Leveraging Large Language Models for Automated Extraction of Information from Provider Notes

Large amounts of health care data remain in an unstructured format in documents from healthcare professionals. Even after text is extracted using optical character recognition (OCR), transforming the contained information into usable format for clinical and research use remains burdensome. Our study aims to explore the usage of large language models (LLMs) to extract and structure information from provider notes in electronic medical records (EMRs). We are focusing on a cohort of patients with head and neck cancer that underwent radiation and/or chemo-radiation therapies. These treatments are often associated with a range of severe and chronic toxicities (mucositis, dysphagia, aspiration, feeding tube use, etc) that can impact patient quality of life and long-term health. By using LLMs we aim to identify and classify treatment related side effects each patient has experienced as well as their respective severities. Preliminary results will summarize the types and frequency of toxicities stratified by different head and neck cancers and demographics.