Integrating Structured and Unstructured Data for Pulmonary Embolism Identification: Implications for Natural Language Processing (NLP) and Large Language Models (LLMs) in Thrombosis Research