Glucose patterns and clinical outcomes in the Dallas Heart Study: A pathway to precision diabetes

Continuous glucose monitoring (CGM) systems have revolutionized glycemic control in people with type 1 diabetes. Prediabetes and type 2 diabetes are heterogenous conditions with significant variation in etiology and rates of disease progression.1-3 Standard glycemic measures such as hemoglobin A1c (HbA1c) do not capture certain elements of glycemia (e.g., variability, hypoglycemia) that are predictive of progression and complications.4-6 CGM systems comprehensively capture glucose patterns and are ideal for precisely identifying high-risk subgroups in persons with and without diabetes. Nonetheless, few studies have conducted standardized CGM assessments and evaluated associations of CGM-defined glucose patterns with progression of glycemia and clinical outcomes outside of type 1 diabetes. Emerging studies of CGM-defined subtypes (or “glucotypes”) are small, focus on patients in clinical settings, use older CGM devices, lack diversity and external validation, do not examine the underlying determinants of subtypes, and have not linked subtypes to progression of glycemia and clinical outcomes.7-13 There is also a paucity of data on glucotypes in persons without diabetes. We proposed to use CGM and novel machine learning (ML) methods to identify clusters of persons with and without diabetes at high risk for poor outcomes.