S1329
Interdisciplinary - Education in radiation oncology
ESTRO 206
our current study — will further clarify the real-world applicability of these models in oncologic practice. References: 1. Alasker A, Alsalamah S, Alshathri N, et al. Performance of large language models (LLMs) in providing prostate cancer information. BMC Urol. 2024;24(1):177. Published 2024 Aug 23. doi:10.1186/s12894-024-01570-02. Lammert J, Dreyer T, Mathes S, et al. Expert-Guided Large Language Models for Clinical Decision Support in Precision Oncology. JCO Precis Oncol. 2024;8:e2400478. doi:10.1200/PO-24-004783. Klang E, Apakama D, Abbott EE, et al. A strategy for cost-effective large language model use at health system-scale. NPJ Digit Med. 2024;7(1):320. Published 2024 Nov 18. doi:10.1038/s41746-024-01315-1 Keywords: artificial intelligence, oncology, cost- efficiency Is seeing believing? How Virtual Reality improved patients' anxiety and perceived preparedness for Deep Inspiration Breath Hold radiation therapy Kathleene Dower 1,2 , Haryana M Dhillon 3 , Kate Mueller 4,5 , Moira O'Connor 2 , Julan Amalaseelan 1 , Huaqiong Zhou 6 , Georgia KB Halkett 2 1 Radiation Oncology, North Coast Cancer Institute, Lismore, Australia. 2 Faculty of Health Sciences, Curtin University, Perth, Australia. 3 School of Psychology, University of Sydney, Sydney, Australia. 4 Faculty of Health Sciences & Medicine, Bond University, Gold Coast, Australia. 5 Research Office, Northern New South Wales Local Health District, Lismore, Australia. 6 School of Nursing, Curtin University, Perth, Australia Purpose/Objective: Deep Inspiration Breath Hold radiation therapy (DIBH- RT) for breast cancer requires patient understanding and cooperation to achieve its benefits, but many patients find it conceptually challenging and stressful (1). Fear of treatment (2) and technical complexity often heighten anxiety and hinder performance (1). This pilot study evaluated whether a virtual-reality (VR) education package could reduce anxiety and improve understanding, confidence, and perceived preparedness for DIBH-RT. Material/Methods: Forty-one breast-cancer patients scheduled for DIBH- RT were enrolled in a prospective, single-arm, mixed- methods study at a regional cancer centre. The intervention included an animated explainer, immersive 360° simulations of CT-planning and treatment, and guided breath-hold coaching. State anxiety was measured using the STAI-6 short form across six timepoints. Open-ended survey questions Digital Poster Highlight 1796
the lowest cost and fastest responses, while ChatGPT- 5 demonstrated the highest Generative and Economic Efficiency, reflecting a balanced trade-off between performance and resource utilization. Claude Opus 4.1, despite its higher cost, remained competitive in response speed (Figure 1).In the stepwise analysis, cost differences were not significant (p = 0.066), but the diagnostic stage showed shorter response times (p = 0.019) and higher generative efficiency (p = 0.021) compared with the treatment and follow-up phases (Tables 1).
Conclusion: This study provides a preliminary yet objective evaluation of LLM performance in structured clinical workflows, highlighting model-specific strengths. Gemini 2.5 was the most cost- and time-efficient, while ChatGPT-5 delivered the best overall performance with superior Generative and Economic Efficiency. The diagnostic stage’s higher productivity likely reflects its focused analytical nature. These findings indicate that future LLM-based clinical decision-support systems should optimize not only for cost and speed but also for clinical accuracy and explainability (1-3). “Ongoing validation of clinical accuracy — already underway in
Made with FlippingBook - Share PDF online