PT - JOURNAL ARTICLE AU - Suh, CH AU - Jung, JY AU - Lee, HY AU - Kim, HA AU - Kim, SS AU - Hong, SJ TI - 235 Hierarchical cluster analysis of systemic lupus erythematosus AID - 10.1136/lupus-2017-000215.235 DP - 2017 Mar 01 TA - Lupus Science & Medicine PG - A108--A108 VI - 4 IP - Suppl 1 4099 - http://lupus.bmj.com/content/4/Suppl_1/A108.1.short 4100 - http://lupus.bmj.com/content/4/Suppl_1/A108.1.full SO - Lupus Sci & Med2017 Mar 01; 4 AB - Background and aims Systemic lupus erythematosus (SLE) is a heterogeneous disorder with diverse manifestations and serologic features. The purpose is to categorise SLE patients into similar initial characteristics.Methods Hierarchical cluster analysis approached to 389 SLE patients and 10 laboratory values. Laboratory values were transformed into Z-score for hierarchical clustering. Ward’s method as agglomeration method was a criterion applied with spearman correlation as distance metric. Clinical characteristics among clusters were examined by ANOVA with Tukey and Fisher’s exact test. To find each SLE cluster using initial laboratory, linear discriminant analysis was applied.Results Three clusters were revealed by initial laboratory data; Cluster 1 had higher anti-dsDNA antibody, ANA titer and ESR, and low complements, lymphocyte, haemoglobin and platelet counts, Cluster 2 had lower anti-dsDNA antibody, ANA titer and ESR, and Cluster 3 had lower anti-dsDNA antibody titer, WBC and lymphocyte counts, and higher ANA titer. As a result from analysing cumulative manifestations and treatment, Cluster 1 showed more frequent malar rash, alopecia and renal disease with higher SLEDAI, and more use of cyclophosphamide and azathioprine. Also, oral ulcer was developed frequently in Cluster 2. During disease duration, total and mean corticosteroids and the number of flare were higher in Cluster 1.Conclusions With initial laboratory values, SLE patients could be divided 3 clusters. Each Cluster showed different characteristics in clinical manifestations and treatment patterns. This predictive model considered disease severity had 84.6% of total predictability.