Subphenotypes of adult-onset diabetes: Data-driven clustering in the population-based KORA cohort

A data-driven cluster analysis in a cohort of European individuals with type 2 diabetes (T2D) has previously identified four subgroups based on clinical characteristics. In the current study, we performed a comprehensive statistical assessment to (1) replicate the above-mentioned original clusters; (2) derive de novo T2D subphenotypes in the Kooperative Gesundheitsforschung in der Region Augsburg (KORA) cohort and (3) describe underlying genetic risk and diabetes complications.

We used data from n = 301 individuals with T2D from KORA FF4 study (Southern Germany). Original cluster replication was assessed forcing k = 4 clusters using three different hyperparameter combinations. De novo clusters were derived by open k-means analysis. Stability of de novo clusters was assessed by assignment congruence over different variable sets and Jaccard indices. Distribution of polygenic risk scores and diabetes complications in the respective clusters were described as an indication of underlying heterogeneity.

Original clusters did not replicate well, indicated by substantially different assignment frequencies and cluster characteristics between the original and current sample. De novo clustering using k = 3 clusters and including high sensitivity C-reactive protein in the variable set showed high stability (all Jaccard indices >0.75). The three de novo clusters (n = 96, n = 172, n = 33, respectively) adequately captured heterogeneity within the sample and showed different distributions of polygenic risk scores and diabetes complications, that is, cluster 1 was characterized by insulin resistance with high neuropathy prevalence, cluster 2 was defined as age-related diabetes and cluster 3 showed highest risk of genetic and obesity-related diabetes.

T2D subphenotyping based on its sample's own clinical characteristics leads to stable categorization and adequately reflects T2D heterogeneity.