527P Title results of the safety run-in part of the TUC-TOC phase II trial (NCT05955170): Tucatinib in combination with oral etoposide (VP16) - trastuzumab in patients with metastatic HER2-positive breast cancer (HER2+ mBC)

frequentist NMA was conducted, and heterogeneity and global inconsistency were assessed.Results: Eighty mBC trials were screened, networks ranged from 18 to 27 trials (Table ).ChatGPT and Claude captured 1 ineligible trial and Grok missed capturing 8.While Gemini missed 4 trials, it had the most reliable network.ChatGPT, Gemini, and Claude ranked trastuzumab rezetecan and trastuzumab botidotin highest, above trastuzumab deruxtecan.Grok excluded all three regimens during trial selection, resulting in a truncated network.Chemotherapy ranked lowest across models, with stable ordering.Heterogeneity was absent-to-moderate with no global inconsistency.Table: 525P Network characteristics and metrics Trials Regimens I 2 (%) 2 Q p ChatGPT 26 18 0 0 0.63 0.43 Gemini 22 18 0 0 0.55 0.46 Claude 27 20 0 0 0.55 0.46 Grok 18 9 28.9 0.01 0.55 0.46 Conclusions: Concordant NMA results from independently created datasets support the robustness of LLM-assisted trial selection.Regimen rankings were stable across networks, with investigational ADCs above the current standard of care.Clinical interpretation of efficacy rankings should account for toxicity, data maturity and cross-trial comparison limitations.