2566 - Genomics-Based Brain Metastasis Risk Model in Stage II-III NSCLC after Concurrent Chemoradiotherapy and Consolidative Immunotherapy
Presenter(s)
K. Olabode1, K. Shin2, L. Martinka1, Simran1,3, J. Gray4, S. Puri4, A. N. Saltos4, T. Tanvetyanon4, B. Creelan4, A. Chiappori4, C. Lu4, M. Shafique4, K. A. Ahmed5, K. Yamoah1, P. C. Rodriguez6, S. A. Rosenberg1, T. J. Dilling1, S. K. Jabbour7, and J. Kim6,8; 1Department of Radiation Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL, 2Division of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 3Department of Radiation Oncology, All India Institute of Medical Sciences (AIIMS), New Delhi, India, 4Department of Thoracic Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL, 5H. Lee Moffitt Cancer Center and Research Institute, Department of Radiation Oncology, Tampa, FL, 6Department of Immunology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL, 7Department of Radiation Oncology, Rutgers Cancer Institute, New Brunswick, NJ, 8Department of Radiation Oncology, Moffitt Cancer Center, Tampa, FL
Purpose/Objective(s):
Brain metastasis (BM) remains a major cause of treatment failure and mortality in locally advanced non–small cell lung cancer (LA-NSCLC) despite the adoption of concurrent chemoradiation (CHRT) followed by an immune checkpoint inhibitor (ICI) per the PACIFIC regimen. The genomic drivers of BM in this setting remain poorly defined. Therefore, we aim to develop a genomics-based risk model to predict BM following CHRT and ICI in LA-NSCLC.Materials/Methods:
We retrospectively analyzed 114 patients with stage II–III NSCLC treated with the PACIFIC regimen from August 2017 to February 2025 who underwent next-generation sequencing using a 252-gene panel. Genes with =5% mutation frequency were analyzed. The overall survival (OS) was stratified by BM status using Kaplan–Meier analysis and log-rank testing. To identify genomic risk features, robust feature selection was performed using a dual-model approach incorporating logistic least absolute shrinkage and selection operator (LASSO) regression with stability selection and Firth’s penalized likelihood regression to minimize overfitting and small-sample bias. A LASSO-derived genomic risk-scoring model was developed and used to stratify patients into high- and low-risk groups.Results:
With a median follow-up of 45.3 months (range, 4.5–85.4), BM developed in 33 patients (28.9%). Median age was 68 years (39–87). 54.4% were female, while 46.5% were former smokers and 29.8% were current smokers. 89.5% of patients had stage III disease, and 54.4% had adenocarcinoma histology. The median radiation dose was 60Gy (60-70). The most common chemotherapy regimen was Carboplatin/Paclitaxel (74.6%). Median OS was significantly worse in patients who developed BM compared with those who did not [31.8 (23.9–NA) vs. 73.6 (34.7–NA) months; p = 0.01]. Multivariable genomic feature selection using LASSO and Firth models (p < 0.05) consistently identified TP53 and BRCA2 mutations as the most robust predictors of BM. TP53 mutations were associated with increased BM incidence, whereas BRCA2 mutations demonstrated a protective association. On brain metastasis–free survival (BMFS) analysis, only 1 of 11 patients with BRCA2 mutations developed brain metastasis, whereas 29 of 78 patients with TP53 mutations experienced brain metastases. A genomics-derived BM risk-scoring model [1.21 × TP53) - (1.55 × BRCA2) + (0.87 × FGF10)] was generated and stratified patients into high (n = 68) and low (n = 46)-risk groups, based on a median score of 1.2083, with significantly different median BMFS (36.8 months vs. not reached; p < 0.002; C-index 0.65).Conclusion:
TP53 and BRCA2 were the most robust genomic predictors of BM risk, with TP53 increasing risk and BRCA2 showing a protective effect. A genomics-based risk score stratified patients into distinct BM risk groups, supporting the potential role of molecular profiling for risk-adapted surveillance and future preventive strategies. Prospective validation is warranted.