Prediabetes Subgroups, Type 2 Diabetes Risk, and Differential Effects of Preventive Interventions

Prior studies have subclassified type 2 diabetes using statistical clustering approaches with clinical data, but few have subclassified prediabetes and assessed the effects of preventive interventions. Our objective was to derive prediabetes subgroups based on clinical biomarkers and assess risk for incident diabetes and differential preventive intervention effects within the derived subgroups, with comparison to more simple modeling approaches.

Baseline data for 3145 participants in the Diabetes Prevention Program trial were used to derive prediabetes subgroups using K-means clustering with data for 22 clinical biomarkers (sex-standardized). Cox proportional hazards regression was used to estimate hazard ratios (HRs) for diabetes and differential intervention effects (intensive lifestyle, metformin, or placebo) by prediabetes subgroups and to compare the clustering strategy to a model with clinical variables.

We identified 2 prediabetes subgroups characterized by severe insulin resistance with severe obesity (subgroup 1, 31% of sample) and moderate insulin resistance with overweight or obesity (subgroup 2, 69%). Subgroup 1 had a 58% higher risk for diabetes (HR: 1.58, 95% confidence interval: 1.31, 1.91) compared to subgroup 2. Randomization to lifestyle (compared to placebo) halved diabetes risk for both subgroups, while metformin provided greater benefit to subgroup 1 vs subgroup 2 (P for interaction <.05). A clinical variable model discriminated diabetes risk better than the clustering strategy.

Pathophysiologically distinct prediabetes subgroups differ in risk for diabetes and preventive benefit from metformin. These results support distinct mechanisms of diabetes susceptibility; however, the use of clinical prediction models to guide treatment decisions may provide adequate risk profiling.