Abstract 3650: AI-based MTB assistant in cancer personalized treatment: Analysis from a real-world, multi-center retrospective study

Abstract Background: Developing personalized treatment plans in cancer care is a complex and time-intensive process. The multidisciplinary Molecular Tumor Board (MTB) approach offers a promising solution for tailoring precision treatments but is often limited by the manual analysis of vast molecular data, which is laborious and error-prone. To address these challenges, we developed SmartMTB, an artificial intelligence (AI)-based assistant, designed to support clinicians in formulating cost-effective and efficient personalized treatment plans. Methods: SmartMTB leverages a medical-field-customized large-language model and a database tailored to the Chinese population. It integrates patient-specific tumor types and gene variation data to suggest personalized treatment strategies. In a real-world, multi-center retrospective study on the precision medicine platform, we included clinical data of 1032 Chinese cancer patients and asked SmartMTB for treatment plan suggestions. Results: The study cohort included 1,032 patients spanning 13 major tumor types, with a median age of 58 years (range: 7-87 years). Clinical stages ranged from I to IV (stage I: 11.4%, stage II: 10.3%, stage III: 21.9%, stage IV: 51.6%, unknown: 4.8%). Among them, 92% of the patients got treatment plan suggestions from SmartMTB. Eventually, 39.4% of the patients received SmartMTB-recommended treatment plans, while 52.5% did not. Patients who received SmartMTB-recommended treatment plans exhibited higher partial response rate (PR, 31.4% vs. 11.3%), objective response rate (ORR, 34.1% vs. 12.3%) and longer median progression-free survival (mPFS, 163 vs. 92 days, P < 0.0001) than those who did not. Similar benefits were observed across cancer types, including lung cancer (PR: 31.9% vs. 10.2%; ORR: 35.2% vs. 11.9%; mPFS: 185 vs. 92 days, P < 0.0001) and hepatobiliary tumors (PR: 28.6% vs. 13.3%; ORR: 31.4% vs. 15%; mPFS: 121 vs. 87 days, P = 0.024). Excluding lung cancer patients treated with EGFR-TKI, SmartMTB's recommendations maintained efficacy for the remaining lung cancer patients (PR: 31.7% vs. 8.9%; ORR: 35% vs. 10.7%; mPFS: 153 vs. 90 days, P < 0.0001). Furthermore, SmartMTB took about 2.5 h to analyze data from 1032 patients and provide treatment suggestions, which was much more efficient than the traditional manual MTB. Conclusion: In this study, patients using treatment strategy SmartMTB recommended had much better clinical treatment results than those who didn't. This study highlights the clinical utility of SmartMTB in enhancing treatment outcomes for cancer patients. By providing timely, accurate, and personalized treatment recommendations, SmartMTB addresses critical gaps in traditional approaches, reducing time and cost burdens. Its potential for widespread clinical adoption could transform precision oncology practices and improve patient care globally. Citation Format: Minghui Wang, Yanhong Shang, Zhengang Yuan, Xin Huang, Qing Hao, Shaohua Yuan, Guobang Shi, Hui Chen, Wenjin Liu, Lili Wang, Shiwang Wen, Huilai Lv, Zhizheng Wang, Dong You, Yang Liu, Fei Pang, Wei Rao, Haitao Wang, Leilei Lu, Kai Wang. AI-based MTB assistant in cancer personalized treatment: Analysis from a real-world, multi-center retrospective study. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3650.