Heterogeneity of diabetes and disease progression with a tree-like representation: findings from the China Cardiometabolic Disease and Cancer Cohort (4C) study.

Abstract

Aims/hypothesis: Diabetes heterogeneity has been modelled as a continuum in European populations, but its phenotypes and long-term comorbidity risks remain unclear in Chinese individuals. This study aimed to identify distinct phenotypes and evaluate their links to future cardiometabolic risks in a large Chinese cohort.

Methods: The discriminative dimensionality reduction with trees (DDRTree) algorithm was used to develop a tree structure based on nine clinical variables. Cox proportional hazard models or logistic regression models were used to analyse probabilities of diabetes-related outcomes.

Results: This study included 19,612 individuals with newly diagnosed diabetes (36.8% male, mean age 59.01 years [SD 8.63]) from the China Cardiometabolic Disease and Cancer Cohort (4C) study. All nine clinical variables used for establishing DDRTree models were gradient distributed across the tree. By overlaying risks of diabetes-related outcomes, we show how these risks differ by participant phenotype. Participants characterised by hyperglycaemia, obesity and dyslipidaemia showed elevated risks of insulin initiation, hypoglycaemia and chronic kidney diseases, while those with hypertension and high creatinine, total cholesterol and alanine aminotransferase levels were associated with a higher risk of CVD. Notably, social determinants and lifestyle factors further contributed to the observed heterogeneity.

Conclusions/interpretation: These findings characterise the heterogeneity of diabetes phenotypes and complication risks in the Chinese population, suggesting potential implications for personalised diabetes care. Given the observed phenotypic differences, management strategies should consider population-specific characteristics.