‍Background

‍

A U.S.-based academic research group conducting retrospective studies on drug repurposing needed to extract clinical variables from a large set of de-identified patient records. Their goal: identify potential new uses for existing medications, using real-world data to surface meaningful insights.

‍

The Challenge

‍

The team needed to determine whether patients had received antipsychotics, undergone EEGs, or had brain imaging — all from de-identified charts. Manual chart review was time-consuming and required highly trained personnel. Scaling up was costly and impractical.

‍

The Solution

‍

The team used Brim to semi-automate the abstraction process. Brim’s LLM-powered pipeline was used to extract and optimize key variables, generate structured labels, and facilitate final review — significantly reducing human effort and improving consistency.

‍

The Results

‍

-4x efficiency. The team was able to review all of the charts in a quarter of the time of manual review.

- Increased accuracy. In cases where Brim and manual review disagreed, Brim was correct in 2 out of 3 discrepancies, identifying evidence missed during manual review.

- Uniformity of abstraction. This team found that Brim was able to apply instructions uniformly across all records, minimizing the chance of introducing reviewer fatigue bias.

‍

Conclusion

‍

Brim enabled this team to complete chart review in a quarter of the time, with comparable accuracy to manual review -- and in some cases, better. For retrospective studies like drug repurposing, this means faster discoveries, lower costs, and the ability to scale research without scaling staff.

‍

“We appreciated the objectivity and capability of Brim compared to our manual reviews – it applies instructions uniformly across all records (minimizing the chance of introducing reviewer fatigue bias)."
- Researcher, Drug Repurposing Team

‍