Journal Article

Data-driven phenotypic profiling of prediabetes reveals heterogeneous cardiometabolic risks in Chinese adults

Zheng, Ruizhi, Tang, Xulei, Liu, Chao, Yang, Tao, Xu, Min, Wan, Qin, Chen, Li, Lu, Jieli, Huo, Yanan, Wang, Tiange, Chen, Lulu, Qiao, Hong, Qin, Guijun, Zhang, Yinfei, Deng, Huacong, Shen, Feixia, Gu, Xuejiang, Wang, Shuangyuan, Bi, Yufang, Ding, Yilan, Jia, Xiaojing, Ye, Zhen, Luo, Zuojie, Wang, Youmin, Zhao, Jiajun, Lai, Shenghan, Wu, Shengli, Lin, Yiting, Hu, Ruying, Wang, Guixia, Lin, Hong, Chen, Gang, Hou, Xinguo, Huang, Feiyue, Li, Donghui, Shi, Lixin, Su, Qing, Ning, Guang, Xu, Yu, Yan, Li, Zeng, Tianshu, Wei, Huapeng, Hu, Chunyan, Wang, Weiqing, Li, Qiang, Qin, Yingfen, Gao, Zhengnan, Mu, Yiming, Yu, Xuefeng, Wang, Jinfeng, Li, Mian

2025

Overview

Background The heterogeneous and complex nature of prediabetes presents a major challenge in identifying individuals predisposed to developing incident diabetes and related complications. We aimed to identify phenotypic subgroups of prediabetes at risk and to explore their distinct associations with cardiometabolic outcomes. Methods This study included 79,000 individuals with prediabetes from the three large-scale prospective cohorts in China. Phenotypic heterogeneity was identified using a soft-clustering algorithm based on the proximity network derived from uniform manifold approximation and projection (UMAP), combined with graph-clustering and Gaussian mixture models. Associations between phenotype probabilities and the incidence of type 2 diabetes (T2D), cardiovascular disease (CVD), and kidney events were assessed to evaluate risk differences across the identified profiles. Results Six phenotypic profiles were identified, including five with distinct metabolic features (representing ~ 70% of the total population), and one without significant features. These profiles demonstrated substantial differences in both baseline cardiometabolic burden and future disease risk. For instance, individuals with a 20% higher probability of belonging to the hypertensive profile had a 9, 6, and 12% higher risk of T2D, CVD, and CKD, respectively, while the profile with high lipids, creatinine, and liver enzyme was associated with an 10% increased risk of T2D and kidney events. Moreover, incorporating phenotypic probabilities into multivariable models significantly improved the prediction of disease risks (likelihood ratio test, P  < 0.05). Conclusions Prediabetes exhibits substantial phenotypic heterogeneity, and delineation of distinct metabolic profiles enables refined risk stratification and informs precision prevention strategies.

• Publisher: BioMed Central,BMC
• Subject: Adult / Aged / Angiology / Biomarkers - blood / Cardiology / Cardiometabolic Risk Factors / Cardiovascular disease / Cardiovascular Diseases - diagnosis / Cardiovascular Diseases - epidemiology / China - epidemiology / Clustering / Diabetes / Diabetes Mellitus, Type 2 - diagnosis / Diabetes Mellitus, Type 2 - epidemiology / East Asian People / Female / Heterogeneity / Humans / Incidence / Kidney events / Male / Medicine / Medicine & Public Health / Middle Aged / Phenotype / Prediabetes / Prediabetic State - blood / Prediabetic State - diagnosis / Prediabetic State - epidemiology / Prognosis / Prospective Studies / Risk Assessment