Artificial intelligence (AI) is rapidly reshaping diabetes care, with expanding roles in complication screening, risk stratification, insulin optimization, and hospital operations, according to a review published in Endocrine Practice.
With 537 million people worldwide living with diabetes and projections reaching 783 million by 2045, the burden of disease continues to grow. Researchers examined how AI and machine learning (ML) are being integrated into clinical workflows to improve early detection, personalize treatment, and streamline health care delivery.
"Artificial intelligence has the potential to significantly enhance diabetes management by enabling proactive risk prediction, optimizing insulin dosing, personalizing treatment plans, and improving clinical decision-making across the care continuum," wrote Rohit Parab, MD, of the Division of Endocrinology at Emory University School of Medicine in Atlanta, and colleagues.
Screening and Early Detection
AI-driven imaging tools are advancing screening for diabetic retinopathy (DR), neuropathy, and foot ulcers. Machine learning models analyzing retinal images have demonstrated 93% sensitivity and 91% specificity for detecting DR. Three fully automated systems have received US Food and Drug Administration clearance for DR diagnosis: IDx-DR (now LumineticsCore), EyeArt, and AEYE Diagnostic Screening.
Beyond retinopathy, ML approaches are being applied to diabetic peripheral neuropathy detection. Algorithms analyzing thermograms, skin images, and corneal confocal microscopy can identify early nerve changes. Some models estimate neuropathy severity using clinical characteristics alone, potentially reducing reliance on specialized testing, noted the review authors.
Digital biomarkers represent another emerging area. Smartwatch data combined with food logs predicted interstitial glucose levels with up to 87% accuracy in one study. Wristband-based photoplethysmography models detected diabetes and prediabetes with more than 84% accuracy. Smartphone-based vascular signal analysis has also demonstrated feasibility for diabetes screening.
Risk Stratification and Disease Progression
AI-based risk engines are being developed to predict progression from prediabetes to diabetes, cardiovascular outcomes, and other complications. Predictive performance varied across models and input variables, with reported area under the curve values ranging from 0.64 to 0.93.
The Building, Relating, Assessing, and Validating Outcomes (BRAVO) diabetes model integrates 17 interrelated risk equations to simulate disease progression and complications. The model has undergone calibration against randomized controlled trial data and national survey data and has been implemented within the electronic health record.
Machine learning models integrating genetic data, polygenic risk scores, imaging, and electronic health record data have further enhanced genomic-based risk prediction, with reported area under the curve values ranging from 0.61 to 0.94.
Automated Insulin Delivery and Glycemic Optimization
Wearable technologies, including continuous glucose monitoring (CGM) and automated insulin delivery (AID) systems, are central to AI-enhanced management. Although commercially available AID systems rely primarily on deterministic control algorithms, AI-based enhancements are emerging.
In a multicenter pilot study of pediatric patients aged 2 to 6 years with type 1 diabetes, an AI-driven digital twin approach optimized pump settings and improved time in range over 8 weeks. A neural-net artificial pancreas system achieved up to 85% time in range in hybrid closed-loop mode and 75% in full closed-loop mode during supervised short-term studies, with subsequent confirmation in home settings.
Decision support systems are also assisting insulin titration. A voice-based conversational AI application improved time to optimal basal insulin dose, adherence, glycemic control, and diabetes-related emotional distress in patients with type 2 diabetes compared with standard care. In youths with type 1 diabetes, an AI-based insulin dose optimization system demonstrated noninferiority to physician-guided titration.
ML models applied to inpatient electronic health record data have predicted insulin needs more accurately than traditional guidelines, noted the researchers
Patient Self-Management and Dietary Analysis
AI-powered smartphone interventions have demonstrated improvements in glycemic outcomes. In one trial, a smartphone-based behavioral coaching intervention reduced hemoglobin A1c by 1.9% compared with 0.7% in a control group. An AI-driven dietary management platform was also associated with improved hemoglobin A1c and greater weight loss compared with routine care.
Deep learning systems analyzing food images can identify food type, portion size, and estimate calories. Emerging models integrate dietary image analysis with historical glucose data and contextual factors such as physical activity and sleep to predict postprandial glucose excursions and alert patients to potential hyperglycemic or hypoglycemic events, reported Dr. Parab and colleagues.
Administrative Applications and Implementation Challenges
AI integration extends beyond direct patient care. Applications include resource allocation, staff scheduling, billing automation, and electronic health record documentation. In one cross-sectional study, large language model–generated discharge summaries were comparable in overall quality to physician-generated summaries, although errors were present and potential harm was rated low.
Despite these advances, challenges remain. Model performance varies based on input variables and data sets, and limited demographic representation may introduce bias. Additional concerns include privacy, data security, interoperability, regulatory standards, and the need for standardized evaluation frameworks.
Efforts to promote responsible AI include explainable models to enhance transparency, frameworks for equitable evaluation of retinal screening systems, and privacy-preserving federated learning approaches that enable multicenter training without direct data sharing.
The investigators reported multiple disclosures that can be found in the published research.