According to arxiv.org, researchers have proposed a lightweight framework that combines retrieval-augmented generation (RAG) and large language models for scalable patient-trial matching. The work addresses challenges in matching patients with clinical trials by processing long electronic health records (EHRs) and complex eligibility criteria.
The framework separates two key components: RAG identifies clinically relevant segments from long EHRs to reduce input complexity, while LLMs encode these selected segments into representations that are refined through dimensionality reduction and modeled using lightweight predictors, according to the paper published on arxiv.org.
The researchers evaluated their approach on multiple public benchmarks including n2c2, SIGIR, TREC 2021/2022, and a real-world multimodal dataset from Mayo Clinic (MCPMD). According to the arxiv.org paper, results showed that retrieval-based information selection “significantly reduces computational burden while preserving clinically meaningful signals.” The study found that frozen LLMs provide strong representations for structured clinical data, whereas fine-tuning is essential for unstructured clinical narratives.
According to the paper, the lightweight pipeline achieves performance comparable to end-to-end LLM approaches “with substantially lower computational cost” compared to existing methods that rely on full-document processing, which are computationally expensive.
Separately, arxiv.org published research on Cooperative Retrieval-Augmented Generation (CoRAG), which treats rerankers and generators as peer decision-makers to improve generation stability by jointly optimizing toward shared task objectives.