SRS229 - Human <i>versus</i> machine: concordance and conflict in spinal oncology management

Abstract Background Spinal metastases are a common complication in cancer patients, requiring multidisciplinary team (MDT) decision-making guided by frameworks like the Spine Instability Neoplastic Score (SINS). This study evaluates the potential of ChatGPT to complement MDT workflows by analyzing concordance rates and identifying areas for improvement. Methods A retrospective analysis of 100 spinal oncology cases was conducted. Each case was anonymized and presented to both ChatGPT and the MDT using identical summaries. Concordance was assessed with Cohen’s Kappa, and discordant cases were analyzed for trends in decision divergence. Results ChatGPT's recommendations aligned with MDT decisions in 86% of cases, with substantial agreement (Cohen’s Kappa = 0.66). Discordant cases (14%) predominantly involved younger patients and rare tumors, where ChatGPT prioritized localized factors such as high SINS scores. Conversely, MDT decisions emphasized systemic prognosis, favoring palliative care in these cases. Conclusions ChatGPT demonstrates significant potential as a decision-support tool in spinal oncology, particularly in cases with straightforward surgical indications. However, discordance in complex cases underscores the need for integrating systemic prognostic tools and dynamic data. These findings highlight ChatGPT’s promise in enhancing clinical workflows and the importance of future refinements.