Free-text evaluation of multimodal clinical question answering (QA) systems remains a central challenge in medical NLP due to the complexity of medical knowledge, the necessity of integrating visual and textual information, and the limitations of existing automatic evaluation metrics for open-ended outputs. In this work, we present a training-free, agentic evaluation framework that formulates response scoring as evidence-guided orchestration of components rather than a task requiring conventional end-to-end fine-tuning of underlying LLMs/VLMs. Our ReAct-style evaluator combines (i) structured reasoning, (ii) multimodal retrieval of similar encounters, (iii) auxiliary explainable feature-based regression models that provide numeric priors and human-interpretable signals, (iv) VLM-generated visual QA references for comparison, and (v) optional image augmentation tools. Unlike standard LLM-as-a-judge approaches that rely on direct generative scoring, our agent decomposes evaluation into modular stages of evidence acquisition, structured feature modeling, and integrative reasoning. We apply this architecture to the MEDIQA-EVAL shared task - a multimodal, multilingual clinical evaluation challenge that assesses system-generated answers for patient queries paired with images along multiple clinical quality dimensions. We report results across both English and Chinese tracks, comparing against baseline prompting methods, and discuss the feasibility and limitations of lightweight agentic systems for clinical QA evaluation.

Discharge instructions are patient-facing, safety-critical documents that guide medication use, follow-up care, and recovery after hospitalization. Because they must synthesize information across the clinical record and often include post-discharge guidance not stated verbatim in the EHR, they are a difficult target for clinical text generation. In this work, we study discharge instructions in MIMIC-IV through a grounding-first lens. Using two LLMs, we decompose each discharge instruction into medically relevant statements and verify them against the Electronic Health Record (EHR). We find that discharge instructions for Surgical admissions are much longer, averaging roughly 24–25 statements per admission versus 11–12 in Non-Surgical cases, while supported content remains similar in absolute count. The additional Surgical content is dominated by statements that are not directly stated in the record or require clinically plausible extrapolation. Through this analysis, we advocate for better grounding and completeness evaluations at a fine-grained level, establishing a foundational step toward safer and more reliable discharge-instruction generation.