Clinical nursing documentation contains detailed observational information that is essential for patient monitoring and clinical decision-making, yet this information is predominantly recorded in free-text form. The MEDIQA-SYNUR shared task addresses this challenge by requiring systems to extract structured nursing observations from clinical transcripts under strict constraints on evidence grounding and value normalization. In this work, we present a multi-agent large language model (LLM)–based system for the MEDIQA-SYNUR task. We utilize the Llama3 open source LLM for this purpose for ease of local deployment within hospital digital infrastructure. Our system decomposes the extraction process into specialized agents responsible for schema-guided extraction, rule-based validation, and precision-oriented filtering. Starting from a baseline multi-agent pipeline, we conduct a systematic error analysis over the entire development set, examining all false positive and false negative predictions. Our final configuration, selected after extensive exploration and error analysis, combined transcript segmentation, the precision agent, and a suppression table derived from development-set analysis. On the development set, this setup achieved an F1 score of 0.6930 (precision = 0.6427, recall = 0.7518). Applying the same configuration directly to the test set, without any additional tuning, yielded an F1 score of 0.5923 (precision = 0.5292, recall = 0.6725). These results represent the most effective balance of precision and recall achieved through our iterative refinements and reflect the final state of the system as submitted for the competition

Extracting structured cancer registry information from pathology and medical reports is challenging due to heterogeneous reporting styles and implicit clinical reasoning. We propose a modular multi-agent framework that decomposes registry abstraction into semantic chunking, retrieval, field-specific extraction, validation, evaluation, and aggregation stages. The dataset includes 818 annotated cancer cases from Sultan Qaboos University Hospital. Evaluation in this study focuses on breast (n=454) and colorectal (n=174) reports across grade, morphology, TNM staging, and laterality extraction tasks. The framework is compared against prompt-based LLaMA 3.3 baselines using accuracy and weighted/macro F1-score metrics. The proposed framework improved performance in context-dependent tasks, particularly grade extraction, where weighted F1-score increased from 0.71 to 0.78 for breast cancer and from 0.56 to 0.67 for colorectal cancer. Improvements were also observed for colorectal laterality extraction. For other extraction tasks, particularly highly structured tasks such as TNM staging and morphology extraction, the multi-agent framework achieved performance comparable to direct prompting. Although the baseline achieved slightly higher average weighted F1-scores overall, the proposed framework provides improved modularity, traceability, and pipeline-level interpretability through explicit intermediate reasoning stages, supporting error analysis and future clinician-guided refinement.