How Brim Works: Turning Clinical Notes into Structured Data

Clinical notes hold some of the most valuable information in healthcare, but they’re notoriously hard to use. Diagnoses, procedures, medications, and timelines are often buried in unstructured text, making it difficult for researchers, registries, and health systems to extract the data they need. Traditional chart abstraction methods rely on humans manually combing through charts, a process that is accurate but painfully slow, expensive, and difficult to scale.

Brim changes that equation. Our AI-powered chart abstraction platform helps researchers and clinicians unlock data trapped in medical records, enabling faster, more accurate, and fully traceable insights. Here’s how it works.

Step 1: Upload Your Notes

‍

Every Brim project begins with the raw material: clinical notes. These might be physician progress notes, discharge summaries, pathology reports, or other types of documentation.

With Brim, you can securely upload de-identified patient notes into the platform. For institutions, Brim is deployed inside your firewall, ensuring that protected health information (PHI) never leaves your environment. That means you get all the benefits of AI abstraction while staying compliant with HIPAA, SOC 2 Type II, and institutional security requirements.

‍

Step 2: Define Your Variables

‍

Once your notes are in the system, the next step is deciding what data points you want to abstract: what we call variables. These can be anything you need, but might include:

• Patient demographics (age, sex, race/ethnicity)
  
  
• Clinical characteristics (diagnosis date, tumor type, disease stage)
  
  
• Treatments and outcomes (chemotherapy regimen, surgical procedures, response)
  

You can define variables in plain English, and Brim’s AI helps you translate them into structured instructions that the system can reliably extract.

Variables can also be imported directly from REDCap data dictionaries, other Brim projects, or shared protocols, making it easy to align your abstraction with existing research standards or registry definitions.

‍

Step 3: Generate and Review a Draft Abstraction

‍

This is where Brim gets really powerful. Brim runs your notes and set of variables through our AI-guided abstraction engine. Using your institution's large language model (LLM) and entirely within your firewall, Brim generates structured outputs for each data point.

• Abstraction is fast: tasks that take humans hours can be completed in seconds.
  
  
• Abstraction is accurate: Brim’s outputs are benchmarked against gold-standard datasets and often catch human errors.
  
  
• Abstraction is traceable: every extracted data point is linked back to its source text in the chart. That means no black box; you always know exactly where the data came from.
  
  

Researchers and curators can then review results in Brim’s Label Review interface, which shows each data point side-by-side with the original chart text. Accept, edit, or reject values with a click. Over time, your team can refine variables for even better performance.

‍

Step 4: Validate and Iterate

‍

Good research requires good data, and validation is baked into Brim.

With the Validation Tool, you can upload a comparison dataset (manual abstractions or a golden dataset) and instantly calculate agreement between Brim and the reference standard.

This enables you to:

• Track improvements over time as you refine your variables.
  
  
• Identify areas where human review remains critical.
  
  
• Build confidence with IRBs, clinical collaborators, and sponsors.
• ‍

Step 5: Integrate With Your Workflow

‍

Brim isn’t just about generating data—it’s about making that data useful. Once abstraction is complete, results can be:

• Exported to REDCap for downstream research workflows.
  
  
• Leveraged in clinical workflows like patient pre-screening, rescheduling, or gap-in-care identification via Brim’s API.
  
  
• Exported to CSV for further analysis and inclusion in research, registry, or other workflows.

Brim was designed to be domain flexible and user friendly. Whether you’re running a pediatric epilepsy case conference, screening patients for a cancer clinical trial, or tracking outcomes in an orthopedic study, Brim adapts to your context.

‍

Real-World Impact of Brim’s AI-Guided Chart Abstraction

‍

Brim is already powering over 100 projects at Vanderbilt and other leading institutions. Examples include:

• Clinical trial recruitment: Matching patients to trials based on nuanced eligibility criteria.
  
  
• Drug repurposing research: Identifying off-label usage patterns across populations.
  
  
• Quality improvement: Tracking post-surgical complications and care pathways.
  
  
• Registry submissions: Automating structured data abstraction for compliance.
  
  

Timing studies show Brim reduces abstraction time by 50–90%, depending on task complexity. Accuracy is on par with (and often exceeds) human inter-rater reliability. Most importantly, Brim enables health systems to scale output without scaling staff, freeing curators and clinicians to focus on higher-value work.

‍

Why Brim Works

‍

At its core, Brim works because it combines three critical ingredients:

1. AI Power & Scale: Large language models and NLP designed for complex, unstructured clinical data.
   
   
2. Human Oversight: Researchers and clinicians remain in the loop, ensuring every output is trustworthy and clinically meaningful.
   
   
3. Workflow Integration: Exports to REDCap and other downstream systems make data actionable.
   
   

This blend of automation, transparency, and flexibility makes Brim a trusted partner for researchers, clinicians, and institutions who need reliable structured data from messy medical charts.

‍

Conclusion

‍

Chart abstraction doesn’t have to be slow, expensive, or inconsistent. With Brim, you can transform unstructured clinical notes into accurate, structured, and traceable data, fueling everything from clinical trial enrollment to registry reporting to retrospective research.

Our mission is simple: to make data abstraction fast, accurate, and secure, so you can focus on discovery, not data entry.

Want to see Brim in action? Get in touch with us to learn how we can streamline your clinical workflows and accelerate your research.