Related Experiment Video
Updated: Jun 19, 2026
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
BRIDGE: benchmarking large language models for understanding real-world clinical practice texts
Jiageng Wu1, Bowen Gu1, Ren Zhou2
1Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA.
Nature Biomedical Engineering
|June 17, 2026
Summary
A new benchmark, BRIDGE, evaluates large language models (LLMs) on real-world clinical data. It shows performance varies, with open-source LLMs matching proprietary ones and updated general models outperforming older specialized ones.
Area of Science:
- Artificial Intelligence
- Medical Informatics
- Natural Language Processing
Background:
- Large language models (LLMs) show potential in medicine, but real-world clinical data benchmarking is lacking.
- Existing benchmarks often use artificial questions or limited datasets, not reflecting clinical practice complexity.
- There's a need for comprehensive evaluation of LLMs on diverse, real-world clinical text.
Purpose of the Study:
- To introduce BRIDGE, a multilingual benchmark for evaluating LLMs on real-world clinical data.
- To assess the performance of various LLMs across different languages, tasks, and clinical specialties.
- To provide a resource for advancing LLM development in clinical text understanding.
Main Methods:
- Developed BRIDGE, a benchmark with 87 tasks from 59 real-world clinical data sources in 9 languages.
- Included 8 task types (e.g., triage, diagnosis, coding) across 14 clinical specialties.
- Evaluated 95 LLMs using multiple inference strategies.
Main Results:
- Significant performance variations were observed across LLM sizes, languages, tasks, and specialties.
- Open-source LLMs demonstrated performance comparable to proprietary models.
- Medically fine-tuned models on older architectures underperformed updated general-purpose LLMs.
Conclusions:
- BRIDGE offers a robust evaluation framework for LLMs in clinical settings.
- The benchmark highlights the need for continuous development and evaluation of LLMs for real-world medical applications.
- BRIDGE and its leaderboard serve as crucial resources for the research community.
