A data-driven approach to identify clusters of HbA1c longitudinal trajectories and associated outcomes in type 2 diabetes mellitus: a large population-based cohort study.

BackgroundWe aimed to identify and characterize common patterns of HbA1c progression among type 2 diabetes mellitus patients who initiate a non-insulin antidiabetic drug (NIAD). MethodsThe IQVIA Medical Research Data incorporating data from THIN, a Cegedim database of anonymized electronic health records, was used to identify a cohort of patients with a first-ever prescription for a NIAD between 2006 and 2019. Trajectory clusters were identified using an Expectation-Maximization algorithm by iteratively fitting k thin-plate splines and reassigning each patient to the nearest cluster. Cox proportional hazards models calculated the hazard ratios (HR) and 95% confidence intervals (CI) for the estimated risk of microvascular (e.g., retinopathy, diabetic polyneuropathy [DPN]) and macrovascular events. FindingsAmong 116,251 new users of NIADs we found five distinct clusters of HbA1c progression, which were characterized as: optimally controlled (OC), adequately controlled (AC), sub-optimally controlled (SOC), poorly controlled (PC), and uncontrolled (UC). The UC and AC clusters had similar index HbA1C (>9%) but the AC cluster achieved HbA1c control (HbA1C <7.5%), while the UC cluster HbA1c remained >9.0%. Compared to the OC cluster, there was a 21% (HR: 1.21, 95% CI: 1.14-1.28) and 30% (HR: 1.30, 95% CI: 1.21-1.40) elevated risk of retinopathy in the AC and UC clusters, respectively. While the PC and UC clusters had a significant 23% (HR 1.23, 95% CI 1.12 - 1.35) and 45% (HR 1.45, 95% CI: 1.27 - 1.64) increased risk of DPN, respectively. InterpretationThe five identified HbA1c trajectory clusters had different risk profiles. Despite achieving diabetic control, patients categorized in the AC cluster had similar outcomes to the UC cluster, suggesting baseline HbA1c is an important indicator of health outcomes. FundingThe Swiss Data Science Centre