182P Evaluating large (LLM) versus small language models (SLM) in summarizing real-life oncology clinical narratives in Spanish

Abstract

to determine trajectories of diarrhea, constipation, nausea/vomiting, and appetite loss up to 6 years post-diagnosis of CC survivors by sex.Methods: Stage I-III CC survivors treated at the Klinikum Stuttgart between 2003 and 2021 who participated in the "Benchmarking in der Patientenversorgung" study of the Onkologischer Schwerpunkt Stuttgart e. V. were included.Participants completed the European Organization for Research and Treatment of Cancer Quality of Life Core-30 (QLQ-C30) questionnaire annually.For the current analysis, we selected survivors who had completed QLQ-C30 surveys between <6 months and 3.5 years after diagnosis.Group Based Trajectory Modeling identified longitudinal patterns in the QLQ-C30 scales.Models were independently calibrated for each sex and outcome variable. Results:In total, 123 women and 158 men, with median age of 67 were included.Mean posterior probabilities per group were at least 0.83 for all models.Five trajectories for diarrhea were identified for both sexes.For constipation, nausea/ vomiting, and appetite loss, identified trajectories differed for females (n=4/4/5) and males (n=5/3/4), respectively.Dominant pattern across all models was the persistent low group.41% of females and 37% of males were in persistent low or medium low groups for all four scores.Females were more likely to be grouped in persistent high trajectories than males (diarrhea: 9% vs 5%; constipation: 11% vs 6%), with 14% in both sexes reporting more problems with diarrhea over time.Males presented with more variable group patterns than females.Conclusions: Machine learning tools can categorize and group trajectory patterns of longitudinal quality of life scores, thus automatically identifying patients with longterm high or increasing scores.Additionally, trajectory groups of men and women may differ, indicating the need for more research and sex-specific post-treatment support.

Ähnliche Arbeiten