Abstract 7430: Multicenter study on an artificial intelligence model for predicting NSCLC recurrence

Abstract Background: Accurate and practical prediction of recurrence in early-stage NSCLC is essential for managing patient outcomes and tailoring personalized treatment strategies. This study updated and validated a previously developed transformer-based model for NSCLC recurrence prediction, which integrates clinical, pathological, laboratory, and radiologic data, while proposing strategies to enhance its generalizability. Methods: The model was developed and internally validated using data from Samsung Medical Center (SMC) and Kyung Hee University Hospital (KH) collected from 2008 to 2024. The primary outcome was predicting recurrence within one year after the monitoring point. This study demonstrates the timely provision of risk scores (RADAR score) and determined thresholds and the corresponding area under the curve (AUC). Results: A total of 16, 828 patients was enrolled (12, 208 with stage I; 2, 702 with stage II; and 1, 918 with stage III) Development set was 11, 341 and validation set was 5, 487. The median age was 64 years (range: 13-96). Females accounted for 44.3%, while males made up 55.7%. The median follow-up duration was 32.9 months, and 18.7% of patients experienced recurrence. The percentage of never-smokers was 49.7%, and ex-smokers or current smokers comprised 50.3%. Median RFS in total population was 140.9 months (95% CI, 134.0-147.9). The AUC for predicting relapse within one year of the monitoring point was 0.804. However, depends on the institute, AUC was 0.855 for SMC data (2008-2022), however 0.756 for SMC data (2023-2024) and 0.719 for KH data. We developed updated version of RADAR model by using multi-center data and validation. The updated RADAR model showed 0.858 of AUC with sensitivity of 74.5% and specificity of 82.1% (AUC=0.865 in stage I, AUC=0.756 in stage II, and AUC=0.731 in stage III).Conclusions:This study addresses the challenges of generalizability in NSCLC recurrence prediction models across diverse data sources. The integrated model showed improved overall performance, emphasizing the importance of multi-institutional data integration for enhancing prediction reliability. Future work will focus on feature selection to mitigate data heterogeneity and optimize model performance. Further external validation and refinement are necessary to support the implementation of precision medicine tailored to individual patients. Citation Format: Junho Lee, Daehwan Lee, Kiwon Lee, Ho Yun Lee, Tae Jung Kim, Yeong Jeong Jeon, Junghee Lee, Seong Yong Park, Jong Ho Cho, Hong Kwan Kim, Yong Soo Choi, Sehhoon Park, Jong-Mu Sun, Se-Hoon Lee, Jin Seok Ahn, Myung-Ju Ahn, Hyun Ae Jung. Multicenter study on an artificial intelligence model for predicting NSCLC recurrence [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 7430.