Faithful by Design: Evaluating and Improving LLM-Generated Clinical Trial Summaries for Multi-Stakeholder Audiences
Quick Answer
This study evaluates the faithfulness of LLM-generated clinical trial summaries using a benchmark framework across three stakeholder audiences.
Quick Take
Models like GPT-4o, Claude Sonnet 4.6, and Gemini 2.5 Flash showed unsupported claims as a major failure mode, with improvements noted when using a knowledge-graph-augmented retrieval system, particularly in NLI-based faithfulness scores.
Key Points
- Introduced a benchmark framework with 200 stratified trials for evaluating summaries.
- GPT-4o, Claude Sonnet 4.6, and Gemini 2.5 Flash scored poorly on unsupported claims.
- Knowledge-graph-augmented retrieval improved NLI-based faithfulness scores significantly.
- Improvements varied by model, with GPT-4o reducing contradictions effectively.
- Study highlights risks of hallucination in clinical trial summaries for stakeholders.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Large language models are increasingly used to summarize clinical trial results for healthcare providers, patients, and payers, but their tendency to hallucinate poses significant risks in this high-stakes context. This study introduces a benchmark evaluation framework for measuring the faithfulness of LLM-generated clinical trial summaries across three stakeholder audiences. The framework consists of 200 stratified trials drawn from the Aggregate Analysis of this http URL database, evaluated using audience-specific prompt templates and a six-dimension faithfulness annotation schema. Baseline measurements were established for GPT-4o, Claude Sonnet 4.6, and Gemini 2.5 Flash across 1,800 generated summaries scored using a cross-encoder natural language inference (NLI) model. Unsupported Claims was identified as the dominant failure mode across all three models, with a mean annotation score of 1.55 out of three. A knowledge-graph-augmented retrieval system was developed and evaluated against the baseline, producing statistically significant improvements in NLI-based faithfulness scores (entailment +0.0125, faithfulness +0.0130, p < 0.0001). Improvement pathways were model-dependent, with GPT-4o improving primarily through contradiction reduction while Claude Sonnet 4.6 and Gemini 2.5 Flash improved through increased entailment.
| Comments: | 8 pages, 8 figures |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.09932 [cs.CL] |
| (or arXiv:2607.09932v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.09932 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Robert Williams [view email]
[v1]
Fri, 10 Jul 2026 19:29:26 UTC (1,810 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.