical research – data doesn’t help researchers if it’s trapped in electronic notes. New software from Vanderbilt, Brim, makes chart abstraction up to 80% faster and often more accurate as well by leveraging AI. So far, it’s been used for drug repurposing, oncology clinical trials, genetics testing, emergency general surgery, pickleball injuries, and many more projects.
Brim is currently free for Vanderbilt faculty, staff, students, and postdocs; use this request form to get started. Those outside of Vanderbilt should request a demo to learn more about using Brim at their institution. Brim prioritizes data security, providing a fully containerized deployment and allowing you to Bring Your Own LLM (BYO LLM), so you can use your institution’s preferred, secure LLM with Brim running on top of it. This creates privacy and protection of health information.
Once you have access to Brim, the process is simple and designed for non-technical users:
- Import patient records from a simple CSV (which can be downloaded via SD Discover, Epic Clarity, or other EHR connectors).
- Define the data you want abstracted. You can do this in a fully manual way, or leverage the tool to generate and optimize draft variables and dependent variables.
- Use Brim to generate abstracted data points, covering hundreds of charts in minutes.
- Give feedback on these data points so Brim can optimize the variables until you reach the desired accuracy. Each data point is linked to evidence from the original text so you can check for hallucinations or false interpretations.
- When you’re satisfied with the extraction, export the data as a CSV file for further analysis.
Brim is managed by Dr. Fabbri in the Department of Biomedical Informatics at VUMC. The project has received support from ARPA-H. Brim is currently under active development, and feedback on the interface and desired functionality/features is very welcome!
