Background <p>The association between rheumatoid arthritis (RA) and lung cancer risk has attracted considerable clinical attention, but conclusions remain inconsistent due to confounding factors such as smoking, and previous studies have paid less attention to specific histological subtypes. This study aims to systematically elucidate the complex relationship between the two through an evidence triangulation strategy combining large-scale clinical cohort studies, machine learning risk prediction, and genetic causal inference.</p> Methods <p>This study adopted an evidence triangulation strategy, integrating evidence from different populations and methods. First layer of evidence (clinical association, Chinese population): We conducted 1:1 propensity score matching analysis on a real-world cohort of 8,867 subjects (4,661 RA patients) (2014–2024, First Teaching Hospital of Tianjin University of Traditional Chinese Medicine), with particular focus on the association between RA and different lung cancer subtypes (adenocarcinoma, squamous cell carcinoma, small cell carcinoma). We also built machine learning models (logistic regression, random forest, XGBoost) based on 22 clinical features from this cohort to explore the feasibility of individualized risk prediction for overall lung cancer. Second layer of evidence (genetic causation, European population): We turned to use completely independent large-scale public GWAS summary data (RA: <i>N</i> = 92,044, European ancestry; squamous cell lung cancer: <i>N</i> = 63,053, European ancestry; smoking: <i>N</i> = 111,752, European ancestry), employing multivariable Mendelian randomization (MVMR) analysis, after adjusting for the genetic effects of smoking, to specifically explore the causal effect of RA on squamous cell lung cancer.</p> Results <p>Clinical evidence reveals that the association between RA and lung cancer exhibits significant subtype specificity: after matching, RA significantly increases squamous cell lung cancer risk (adjusted OR = 2.415, 95% CI: 1.104–5.283, <i>P</i> = 0.027), but shows no significant association with adenocarcinoma (<i>P</i> = 0.437) or small cell carcinoma (<i>P</i> = 0.564). Machine learning models based on clinical features showed limited predictive ability (AUC = 0.57–0.68), revealing the challenge of translating population associations into individual predictions. In an independent European population, genetic analysis presents an instructive paradox: preliminary univariable MR analysis shows RA has a genetic protective effect on squamous cell lung cancer (OR = 0.985), while linkage disequilibrium score regression (LDSC) shows no significant genetic correlation between the two (rg = -0.019, <i>P</i> = 0.806). After simultaneously adjusting for the genetic effects of smoking in multivariable MR analysis, the causal promoting effect of RA on squamous cell lung cancer was revealed (OR = 1.02, 95% CI: 1.00–1.04, <i>P</i> = 0.046), explaining that the apparent protective effect observed in univariable analysis was actually caused by smoking confounding.</p> Conclusion <p>RA specifically increases squamous cell lung cancer (but not other subtypes) risk. Multivariable Mendelian randomization resolves the causal paradox caused by smoking confounding. Clinical evidence (China) and genetic evidence (Europe) come from different populations and require further validation. Results support implementation of targeted squamous cell lung cancer screening for high-risk RA patients.</p>