This research introduces a novel framework for clinical named entity recognition without requiring domain-specific training data.

Key Findings:

• Open-source NER-specific LLMs show variable performance across different clinical entities
• Entity decomposition with filtering significantly improves extraction accuracy
• Zero-shot approach eliminates need for labeled clinical data
• Framework demonstrates practical utility for medical information extraction

Why It Matters: Accurate entity recognition in clinical narratives is crucial for healthcare data structuring, decision support, and research. This approach makes advanced NLP accessible for medical applications without costly proprietary models or extensive training data.

Entity Decomposition with Filtering: A Zero-Shot Clinical Named Entity Recognition Framework