Data-driven cluster analysis of adults with prediabetes: Findings from the Japan diabetes prevention study

Abstract

Aim: To identify distinct metabolic subtypes among Japanese adults with prediabetes and examine their association with type 2 diabetes (T2D) risk and response to lifestyle interventions. Methods: Data from 295 adults with impaired glucose tolerance (mean age 50.4 ± 14.6 years) enrolled in the Japanese Diabetes Prevention Program, a randomized controlled trial with up to 6 years of follow-up, were analyzed. Participants with a history of diabetes, gastrectomy, or exercise contraindications were excluded from the study. K-means clustering (k = 2–5) was applied to baseline indices, including HbA1c, HOMA-β, HOMA-IR, and HDL-cholesterol. Principal component analysis (PCA) was used for interpretation. Incident T2D was defined as a repeat 75-g OGTT. Kaplan–Meier and Cox models were adjusted for baseline HbA1c estimated cumulative incidence and hazard ratios (HRs). Results: Three clusters emerged at k = 3, and PCA (two components explaining 68.0% of the variance) clearly separated the clusters based on insulin resistance and glycemic indices. Cluster 1, Metabolically Resilient Prediabetes (n = 127, MRP), showed the most favorable metabolic profile and lowest T2D risk. Cluster 2, Insulin-Insufficient Prediabetes (n = 109, IIP), exhibited reduced β-cell function and the highest T2D risk, but derived the greatest benefit from lifestyle intervention (adjusted HR 0.40, 95% CI 0.17–0.95, P < 0.05). Cluster 3, Severe Insulin-Resistant Prediabetes (SIRP; n = 59), displayed obesity-related insulin resistance and an intermediate risk. Conclusions: Unsupervised clustering identified three biologically distinct prediabetic subtypes that differed in diabetes risk and intervention responsiveness, supporting a precision prevention framework for Japanese adults with prediabetes.