Comparison of clustering and phenotyping approaches for subclassification of type 2 diabetes and its association with remission in Indian population

Abstract

Abstract Identification of novel subgroups of type 2 diabetes (T2D) has helped improve its management. Most classification techniques focus on clustering or subphenotyping but not on both. This study aimed to compare both these methods and examine the rate of T2D remission in these subgroups in the Indian population. K-means clustering (using age at onset, HbA1C, BMI, HOMA2 IR and HOMA2%B) and subphenotyping (using homeostatic model assessment (HOMA) estimates) analysis was done on the baseline data of 281 patients with recently diagnosed T2D who participated in a 1-year online diabetes management program. Cluster analysis revealed three distinct clusters: severe insulin-deficient diabetes (SIDD), severe insulin-resistant diabetes (SIRD), and mild obesity-related diabetes (MOD) while subphenotyping showed four distinct categories: hyperinsulinemic, insulinopenic, classical, and nascent T2D. Comparison of the two approaches revealed that the clusters aligned with phenotypes based on shared characteristics of insulin sensitivity (IS) and beta cell function (BCF). Clustering correctly identified individuals in nascent group (high IS and BCF) as having mild obesity related diabetes which subphenotyping did not. Post-one-year intervention, higher remission rates were observed in the MOD cluster (p = 0.383) and the nascent phenotype showing high IS and BCF (p = 0.061, Chi-Square test). In conclusion, clustering based on a comprehensive set of parameters appears to be a superior method for classifying T2D compared with pathophysiological subphenotyping. Personalized interventions may be highly effective for newly diagnosed individuals with high IS and BCF and may result in higher remission rates in these individuals. Further large-scale studies are required to validate these findings.