Trajectory of Glycated Hemoglobin Over Time Among Obese Type 2 Diabetes Patients on U-100 Basal-Bolus Insulin Regimen Using Real-World Data

• Published in: Journal of the Endocrine Society 2021-05, Vol.5 (Supplement_1), p.A480-A481
• Main Authors: Chen, Jieling, Nelson, David R, Borra, Sujana, Liu, Yi, Maughn, Keshia, Fan, Ludi, Rey, Gabriel G, Pollom, Roy D, Hood, Robert C
• Format: Article
• Language: English
• Subjects: Diabetes Mellitus and Glucose Metabolism
• ISSN: 2472-1972; 2472-1972
• DOI: 10.1210/jendso/bvab048.982

Introduction: There is a consistent increase in prevalence of obese type 2 diabetes (T2D) patients, many of whom require insulin treatment. As endogenous insulin secretion dwindles and progressively increasing body weight worsens insulin resistance, exogenous insulin doses can be quite high. This presents unique disease management challenges to patients, treating physicians, and the health systems. These challenges could lead to therapeutic inertia and result in poor glycemic control (continuous increase or persistently high glycated hemoglobin [A1c]). Therefore, examining the A1c trajectory over a significant period could detect the existence and magnitude of therapeutic inertia. Further segmenting the patients based on their A1c trajectory over time would help formulate management strategies with tailored interventions to targeted patient segments with signs of therapeutic inertia. Objective: To segment obese patients with T2D on U-100 basal-bolus regimen based on A1c trajectory over a 3-year period. Methods: Adults with ≥2 T2D claims who were on U-100 basal-bolus regimen and with body mass index ≥30 kg/m2 or diagnosis codes for obesity during the identification period (APR2014-SEP2015) in the Veterans Health Administration database were included. The study period was OCT2013-SEP2018 and patients were required to have continuous enrollment for ≥6 months pre- and ≥3 years post-index periods. We captured the A1c pattern at 6-month intervals over a 3-year period. Only patients with A1c in at least 4 of the 6 time periods were included. A longitudinal unsupervised trajectory clustering method using the traj R package was implemented. Twenty-four features of A1c trajectory were examined followed by feature reduction using factor analysis. Based on the selected features, K-means clustering was used to identify patient segments based on the A1c trajectory. We extended the approach by repeating the same process as above among patient cluster with stable A1c trajectory to detect further A1c patterns. Results: A total of 45,520 patients were included. Four patient clusters were identified based on distinct patterns of A1c trajectory. The first cluster has descending A1c over time (N=8,325; 18.3%) while the second one has ascending A1c over time (N=8,123; 17.8%). Among patients with stable A1c trajectory, two more clusters were identified: stable high (N=10,654; 23.4%) with persistently high A1c around 9.0% and stable low (N=18,378; 40.4%) with persistently low A1c