2026· Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks· pp. 93-102· 0 citations· 23 references
TL;DR
This work introduces MedFaith-F1, a category-level faithfulness metric over four clinically salient categories: clinical problems, medications, procedures, and follow-up intent, and proposes a hybrid Evidence and Knowledge-Grounded Retrieval-Augmented Generation ( EKG-RAG), an evidence and knowledge-grounded framework combining hybrid retrieval over PubMed and MedlinePlus resources with UMLS-aligned ontology grounding.
Abstract
Medical query rewriting transforms verbose consumer health questions into concise clinical queries, a critical step in health information retrieval. Large language models (LLMs) perform well on this task by standard metrics, yet high ROUGE or BERTScore does not guarantee preservation of clinical content. To address this issue, we introduce MedFaith-F1 , a category-level faithfulness metric over four clinically salient categories: clinical problems, medications, procedures, and follow-up intent. We further propose a hybrid Evidence and Knowledge-Grounded Retrieval-Augmented Generation ( EKG-RAG ), an evidence and knowledge-grounded framework combining hybrid retrieval over PubMed and MedlinePlus resources with UMLS (Unified Medical Language System)-aligned ontology grounding. Evaluating LLMs LLaMA-3 and Qwen2.5 across zero-shot, few-shot, and QLoRA settings on MeQSum and medical question-pair (MQP) datasets revealed that base models exhibit category-level faith-fulness failure rates (CHR) exceeding 40%, invisible to standard metrics, while EKG-RAG with QLoRA reduces CHR to 26.75%, achieving MedFaith-F1 of 0.73. Our findings call for faithfulness-aware evaluation in clinical query rewriting, and MedFaith-F1 provides a reproducible step in that direction.
The proposed framework offers a practical and scalable approach to mitigating hallucinations without requiring task-specific fine-tuning, highlighting the potential of retrieval-augmented approaches for trustworthy artificial intelligence (AI)-assisted healthcare applications.
Medical visual question answering (Med-VQA) is often assumed to require medical fine-tuning, large models, or complex multi-agent pipelines. We revisit this assumption with \textbf{MedProb}, a lightweight probing framework that predicts multiple-choice Med-VQA answers from frozen VLM representations without free-text generation. Across PATH-VQA, SLAKE, and VQA-RAD, MedProb recovers substantially more answer-relevant signal than prompting and performs stronger than medical VLMs and agentic systems. Probing also reduces the apparent gap between small and large models compared to prompting, suggesting that smaller VLMs contain more recoverable Med-VQA signal than generation-based evaluation reveals. Across 14 matched general-purpose and medical VLM pairs, medical adaptation does not consistently improve this linear decodability. Finally, free-text generation exhibits an answer-position bias of up to 10 percentage points, whereas MedProb also has positional bias, however, it is impacted differently than prompting. Our main results target the multiple-choice/multiclass Med-VQA setting; we additionally show the probe can be extended to open-ended generation via a rejection-sampling scoring procedure.
Erfan Nourbakhsh, Ke Yang, Anthony Rios· 0 citations
While general-purpose large language models (LLMs) demonstrate remarkable capabilities, their clinical application demands rigorous adaptation to ensure safety and accuracy. This review presents a comprehensive framework for transforming LLMs into trustworthy medical specialists. We detail three core knowledge-injection strategies-(1) static embedding to internalize foundational biomedical knowledge; (2) behavioral alignment to enforce clinical safety and verifiable diagnostic logic; and (3) dynamic injection, such as retrieval-augmented generation, for real-time evidence grounding-together with multimodal integration as a complementary perception-injection paradigm extending the input space beyond text to imaging, biosignals, and tabular data. Building on these strategies, we further explore the evolution toward agentic AI systems that orchestrate them for autonomous, collaborative clinical decision-making. Finally, we discuss critical challenges, including model calibration, resource constraints, standardized reporting, and robust safety protocols. Combining these complementary strategies is essential for developing deployable, domain-specialized clinical AI systems.
Background/Objectives: Large language models (LLMs) show promise for clinical decision support, yet their accuracy in interpreting specialized medical guidelines remains uncertain. Retrieval-augmented generation (RAG) may enhance performance by grounding responses in authoritative knowledge bases. This study aimed to compare the accuracy, comprehensiveness, and safety of RAG-enhanced versus standard LLMs for answering clinical questions derived from the German S3 guideline for oral cavity carcinoma. Methods: We conducted a prospective, single-blind benchmark study evaluating six LLMs: one RAG-enhanced model (Custom GPT with guideline access), one consensus-based model (ConsensusGPT), and four standard models (DeepSeek-V3.2, Mistral Small 3.2, Qwen3-Next-80B, GPT-OSS-120B). Fifty clinical questions covering 17 guideline domains were presented to each model three times, yielding 900 evaluations. Three expert reviewers assessed responses using 5-point Likert scales for accuracy, comprehensiveness, and clarity, under a single-blind procedure, the effectiveness of which was tested by a pre-specified manipulation check. We then ran a paired within-model experiment in which each base model was queried with and without guideline access through a transparent, openly released retrieval pipeline, and scored every response with a condition-blind automated judge alongside deterministic retrieval metrics computed from the logs. Secondary outcomes included hallucination rates and guideline citation behavior. Inter-rater reliability was assessed using intraclass correlation coefficients (ICCs). Results: In a paired within-model design that held each base model fixed, adding transparent guideline retrieval improved accuracy—significantly in the three weaker open-weight models (Mistral, Qwen3, and GPT-OSS) and directionally in the already-strong DeepSeek and GPT-5 bases. Because a pre-specified blinding check found that experts could still identify retrieval-augmented answers with 98.5% accuracy, we anchored causal interpretation on measures that do not depend on the human raters, ranked by their independence: deterministic, log-derived retrieval metrics first, and then an automated, condition-blind LLM judge, whose agreement with the experts (Spearman ρ = 0.81, 95.7% within-one agreement) establishes shared calibration rather than independence from their bias. Deterministically from the retrieval logs, citation groundedness rose from 0% to 51–89% and retrieval recall@5 was 92%. On the judge, content-level hallucination fell from 42% to 4% and accuracy rose by a pooled +0.64 points (95% CI 0.47–0.80); the accuracy gain persisted after adjustment for response length (+0.48, 95% CI 0.22–0.73), which retrieval shortened rather than lengthened. The accuracy gain was large for weaker base models and small or non-significant for already-strong ones, whereas the hallucination and auditability gains were consistent across all models. The human ratings reproduced the judge’s accuracy effect (+0.61, 95% CI 0.49–0.74), and GPT-5 run through the transparent pipeline showed no significant difference from the proprietary Custom GPT (judge accuracy 4.48 vs. 4.58). Conclusions: Guideline retrieval yields a reproducible, largely base-independent improvement in the safety and auditability of LLM answers to clinical guideline questions, with accuracy gains concentrated in weaker base models. Because retrieval-augmented answers are recognizable to experts, rigorous evaluation should rely on rater-independent measures, and residual hallucination continues to require human oversight.
Andreas Vollmer, Lara Schorn, Felix Schrader et al.· Diagnostics· 0 citations
Understanding medical search intent is critical for improving biomedical information retrieval, particularly for identifying evidence-seeking queries. However, existing resources are limited in scale and lack structured annotation aligned with clinical evidence needs. We present MedQueryIntent, a large-scale benchmark dataset consisting of a human-annotated real-world query set and an LLM-augmented evidence query set. The human-annotated portion, MedQueryIntent-Real, contains 12,000 medical queries collected from Trip and PubMed and is labeled with a hierarchical taxonomy distinguishing evidence-seeking and non-evidence-seeking intent. The annotation follows structured guidelines based on clinical research frameworks (e.g., PICO, PEO, PCC), ensuring consistent and interpretable labeling. To address the limitations of short and ambiguous real-world queries, we further construct MedQueryIntent-Synth, an LLM-augmented evidence query dataset containing 7,633 unique queries, generated by transforming systematic review content into natural query forms via controlled prompting. Together, the two resources form the complete MedQueryIntent benchmark with 19,633 queries. We benchmark multiple models, including biomedical encoders and large language models, demonstrating that MedQueryIntent provides a challenging and realistic testbed. Results show that incorporating MedQueryIntent-Synth improves classification performance and robustness. Our dataset supports research in medical query understanding and downstream applications such as evidence retrieval and clinical decision support. The dataset is publicly available at https://github.com/yingchengsun/MedQueryIntent.
S. Schnell, Supriya Kottam, Yingcheng Sun· International Conference on...· 0 citations
The results show that QA evaluation should account for how inferential relations compose across a reasoning trace, rather than relying only on final answers or LLMs as verifiers.
Guneet Singh Kohli, Yuxiang Zhou, M. Schlichtkrull et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
