MatchMiner-AI: Open-Source AI Tool for Cancer Clinical Trial Matching
MatchMiner-AI, a novel open-source artificial intelligence tool, is designed to enhance patient enrollment in cancer clinical trials by effectively matching individuals to appropriate trials. This system, created by researchers in collaboration with practicing clinical oncologists, utilizes open-weight large language models (LLMs) to analyze unstructured electronic health record (EHR) data and identify eligible patient populations from trial criteria. It incorporates embedding and re-ranking models for retrieving and prioritizing trial and patient recommendations. Trained on synthetic EHR data, its effectiveness was assessed through retrospective analyses of distillation fidelity, with a closed-source LLM serving as a benchmark for summarization and matching, alongside evaluations by oncologists. The findings, which demonstrate superior performance compared to baseline evaluations, are detailed in a paper available on arXiv (arXiv:2412.17228). The study emphasizes privacy-preserving and open-source methods to broaden access to trial opportunities, addressing the fact that fewer than 10% of adults with cancer participate in therapeutic trials. This initiative reflects the increasing integration of AI in healthcare to enhance patient access to clinical research.
Key facts
- MatchMiner-AI is an open-source, privacy-preserving AI tool for cancer clinical trial matching.
- It was co-developed with practicing clinical oncologists.
- It uses open-weight LLMs to summarize patient histories from unstructured EHR data.
- It extracts target populations from trial eligibility documents.
- Embedding and re-ranking models are used to retrieve and rank trial and patient suggestions.
- The tool was trained on synthetic EHR data.
- Evaluation included retrospective distillation fidelity, closed-source LLM judge, and oncologist review.
- Fewer than 10% of adults with cancer enroll in therapeutic trials.
Entities
—