Abstract
Type 2 diabetes (T2D) is a heterogeneous condition, but its phenotypic variation and links with mortality are unclear. We apply the discriminative dimensionality reduction with trees (DDRTree) algorithm to seven clinical variables in 10,091 adults with newly diagnosed T2D from a nationally representative Chinese cohort. Distinct mortality patterns are observed across phenotypes. Cardiovascular mortality is highest in the most hypertensive and obese individuals, while diabetic ketoacidosis/coma mortality is largely driven by the combination of hyperglycemia and dyslipidemia. Additionally, chronic obstructive pulmonary disease mortality is higher in those with elevated high-density lipoprotein (HDL) and total cholesterol levels. These patterns are similar in UK Biobank, though cardiovascular mortality is highest in those with dyslipidemia and obesity. Predictive models incorporating these variables show good performance and an online tool is provided for individual risk prediction. Overall, this study visualizes phenotypic variation in T2D and its impact on mortality, underscoring the need for personalized treatment strategies.</p>