AI tools advance diabetes management

Artificial intelligence in diabetes care enhanced automated insulin delivery systems, improved glycemic outcomes, and showed promise for diabetes education and prevention programs, according to findings in a mini-review of recent diabetes technology research in The Journal of Clinical Endocrinology & Metabolism.

The review, led by Priya Prahalad, MD, PhD, a clinical professor and pediatric endocrinologist with Stanford Children's Medical Health in Paolo Alto, Calif., summarized studies published between March 2024 and December 2025 examining advancements in several emerging diabetes technology, including artificial intelligence (AI).  

The researchers noted that one of the earliest clinical applications of AI in diabetes care was diabetic retinopathy screening. The first FDA-authorized autonomous AI diagnostic system for diabetic retinopathy was cleared in 2018, with two additional systems subsequently authorized.  

The review highlighted developments in other areas of AI, such as automated insulin delivery algorithms, digital twin simulations, large language models (LLMs), and AI-guided lifestyle interventions. Here are the main takeaways.  

AI-assisted AID algorithms show promise. In the single-group pilot PEDAP-AI study, researchers incorporated AI into automated insulin-delivery (AID) algorithms for children aged 2 to 6 years with type 1 diabetes. After eight weeks, time in range improved 6 percentage points from baseline, and time below 54 mg/dL and time above 250 mg/dL were noninferior to baseline. However, more work is needed before bringing AI-powered AID algorithms into clinical practice, according to the researchers. 

Digital twins may enhance glucose management. A digital twin is a computer simulation that integrates a patient's genotype, phenotype, physiology, and behavior. In 1 randomized clinical trial, integrating digital twin technology with automated insulin delivery systems increased time in range from 72% to 77% and reduced glycated hemoglobin (HbA1c) levels from 6.8% to 6.6%. 

Another randomized trial, in which 28 adults were enrolled, evaluated a digital twin-enabled bolus calculator in adults with type 1 diabetes and reported a 7% increase in time in range and lower rates of hypoglycemia compared with traditional carbohydrate counting approaches. 

Digital twin interventions have also been associated with improvements in patients with type 2 diabetes. In a retrospective real-world study, users of a digital twin program incorporating precision nutrition, physical activity, sleep guidance, and breathing exercises experienced a mean 1.8% decline in HbA1c levels after one year, with 89% of participants achieving levels below 7%. Participants also lost a mean of 4.8 kg and demonstrated improvements in markers of insulin resistance and beta-cell function. 

LLMs support diabetes education. A study comparing responses from 5 large language models (LLMs) to 20 common questions asked by pediatric patients with type 1 diabetes and their parents, and rated by pediatric endocrinologists, found that ChatGPT-4o and Gemini Advanced generated above-average, accurate, and patient-friendly responses. 

The researchers noted that diabetes-specific chatbots may become a future application of LLMs. However, studies of AI-assisted carbohydrate estimation suggest performance varies by model, meal type, and the amount of contextual information and shows enough error that validation is needed before use for insulin-dosing decisions. 

AI behavioral tools show early promise. AI-based behavioral interventions also showed feasibility in diabetes prevention and psychosocial support. In a randomized clinical trial involving adults with prediabetes and overweight or obesity, an AI-led Diabetes Prevention Program was noninferior to human-led coaching, although participants still preferred human coaches. 

In a 12-week, single-group feasibility pilot study of 40 adolescents aged 12 to 16 years with type 1 diabetes, use of the self-compassion chatbot COMPASS was associated with reductions in diabetes distress and improvements in resilience, self-efficacy, self-compassion, and emotional well-being.  

While the review focused on summarizing the advancements and positive outcomes associated with diabetes care technology, researchers also emphasized broader concerns regarding equitable access to emerging diabetes technologies, including AI-enabled tools, particularly in low-income countries and underserved populations. 

The researchers disclosed research support, consulting relationships, advisory board participation, and speaking honoraria involving multiple diabetes technology and pharmaceutical companies.