2249-PUB: Clustering Analysis as a Tool to Define Endotypes in Type 1 Diabetes (T1D)

Abstract

Due to heterogeneity in pediatric T1D, uniform treatment strategies are unlikely to benefit all patients equally. T1D endotypes, or subgroups with distinct characteristics amenable to particular interventions, are a potential solution, but defining endotypes remains a challenge. We used electronic medical records (EMR) to define clusters of distinct presentations of pediatric T1D in a diverse population with a high prevalence of obesity. We analyzed data collected through an EMR Population Health Diabetes Registry of 604 children newly diagnosed with T1D at a large academic hospital in southwestern USA. Using PROC Cluster (SAS 9.4), we applied the Centroid Clustering method to find the optimal number of clusters that minimized the sum of squared distance, then used SAS PROC Tree to identify subjects in each cluster. Variables used were age, DKA at diagnosis, weight, sex, race/ethnicity, c-peptide, hemoglobin A1c (A1c), thyroglobulin antibody (TgAb), and thyroid peroxidase antibody (TPO). Due to multiple testing, p-values <0.01 were statistically significant. This yielded 6 clusters: Cluster 1 (n=424), was characterized by low C-peptide, normal weight, negative TgAb and TPO, intermediate or high A1C; Cluster 2 (n=73) by obesity, low or intermediate C-peptide, and negative TgAb and TPO; Cluster 3 (n=42) by positive TgAb; Cluster 4 (n=33), by positive TPO and negative TgAb; Cluster 5 (n=19) by high or intermediate C-peptide without obesity or intermediate C-Peptide with obesity and negative TgAb and TPO; Cluster 6 (n=13) by obesity, high C-peptide and negative TgAb and TPO. After adjusting for age, BMI Z-score and race/ethnicity, clusters were statistically different for A1c, C-peptide and TPO and TgAb status (all p<0.01), but not for glucose or DKA. This strategy identifies distinct groups in T1D based on objective criteria. Endotypes from clustering analysis of more EMR data could be used for targeting therapeutic strategies to groups with similar characteristics and help personalize T1D treatment and care. Disclosure A.F. Siller: None. X. Gu: None. M. Tosur: None. M. Astudillo: None. A. Balasubramanyam: None. M. Bondy: None. M.J. Redondo: None. Funding Baylor College of Medicine