Immunotherapy, such as immune checkpoint inhibitors (ICIs), is a validated strategy for treating lung adenocarcinoma (LUAD) patients. One of the main challenges in ICIs treatment is the lack of efficient biomarkers for predicting response or resistance. Metabolic reprogramming has been proven to remodel the tumor microenvironment, altering the response to ICIs. We constructed a prognostic model as metabolism-related gene (MRG) of four genes by using weighted gene co-expression network analysis (WGCNA), the nonnegative matrix factorization (NMF), and Cox regression analysis of a LUAD dataset (n = 500) from The Cancer Genome Atlas (TCGA), which was validated with three Gene Expression Omnibus (GEO) datasets (n = 442, n = 226 and n = 127). The MRG was constructed based on BIRC5, PLK1, CDKN3, and CYP4B1 genes. MRG-high patients had a worse survival probability than MRG-low patients. Furthermore, the MRG-high subgroup was more associated with cell cycle-related pathways; high infiltration of activated memory CD4+T cells, M0 macrophages, and neutrophils; and showed better response to ICIs. Contrarily, the MRG-low subgroup was associated with fatty acid metabolism, high infiltration of dendric cells, and resting mast cells, and showed poor response to ICIs. MRG is a promising prognostic index for predicting survival and response to ICIs and other therapeutic agents in LUAD, which might provide insights on strategies with ICIs alone or combined with other agents.