<p>Non-small cell lung cancer (NSCLC) remains a leading cause of cancer mortality, and it remains challenging to predict immunotherapy responses. This study integrates RNA sequencing data from five NSCLC immunotherapy cohorts to identify three molecular subtypes, with a copper-dependent proliferation subtype showing poor prognosis and an immunosuppressive tumor microenvironment. We developed a prognostic model that stratifies patients into high- and low-risk groups by a machine learning pipeline combining 101 algorithmic models. The low-risk group exhibited higher immune infiltration and better progression-free survival, characterized by activation of immune-related pathways, such as IL-2/STAT5 and IFN-γ signaling. CEACAM5⁺ epithelial cells were identified as a high-risk subgroup linked to poorer survival and immunotherapy response via mapping the score of the model and clinical information into single-cell sequencing data. Finally, analysis of clinical specimens with different immunotherapy responses confirmed, by western blot and immunohistochemistry, that expression of CEACAM5⁺ epithelial cells related markers was significantly higher in epithelial cells of the non-MPR group compared with the MPR group. Our findings highlight the importance of genes related to cuproptosis and copper hemostasis as biomarkers for immunotherapy prediction and prognosis stratification.</p>