ESTRO 2025 - Abstract Book

S2149

Interdisciplinary – Education in radiation oncology

ESTRO 2025

170

Digital Poster Large language models for enhancing pretreatment education in pediatric radiation oncology Dominik Wawrzuta 1 , Aleksandra Napieralska 2,3,4 , Katarzyna Ludwikowska 1 , Laimonas Jaruševičius 5 , Anastasija Trofimoviča-Krasnorucka 6,7 , Gints Rausis 6 , Agata Szulc 8 , Katarzyna Pędzwiatr 1 , Kateřina Poláchová 9,10 , Justyna Klejdysz 11,12 , Marzanna Chojnacka 1 1 Department of Radiation Oncology, Maria Sklodowska-Curie National Research Institute of Oncology, Warsaw, Poland. 2 Radiotherapy Department, Maria Sklodowska-Curie National Research Institute of Oncology, Gliwice, Poland. 3 Department of Oncology, Maria Sklodowska-Curie National Research Institute of Oncology, Cracow, Poland. 4 Faculty of Medicine & Health Sciences, Andrzej Frycz Modrzewski Krakow University, Cracow, Poland. 5 Oncology Institute, Lithuanian University of Health Sciences, Kaunas, Lithuania. 6 Department of Radiation Oncology, Riga East University Hospital, Riga, Latvia. 7 Department of Internal Diseases, Riga Stradiņš University, Riga, Latvia. 8 Department of Radiation Oncology, Lower Silesian Center of Oncology, Pulmonology and Hematology, Wroclaw, Poland. 9 Department of Radiation Oncology, Masaryk Memorial Cancer Institute, Brno, Czech Republic. 10 Department of Radiation Oncology, Masaryk University, Brno, Czech Republic. 11 Department of Economics, Ludwig Maximilian University of Munich, Munich, Germany. 12 ifo Institute, ifo Institute, Munich, Germany Purpose/Objective: Pediatric radiotherapy patients and their parents often become aware of the need for treatment early on, but consultations with radiation oncologists typically occur later in the care process. Consequently, they search for information online, often encountering unreliable sources [1]. Large language models (LLMs) have the potential to serve as an educational pretreatment tool, providing reliable answers to their questions [2,3]. Our objective was to evaluate the quality of the responses provided by LLMs for patients and parents seeking information on pediatric radiation oncology. Material/Methods: We collected 80 pretreatment questions related to radiotherapy from pediatric patients and their parents. Responses were generated using GPT-3.5, GPT-4, and a fine-tuned GPT-3.5 model, specifically trained using pediatric radiotherapy guides from various institutions. In addition, a radiation oncologist (RO) provided expert answers as the gold standard. Thus, four distinct answers were obtained for each of the 80 questions. A multi-institutional group of nine pediatric radiotherapy experts conducted a blind review, evaluating responses based on reliability, conciseness, and comprehensibility using Likert scales (1-5 for reliability, 1-3 for conciseness, and 1-3 for comprehensibility). Each response was reviewed by five different experts. A composite score (scale 0-1) was calculated by averaging the standardized scores across the three dimensions. The flowchart of the methodological steps followed in the study is presented in Figure 1.