Shared transcriptional signatures between rheumatoid arthritis and lung cancer: a multi-cohort machine learning study identifies CKS2 as a candidate cell cycle marker associated with poor prognosis in lung cancer

Patients with rheumatoid arthritis (RA) have an increased risk of developing lung cancer (LC), yet the molecular features that may underlie this epidemiological association remain poorly characterized. Given the shared inflammatory and immune dysregulation features observed in both conditions, integrative analysis of transcriptomic data from independent RA and LC cohorts may help identify overlapping molecular signatures that could inform future studies on comorbidity risk. In study 1, microarray data from peripheral blood samples of an RA training cohort (RAtrc, n = 275) and lung tissue samples of an LC training cohort (LCtrc, n = 194) were analyzed to identify differentially expressed genes (DEGs) in both diseases relative to their respective controls. Machine learning algorithms, including support vector machine recursive feature elimination (SVM-RFE), random forest (RF)-RFE, and extreme gradient boosting (XGB)-RFE were applied to identify risk signatures and establish model RAtrc and LCtrc. A repeated double cross-validation strategy was implemented, model RAtrc was internally validated in independent RA cohorts (RAtec1: n = 398; RAtec2: n = 135), and externally evaluated using LC cohorts (LCtec1: n = 166; LCtec2: n = 120) to assess its generalizability across diseases. Conversely, model LCtrc was validated in LC cohorts and evaluated in RA cohorts. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), and calibration was assessed using a calibration curve with the mean absolute error (MAE) reported. In response to class imbalance, we used Precision-Recall AUC (PR-AUC) and Weighted F1 Score as additional metrics. Prognostic associations of risk scores were evaluated in LC cohorts using Lasso, multivariate Cox, and stepwise regression models. In study 2, microarray data from 2076 LC patients and 1349 controls were pooled to evaluate the prognostic relevance of the identified signature genes in LC. Transcriptional analyses revealed shared dysregulation in pathways related to N6-adenosine methylation, immune checkpoints, ferroptosis, immune infiltration, DNA repair and G2M checkpoint between RA and LC. Model RAtrc (AUC = 0.956, MAE = 0.019) achieved high performance in internal validation in RAtec1 (Weighted F1 = 0.913, PR-AUC = 0.996), RAtec2 (Weighted F1 = 0.762, PR-AUC = 0.985), but showed limited generalizability when applied to LCtec1 (AUC = 0.617, MAE = 0.013), and LCtec2 (AUC = 0.661, MAE = 0.021). model LCtrc (AUC = 0.995, MAE = 0.023) demonstrated strong performance in LCtec1(AUC = 0.702, MAE = 0.016), LCtec2 (AUC = 0.930, MAE = 0.013), and unexpectedly high discrimination in RAtec1 (Weighted F1 = 0.730, PR-AUC = 0.986), and RAtec2 (Weighted F1 = 0.740, PR-AUC = 0.966). Risk scores derived from the signatures correlated with oncogenic pathway activity and poorer survival outcomes in LC patients. CKS2, a gene common to both signatures, emerged as a candidate cell cycle regulator associated with poor prognosis in LC, warranting further investigation in the context of RA comorbidity. This study identifies overlapping gene expression signatures in RA and LC, particularly in immune and cell cycle pathways, providing a resource for further exploration of the links between chronic inflammation and tumorigenesis. These findings may eventually inform risk stratification strategies in RA patients, though prospective validation in longitudinal cohorts is required before clinical translation. The signatures show prognostic value in LC, suggesting potential utility in personalized oncology. Further studies are needed to validate CKS2 as a prognostic biomarker and to examine its role in patients with both RA and LC.