Identifying Severity Clusters in SLE Patients

Abstract

Machine learning (ML) has a successful impact in healthcare data mining. We use unsupervised ML methods to extract features and identify subgroups of Systemic Lupus Erythematosus (SLE) patients related to the disease severity. We analyze the similarity between SLE patients within these clusters. Finally, we evaluate the clustering results, using two types of cluster validation, internal cluster validation, and external cluster validation. The clustering analysis results show two separate patients clusters which are mild and severe subgroups. Patients in the severe subgroup have a higher prevalence of the renal disorder, hemolytic anemia, anti-dsDNA anti- body, and low complements (C3, C4). The severe subgroup of patients suffer from malar rash and proteinuria with higher use of cyclophosphamide, mycophenolate mofetil, and azathioprine. The second cluster is mild disease activity, and it is associated with joint pain, low complements (C3, C4), and a positive anti-dsDNA antibody.