SAT-579 - Performance of Artificial Intelligence and Machine Learning Models in Predicting Cardiovascular Outcomes in Type 2 Diabetes: A Systematic Review

Introduction
Type 2 diabetes mellitus (T2DM) confers 2-4 fold increased cardiovascular disease (CVD) risk. Traditional prediction models demonstrate suboptimal performance due to complex non-linear interactions. This systematic review evaluated Machine Learning (ML)/ Artificial Intelligence (AI)-based models for CVD prediction in T2DM patients.

Methods
Following PRISMA 2020 guidelines, we searched PubMed, Embase, Scopus, and Cochrane databases through December, 2025. From 84 studies identified, 8 met inclusion criteria reporting discrimination metrics (AUC/C-statistic) for CVD prediction in T2DM. Data extracted included sample size, demographics, and model performance.

Results
Eight studies encompassing 162,808 T2DM patients were included. Mean age ranged from 58.3-65.6 years; % female was 38.0-55.2%. Algorithms included XGBoost, random forest, gradient boosting, ensemble methods, and deep learning. AUC/C-statistics ranged from 0.669-0.987 in validation cohorts. The largest study (Weinan 2024; n=141,516) achieved Harrell's C>0.80 for 10-year CVD prediction (female:0.764; male:0.741). Highest discrimination was Yuan 2025 (n=2,895) using ensemble learning (AUC:0.987). Most studies (6/8) achieved C-statistics>0.70. Key predictors included age, diabetes duration, renal function (eGFR, ACR), HbA1c variability, blood pressure, and inflammatory markers. Novel predictors incorporated TCM features, NMR-lipoprotein profiles, and DL-based coronary calcium scoring.

Conclusions
ML/AI models demonstrate good-to-excellent discrimination (C-statistic>0.70) for CVD risk stratification in T2DM, often outperforming traditional tools. Performance heterogeneity (AUC:0.669-0.987) reflects population differences, feature selection, and validation rigor. Integration of novel biomarkers may enhance prediction beyond conventional variables. External validation and prospective implementation studies are needed to establish real-world clinical utility in diverse T2DM populations.

*Unless otherwise noted, all abstracts presented at ENDO must not be released to the press or the public until the date and time of presentation. For oral presentations, the abstracts are embargoed until the session begins. The Endocrine Society reserves the right to lift the embargo on specific abstracts that are selected for promotion prior to or during ENDO.*