RESUMEN
AIM: To cluster adults with diabetes using variables from real-world clinical care at manifestation. MATERIALS AND METHODS: We applied hierarchical clustering using Ward's method to 56 869 adults documented in the prospective Diabetes Follow-up Registry (DPV). Clustering variables included age, sex, body mass index (BMI), HbA1c, diabetic ketoacidosis (DKA), components of the metabolic syndrome (hypertension/dyslipidaemia/hyperuricaemia) and beta-cell antibody status. Time until use of oral antidiabetic drugs (OADs), use of insulin, chronic kidney disease (CKD), cardiovascular disease (CVD), retinopathy or neuropathy were assessed using Kaplan-Meier analysis and Cox regression models. RESULTS: We identified eight clusters: four clusters comprised early diabetes onset (median age 40-50 years) but differed with regard to BMI, HbA1c, DKA and antibody positivity. Two clusters included adults with diabetes onset aged in their early 60s who met target HbA1c, but differed in BMI and sex distribution. Two clusters were characterized by late diabetes onset (median age 69 and 77 years) and comparatively low BMI, but differences in HbA1c. Earlier insulin use was observed in adults with high HbA1c, and earlier OAD use was observed in those with high BMI. Time until CKD or CVD was shorter in those with late onset, whereas retinopathy occurred earlier in adults with late onset and high HbA1c, and in adults with early onset, but high HbA1c and high percentage of antibody positivity. CONCLUSIONS: Adult diabetes is heterogeneous beyond classical type 1/type 2 diabetes, based on easily available variables in clinical practice using an automated clustering algorithm that allows both continuous and binary variables.