Researchers have leveraged large-scale Chinese population cohorts to systematically investigate metabolic heterogeneity prior to diabetes onset and its association with the future risk of diabetes and related complications. The study proposes an innovative data-driven subtyping framework that enables more precise short-term risk prediction for diabetes and its major complications.
Diabetes and its complications represent a growing global public health burden. Importantly, substantial metabolic differences already exist among individuals before the clinical onset of diabetes. In a recent study published in the Chinese Medical Journal on February 10, 2026, Chinese investigators applied machine learning-based approaches to identify distinct metabolic subgroups in diabetes-free individuals, revealing how these latent phenotypes are associated with the 3-year risk of developing diabetes as well as major complications, including cardiovascular disease, fatty liver disease, and stroke. The findings underscore the limitations of relying on single glycemic markers alone and highlight the importance of multidimensional clinical profiling for early and accurate risk identification.
The development of diabetes is driven by complex and heterogeneous pathophysiological processes. Even in the prediabetic or diabetes-free stage, individuals differ markedly in adiposity, blood pressure, lipid metabolism, and liver and kidney function. Traditional risk assessment strategies often depend on empirically selected indicators and may fail to capture this underlying heterogeneity. To address this gap, the research team developed a novel analytical framework that combines complication-oriented clustering with weighted naive Bayes classification. By first clustering individuals based on future complication outcomes and subsequently identifying the most informative clinical features, this approach yields metabolically meaningful and clinically implementable subtypes.
The study was conducted through a multi-institutional collaboration involving Beijing Hospital, Peking University, and the Guangzhou National Laboratory. A total of 13,829 diabetes-free adults from the Kunshan Aging Research with E-health (KARE) cohort were analyzed, with independent external validation performed in a separate cohort from Beijing Jiuhua Hospital. Thirteen routinely measured clinical indicators—including age, sex, body mass index, waist circumference, blood glucose, lipid profiles, blood pressure, liver enzymes, uric acid, and kidney function markers—were used to derive three distinct metabolic subgroups with clearly differentiated clinical characteristics.
The results demonstrated marked differences in diabetes and complication risks across the identified subtypes. The low-risk subgroup exhibited generally favorable metabolic profiles and the lowest incidence of diabetes and cardiometabolic events. The high-risk subgroup was characterized by poor glycemic and lipid control and showed the highest 3-year risk of developing diabetes and fatty liver disease. A third subgroup displayed an intermediate diabetes risk but was distinguished by older age, elevated blood pressure, and obesity, conferring the highest risks of cardiovascular disease and stroke. Survival analyses confirmed stable and reproducible risk gradients across diabetes, fatty liver disease, cardiovascular disease, and stroke.
Notably, the investigators further integrated polygenic risk score analyses to examine genetic susceptibility across subtypes. Genetic risk for diabetes and cardiovascular disease was more pronounced in the corresponding high-risk subgroups, largely mirroring observed clinical outcomes. These findings suggest that genetic predisposition and metabolic phenotype act synergistically in shaping disease trajectories.
The authors emphasize that this data-driven subtyping strategy, based exclusively on routinely available clinical indicators, offers a practical tool for identifying high-risk individuals before diabetes onset. By enabling stratified and targeted prevention strategies, this approach holds considerable promise for implementation in primary care, community health screening, and population-based chronic disease management programs.
Overall, this study reveals substantial metabolic heterogeneity in diabetes-free adults and translates it into an actionable risk stratification framework. The findings provide important evidence to support precision prevention and early intervention strategies for diabetes and its major complications, advancing the integration of artificial intelligence into real-world metabolic disease prevention.
Reference
DOI: https://doi.org/10.1097/cm9.0000000000003953
About Lixin Guo from Beijing Hospital
Lixin Guo is currently a professor at Beijing Hospital. Main research directions include Pathogenesis and intervention of diabetic angiopathy, diabetes-related diseases, endocrine and metabolic diseases in the elderly.He has published more than 210 research papers and edited more than 20 endocrine monographs.
Chinese Medical Journal
Observational study
People
Data-driven subgroups for 3-year risk stratification of incident diabetes and complications in diabetes-free Chinese adults
10-Feb-2026
None