Key points are not available for this paper at this time.
Introduction and Objective: Many studies have subclassified T2D with various physiological factors to better understand disease heterogeneity but few have subclassified prediabetes, a period critical to prevention of T2D. We derived prediabetes subgroups and assessed risk for T2D. Methods: We included 3043 participants from the Diabetes Prevention Program. We used baseline data for 35 clinical variables (table footnote) to compare two clustering methods: 1) kmeans using model statistics to identify the optimal cluster/subgroup number and 2) Uniform Manifold Approximation Projection data reduction with Hierarchical Density Based Spatial Clustering of Applications with Noise (HDBSCAN). We used Cox regression to assess 3-year T2D risk and intervention effects, restricting to 2549 participants randomized to the placebo, metformin, or lifestyle arms (with selection weighting). Results: Both kmeans and HDBSCAN methods supported two subgroups as optimal. Despite general subgroup similarities between methods (high obesity in subgroup 1), the kmeans subgroups differed more notably in glucose/insulin measures and HDBSCAN subgroups differed in adiposity measures (table). Within clustering method, subgroups differed in both risk for T2D (32% and 88% higher risk, respectively) and intervention effects. Conclusions: Prediabetes physiological subgrouping may inform T2D risk and targeted prevention strategies. Disclosure R. Casanova: None. J.M. Stafford: None. Y. Demesie: None. B.C. Jaeger: None. B.J. Wells: None. M. Bancks: None. Funding American Diabetes Association (11-22-ICTSPM-18)
Casanova et al. (Fri,) studied this question.