B-079 Identifying Heterogeneity of Systemic Connective Tissue Diseases by Applying Joint Dimension Reduction and Cluster Analysis to Immunomarkers

Abstract

Abstract Background Accurate identification of systemic connective tissue diseases (SCTDs) can be challenging due to the heterogeneity of each disease. This study aims to explore the heterogeneity of different SCTDs by applying non-supervised learning to immunomarker data and proposes an alternative classification. Additionally, the study investigates the clinical implications of these different groups. Methods We utilized multiple correspondence analysis and k-means clustering to analyze the immunomarker data of patients with systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), and Sjögren’s syndrome (SS) in Taiwan between 2001 and 2016. To investigate the clinical correlations of the classification results, we conducted a comprehensive review of the clinical findings and compared the results across different groups. Results A total of 11 923 patients with the three SCTDs were classified into six distinct clusters based on their immunomarker patterns. In this clustering system, RA patients were predominantly grouped in cluster 1, whereas SLE and SS patients had a more dispersed distribution. Among patients with SLE, renal involvement was more frequent in clusters 3 and 6 (52% and 51%, respectively) than in other clusters, and discoid lupus was more frequent in cluster 3. Among SS patients, typical SS findings such as inflammation or infection of eyes and allergic reactions were more common in cluster 2 (54% and 51%, respectively). In contrast, patients in cluster 3 had more unspecified benign neoplasms (58%) and were more likely to experience immune disorders (63%). Conclusion The immunomarker-driven clustering yields an alternative classification result with new clinical insights. The data-driven approach would provide a more evidence-based classification and aid in a more accurate diagnosis and management for SCTDs.