Аңдатпа
Abstract Background Diabetes is a global pandemic, affecting 537 million adults worldwide in 2021. Diabetes is much more complex than the classification into type 1 diabetes and type 2 diabetes suggests. We aimed to identify homogeneous groups of patients with diabetes based on routinely collected measurements using unsupervised, data-driven cluster analysis. Methods We analyzed the data from Electronic Medical Records of 558 patients with newly diagnosed diabetes from 4 outpatient clinics in Astana. We performed a k-means cluster analysis of 9 routinely measured variables using the KMeans function (iterations=3 million) from the Scikit-Learn Python library. Results Cluster 1, consisting of 176 (31,5%) patients, was defined by late-onset diabetes, relatively low blood pressure (BP) and body mass index (BMI), moderate metabolic derangements, and comparably low glomerular filtration rate (GFR). The 83 (14,9%) patients in cluster 2 were characterized by a late manifestation of diabetes, high BP and BMI, relatively poor glycemic control, moderate dyslipidemia, and comparably low GFR. Cluster 3, including 98 (17,6%) patients, was represented by a low age at onset, comparably low BP and BMI, moderately poor glycemic and lipid controls, and high GFR. The 110 (19,7%) patients in cluster 4 had late-onset diabetes, high BP and BMI, subcompensated diabetes, poor lipid control, and relatively low GFR. Cluster 5, which involved 91 (16,3%) patients, was characterized by comparably low BP and BMI, poor glycemic control, modest dyslipidemia, and high GFR. Conclusions We intend to continue the project for further conducting a prospective study with dynamic monitoring of study participants, tracking specific patterns in the development of diabetic complications in each cluster. The results of this work will contribute to the development of a model for identifying patients with an initially predictable course of diabetes and a high risk of its complications at the time of diagnosis. Key messages • We identified five clusters of patients with diabetes with clinically significant characteristics in each cluster. • Identification of subgroups of patients with an initially predictable course can be useful for recognizing those at highest risk of complications and applying personalized treatment strategies.