JISARA

Abstract: The rapid growth of data in the health sector has made it crucial to communicate essential information quickly and succinctly. The vast amount of textual data from electronic health records tends to overwhelm healthcare professionals which reduces the time they can dedicate to patient care. This massive amount of complex qualitative data causes physicians to struggle with the decision-making process which had traditionally relied on human evaluation. This study addresses the urgent need for effective summarization of health records to improve patient outcomes and clinical decision-making. We highlight the use of large language models (LLMs) to produce concise summaries of patients' medical oncology reports. Specifically, we utilized pre-trained transformer models, including BART, T5, and Pegasus, to summarize patient clinical notes. The performance of these models was evaluated using BLEU, ROUGE, and BERT scores on CORAL expert-curated medical oncology reports that were de-identified using Philter. The results show that the BART and T5 models performed the best, with the generated summaries being shorter than the original oncology reports. This approach reduces information overload and enhances patient care by providing concise and informative summaries.

Download this article: JISARA - V18 N2 Page 20.pdf

Recommended Citation: Izuchukwu , C., Wimmer, H., Rebman Jr., C.M., (2025). A Comparison of Large Language Models for Oncology Clinical Text Summarization. Journal of Information Systems Applied Research and Analytics 18(2) pp 20-29. https://doi.org/10.62273/IGMU6476