TITLE:
Advances in Cardiovascular Disease Risk Prediction in People with Diabetes
AUTHORS:
Yuhuan Li, Junli Xue
KEYWORDS:
T2DM, CVD, Risk Prediction, Machine Learning
JOURNAL NAME:
Journal of Biosciences and Medicines,
Vol.14 No.9,
September
10,
2026
ABSTRACT: Cardiovascular Disease (CVD) is one of the primary complications in patients with T2DM and a major cause of death among them. However, there are significant differences in CVD risk among patients with T2DM, thus early risk prediction in this population is of critical clinical importance. In this review, CVD is used as an umbrella term; where evidence permits, outcomes are distinguished as atherosclerotic cardiovascular disease (ASCVD; principally coronary heart disease and ischemic stroke), heart failure (HF), or a study-specific composite of major adverse cardiovascular events (MACE). In recent decades, approaches for cardiovascular risk prediction in patients with diabetes have evolved substantially. With advances in medical technology, the ways to predict and assess cardiovascular diseases in this group keep improving. Initially, risk was assessed using just a single risk factor, then multi-indicator joint assessment models appeared, followed by the inclusion of biomarkers and imaging results into the models. In recent years, with the rise of AI algorithms like machine learning (ML) and deep learning (DL), lots of clinical, imaging and omics data have been used for risk prediction. Research has already shown that blood sugar, blood pressure and blood fat levels are linked to cardiovascular events, but the predictive power of a single indicator is limited. The prediction performance of the multi-indicator combined scoring model is better than single-indicator assessment, but when applied to diabetic patients, the model’s discrimination and calibration are still not ideal. Biomarkers and imaging may improve risk stratification in selected settings, but evidence that these gains can be generalised across populations or improve clinical outcomes remains limited. ML models can capture nonlinear and longitudinal patterns; however, their incremental value over well-specified statistical models is often modest when the same conventional predictors are used. Reported performance should therefore be compared only when the population, endpoint definition, prediction horizon, and validation design are sufficiently aligned. This review synthesizes advances in CVD risk prediction for T2DM and examines methodological quality, temporal data leakage, external validation, calibration, fairness, and clinical implementation.