Exploring the Utility of ChatGPT‐4o and DeepSeek‐R1 for Multidisciplinary Uro‐Oncology Tumor Boards: A Tumor Type–Specific Analysis

OpenAlex İngilizce

Introduction This study demonstrates the potential benefits of large language models (LLMs) such as ChatGPT‐4o and DeepSeek‐R1 in supporting multidisciplinary uro‐oncologic tumor boards (MUTBs) discussing complex urological‐oncological cases. In particular, DeepSeek‐R1 has demonstrated better performance than ChatGPT‐4o in specific types of urological malignancies and specific treatment protocols. Objective This study examined the agreement between the AI‐based systems ChatGPT‐4o and DeepSeek‐R1 and the decision making of a MUTB, which is a team of experts who assess uro‐oncological cases. Methods A total of 97 uro‐oncological cases covering a period of one year were retrospectively evaluated. The treatment plans for the cases were determined using traditional tumor board methods and separately presented to ChatGPT‐4o and DeepSeek‐R1. Comprehensive medical records were created for each case, and the planned treatments were classified as surgical intervention, radiotherapy/chemotherapy (RT‐CT), advanced imaging, or follow‐up. Agreement between the board and AI decisions was analyzed using Cohen’s kappa and with precision, recall, and F1 scores. Results In prostate cancer cases, DeepSeek‐R1 showed substantial agreement ( κ = 0.787, p < 0.001) with the board decisions, while ChatGPT‐4o showed moderate agreement ( κ = 0.593, p < 0.001). For renal malignancies, DeepSeek‐R1 showed moderate agreement ( κ = 0.603, p < 0.001), while ChatGPT‐4o showed slight agreement ( κ = 0.172, p < 0.001). In bladder cancer cases, DeepSeek‐R1 showed moderate agreement ( κ = 0.604, p < 0.05), but ChatGPT‐4o did not show statistically significant agreement ( κ = 0.217, p > 0.05). When all cases were considered together, DeepSeek‐R1 showed high agreement, while ChatGPT‐4o showed moderate agreement (DeepSeek‐R1: κ = 0.773, p < 0.001; ChatGPT‐4o: κ = 0.608, p < 0.001). Conclusion Our findings suggest that DeepSeek‐R1’s decision support performance is better than that of ChatGPT‐4o in various oncological scenarios. The results highlight the potential for publicly available AI models to be used in complex clinical fields, such as oncology.