Harnessing NLP and Large Language Models for Pattern Discovery and Information Extraction in Electric Health Reports

In this work, we report on a series of natural language processing tools and models to improve the efficiency and accuracy of information discovery from clinical trials and pharmacological studies. Our main contributions are: 1. The development of an open-source platform Tri-AL that • Enables dynamic tracking of clinical trials information over time, • Excels in data visualization and user interaction with a particular emphasis on enhancing the analysis and representation of race and ethnicity data to foster equity in clinical research, and • Includes a predictive model utilizing machine learning to decipher drug mechanisms of action. 2. Heterogeneous Graph Neural Network for Gene-Chemical Entity Relation Extraction: We created a supervised deep learning model that adapts a heterogeneous Graph Neural Network to extract gene-chemical components. This model augments word representations using message passing that accurately identifies gene-chemical named entities and their relationships class. 3. Bipartite Graph Model for Evaluating Summarization Performance: We proposed a bipartite graph model to evaluate the performance of large language models in summarizing clinical trials. This model provides a robust framework to assess the accuracy and effectiveness of automated summarization tools in the medical domain.