When Retrieval Helps and Distracts: Evaluating Evidence-Generating LLMs for Biomedical Claim Verification

2026-08-02

AI Summary

This study addresses the challenge of generating faithful, complete, and verifiable evidence for biomedical fact-checking, where existing approaches suffer from limitations in both evidence generation quality and retrieval effectiveness. The authors systematically evaluate various large language models—including instruction-tuned variants, PubMed-retrieval-augmented models, and classifiers—on the CARE-XAI benchmark. They introduce Bio-GRACE, a novel evaluation metric that measures the decision utility of retrieved evidence by normalizing against gold-standard references. Experimental results demonstrate that fine-tuned models achieve the best performance in evidence generation; PubMed retrieval benefits only claims aligned with PubMed content and may introduce noise for public health statements; and Bio-GRACE effectively exposes the shortcomings of conventional metrics based on recall and lexical overlap, underscoring the critical role of selective retrieval.

Abstract

Biomedical fact-checking systems must do more than predict whether a claim is supported, contradicted, or unaddressed: they should also produce evidence that is faithful, complete, and useful for verification. We study this evidence-generation setting on CARE-XAI, a unified benchmark spanning five biomedical and health fact-checking sources. We compare base instruction LLMs, PubMed retrieval-augmented LLMs, fine-tuned LLMs, label-only LLMs, and biomedical encoder classifiers under a shared evaluation protocol. Biomedical classifiers remain strongest for verdict-only prediction, while fine-tuned LLMs are the strongest evidence-generating systems. PubMed retrieval is mixed: it helps PubMed-aligned sources such as PubMedQA and SciFact, but can distract models on broader public-health claims. We introduce Bio-GRACE, a gold-reference-normalized diagnostic for measuring whether retrieved evidence recovers the decision benefit of reference evidence. Bio-GRACE shows that retrieval utility is source-dependent, motivates selective retrieval, and exposes why retrieval recall and lexical evidence overlap are insufficient for biomedical fact-checking.

Problem

• biomedical claim verification
• evidence generation
• retrieval-augmented LLMs
• fact-checking
• evidence faithfulness

Innovation

• evidence generation
• retrieval-augmented LLMs
• biomedical fact-checking
• Bio-GRACE
• CARE-XAI