Skip to main content
ENDO 2026 Main banner

General Endocrinology

SAT-831 - Artificial Intelligence–Driven Analysis of Continuous Glucose Monitoring Data for Prediction of Cardiometabolic Outcomes in Type 2 Diabetes: A Systematic Review and Meta-Analysis

Saturday, June 13, 2026
9:30 AM - 4:30 PM CT


Introduction: Type 2 diabetes mellitus (T2DM) is associated with cardiovascular events and microvascular complications, resulting in high morbidity and mortality. Traditional HbA1c and capillary glucose monitoring do not capture intraday glycemic variability or hypoglycemia. Continuous glucose monitoring (CGM) provides dynamic metrics such as time in range (TIR), time below range (TBR), and glycemic variability, showing an HbA1c reduction of 0.31% (95% CI: −0.43% to −0.19%) compared to traditional methods. Artificial intelligence (AI) and machine learning (ML) improve interpretation through predictive models capable of identifying complex patterns, with hypoglycemia prediction sensitivity of 0.80 (95% CI: 0.72–0.86) and specificity of 0.92 (95% CI: 0.87–0.96). However, no systematic review has confirmed AI superiority for cardiometabolic outcomes in T2DM.

Objective: To evaluate AI/ML models applied to CGM data in predicting cardiometabolic outcomes in adults with T2DM, compare predictive accuracy (AUC-ROC, sensitivity, specificity), analyze the impact of CGM-derived metrics, and assess methodological quality using PROBAST and TRIPOD.
Methodology: A systematic review with meta-analysis was conducted following PRISMA 2020. Searches in PubMed, Cochrane, and Embase identified 82 studies, with 10 included after double-blind Rayyan screening. Studies of T2DM patients using CGM were analyzed, comparing AI-based and traditional models. Quality was assessed with PROBAST and reporting followed TRIPOD. A random-effects model was applied using RR, OR, or HR (95% CI), with heterogeneity assessed by I².

Results: AI/ML models using CGM data outperformed traditional methods in predicting cardiometabolic outcomes, with higher accuracy (AUC-ROC, sensitivity, specificity) by integrating TIR, glycemic variability (SD, CV, MAGE), and hypo/hyperglycemia. Deep neural networks and ensemble methods with clinical variables showed best performance. Higher TIR was associated with reduced albuminuria, retinopathy, cardiovascular events, and all-cause mortality. MAGE independently predicted 10-year cardiovascular risk (OR: 1.623; 95% CI: 1.198–2.316; p< 0.001). ML identified metabolic subphenotypes (muscle insulin resistance AUC 95%, beta-cell dysfunction AUC 89%, incretin deficiency AUC 88%), enabling home-based risk stratification. PROBAST indicated high risk of bias: inadequate sample size (56%), missing data (41%), and overfitting issues (39%).

Conclusion: AI-driven CGM analysis shows strong potential for predicting cardiometabolic outcomes in T2DM, outperforming conventional methods in assessing glycemic variability, TIR, and cardiovascular risk. ML models enable accurate risk stratification (AUC >85%), but clinical use requires external validation, standardization, and reduction of bias.

*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.*