Abstract
Lung adenocarcinoma (LUAD) exhibits marked prognostic heterogeneity. Existing methylation-based prognostic models lack robustness, accuracy, or external validation, limiting their clinical utility. We develop MethPro-LUAD, a prognostic model based on eight 5-methylcytosine features, using TCGA-LUAD data (training, N = 269; testing, N = 180), and validate the model across two independent hospital cohorts (Beijing, N = 195; Nanjing, N = 88) and two external GEO datasets (N = 82 and N = 155) to ensure its generalizability and effectiveness. MethPro-LUAD stratifies patients into high- and low-risk groups, with high-risk individuals consistently showing significantly shorter overall survival across all cohorts. In addition, the model's predictive performance remains robust in subgroups defined by age, sex, stage, and EGFR mutation and in disease-free survival analyses. Compared with other reported LUAD prognostic models, MethPro-LUAD demonstrates better performance across cohorts, which offers valuable support for personalized postoperative treatment and follow-up, and warrants further prospective evaluation, highlighting its strong potential for clinical translation through feasible targeted assays.