Clinical Characterization of Data-Driven Diabetes Clusters of Pediatric Type 2 Diabetes

Background. Pediatric Type 2 diabetes (T2D) is highly heterogeneous. Previous reports on adult-onset diabetes demonstrated the existence of diabetes clusters. Therefore, we set out to identify unique diabetes subgroups with distinct characteristics among youth with T2D using commonly available demographic, clinical, and biochemical data. Methods. We performed data-driven cluster analysis (K-prototypes clustering) to characterize diabetes subtypes in pediatrics using a dataset with 722 children and adolescents with autoantibody-negative T2D. The six variables included in our analysis were sex, race/ethnicity, age, BMI Z-score and hemoglobin A1c at the time of diagnosis, and non-HDL cholesterol within first year of diagnosis. Results. We identified five distinct clusters of pediatric T2D, with different features, treatment regimens and risk of diabetes complications: Cluster 1 was characterized by higher A1c; Cluster 2, by higher non-HDL; Cluster 3, by lower age at diagnosis and lower A1c; Cluster 4, by lower BMI and higher A1c; and Cluster 5, by lower A1c and higher age. Youth in Cluster 1 had the highest rate of diabetic ketoacidosis (DKA) (p=0.0001) and were most prescribed metformin (p=0.06). Those in Cluster 2 were most prone to polycystic ovarian syndrome (p=0.001). Younger individuals with lowest family history of diabetes were least frequently diagnosed with diabetic ketoacidosis (p=0.001) and microalbuminuria (p=0.06). Low-BMI individuals with higher A1c had the lowest prevalence of acanthosis nigricans (p=0.0003) and hypertension (p=0.03). Conclusions. Utilizing clinical measures gathered at the time of diabetes diagnosis can be used to identify subgroups of pediatric T2D with prognostic value. Consequently, this advancement contributes to the progression and wider implementation of precision medicine in diabetes management.