Abstract
To investigate ethnic differences inprogression from prediabetes to diabetes in a multi-ethnic community, and develop prediction models for diabetes risk and glycemic progression. This 4.7-year retrospective cohort included 4,726 non-diabetic Han, Yi, and Dai individuals. Cox regression was used to analyze the cumulative incidence of diabetes. Two prediction models were developed and validated: a diabetes risk model and GlycoStep, a model for glycemic progression. Prediabetes was significantly associated with diabetes development (HR = 2.25). Over 4.7 years, the cumulative diabetes incidence was 12.7% in the prediabetes group versus 5.5% in normoglycemic individuals. The Yi had the highest incidence, followed by Han and Dai. The diabetes risk model achieved a C-index of 0.76 and AUCs of 0.65-0.80 over 2-4 years, demonstrating effective predictive performance. Similarly, the GlycoStep model exhibited strong predictive ability (C-index: 0.66; AUC range: 0.68-0.77), with ethnic-specific variations. These findings highlight ethnic disparities in diabetes progression and support individualized prevention strategies in multi-ethnic communities.