Rheumatoid arthritis (RA) is a chronic, heterogeneous autoimmune disease. Its high disability rate has a serious impact on society and individuals, but there is still a lack of effective and reliable diagnostic markers and therapeutic targets for RA. In this study, we integrated RA patient information from three GEO databases for differential gene expression analysis. Additionally, we also obtained pan-cancer-related genes from the TCGA and GTEx databases. For RA-related differential genes, we performed functional enrichment analysis and constructed a weighted gene co-expression network (WGCNA). Then, we obtained 490 key genes by intersecting the significant module genes selected by WGCNA and the differential genes. After using the RanddomForest, SVM-REF, and LASSO three algorithms to analyze these key genes and take the intersection, based on the four core genes (BTN3A2, CYFIP2, ST8SIA1, and TYMS) that we found, we constructed an RA diagnosis. The nomogram model showed good reliability and validity after evaluation, and the ROC curves of the four genes showed that these four genes played an important role in the pathogenesis of RA. After further gene correlation analysis, immune infiltration analysis, and mouse gene expression validation, we finally selected CYFIP2 as the cut-in gene for pan-cancer analysis. The results of the pan-cancer analysis showed that CYFIP2 was closely related to the prognosis of patients with various tumors, the degree of immune cell infiltration, as well as TMB, MSI, and other indicators, suggesting that this gene may be a potential intervention target for human diseases including RA and tumors.