Type 2 diabetes classification: a data-driven cluster study of the Danish Centre for Strategic Research in Type 2 Diabetes (DD2) cohort

Abstract

Introduction: A Swedish data-driven cluster study identified four distinct type 2 diabetes (T2D) clusters, based on age at diagnosis, body mass index (BMI), hemoglobin A1c (HbA1c) level, and homeostatic model assessment 2 (HOMA2) estimates of insulin resistance and beta-cell function. A Danish study proposed three T2D phenotypes (insulinopenic, hyperinsulinemic, and classical) based on HOMA2 measures only. We examined these two new T2D classifications using the Danish Centre for Strategic Research in Type 2 Diabetes cohort.

Research design and methods: In 3529 individuals, we first performed a k-means cluster analysis with a forced k-value of four to replicate the Swedish clusters: severe insulin deficient (SIDD), severe insulin resistant (SIRD), mild age-related (MARD), and mild obesity-related (MOD) diabetes. Next, we did an analysis open to alternative k-values (ie, data determined the optimal number of clusters). Finally, we compared the data-driven clusters with the three Danish phenotypes.

Results: Compared with the Swedish findings, the replicated Danish SIDD cluster included patients with lower mean HbA1c (86 mmol/mol vs 101 mmol/mol), and the Danish MOD cluster patients were less obese (mean BMI 32 kg/m² vs 36 kg/m²). Our data-driven alternative k-value analysis suggested the optimal number of T2D clusters in our data to be three, rather than four. When comparing the four replicated Swedish clusters with the three proposed Danish phenotypes, 81%, 79%, and 69% of the SIDD, MOD, and MARD patients, respectively, fitted the classical T2D phenotype, whereas 70% of SIRD patients fitted the hyperinsulinemic phenotype. Among the three alternative data-driven clusters, 60% of patients in the most insulin-resistant cluster constituted 76% of patients with a hyperinsulinemic phenotype.

Conclusion: Different HOMA2-based approaches did not classify patients with T2D in a consistent manner. The T2D classes characterized by high insulin resistance/hyperinsulinemia appeared most distinct.

Keywords: classification; clusters; cohort; type 2 diabetes.

Figure 1

Patient and cluster variable distribution. (A) De novo replication of the four Swedish T2D clusters in the DD2 cohort and (B) New alternative k-value DD2 Clusters. BMI, body mass index; DD2, The Danish Centre for Strategic Research in Type 2 Diabetes Cohort; HbA1c, hemoglobin A1c; HOMA2-B, homeostatic model assessment 2 estimate of beta-cell function; HOMA2-IR, homeostatic model assessment 2 estimate of insulin resistance; MARD, mild age-related diabetes; MOD, mild obesity-related diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes; T2D, type 2 diabetes.

Figure 2

Sankey diagrams showing the flow of patients between (A) the clusters identified in the de novo analysis replicating the Swedish clusters and the New DD2 Clusters; (B) the clusters identified while replicating the Swedish clusters and Danish phenotypes; (C) the New DD2 Clusters and the Danish phenotypes; and (D) the clusters identified in the de novo analysis replicating main analysis and those identified using information on centroids and means identified in the original Swedish ANDIS cohort. A total of 15 (SIDD: n=2, SIRD: n=1, MOD: n=5, MARD: n=7) individuals had high insulin sensitivity and high beta-cell function (ie, neither insulinopenic, classical, or hyperinsulinemic type 2 diabetes). In the previous DD2 phenotype study, this was considered to be HOMA2 values in the non-diabetes area, and the individuals were therefore not classified. We did not exclude these 15 individuals in our study; however, they are not shown in figure parts B and C. ANDIS, All New Diabetics in Scania; DD2, Danish Centre for Strategic Research in Type 2 Diabetes; HOMA2, homeostatic model assessment 2; MARD, mild age-related diabetes; MOD, mild obesity-related diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.

Figure 3

Plot of insulin sensitivity and beta-cell function. The lines mark the distinction between the three Danish phenotypes, and the colors mark the four clusters identified in the main analysis. ^aA total of 15 (SIDD: n=2, SIRD: n=1, MOD: n=5, MARD: n=7) individuals had high insulin sensitivity and high beta-cell function (ie, neither insulinopenic, classical, or hyperinsulinemic type 2 diabetes). In the previous DD2 phenotype study, this was considered to be HOMA2 values in the non-diabetes area, and the individuals were therefore not classified. We did not exclude these 15 individuals in our study. DD2, Danish Centre for Strategic Research in Type 2 Diabetes; HOMA2, homeostatic model assessment 2; MARD, mild age-related diabetes; MOD, mild obesity-related diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.