How are AI models and new clinical treatments reshaping diabetes care after the 2025 ADA conference?

At the 85th Scientific Sessions of the American Diabetes Association (ADA), held June 20–23 in Chicago, artificial intelligence-driven diagnostics, advanced patch-based interventions, and promising injectable therapies took center stage. Multinational pharmaceutical developers and digital health innovators unveiled data-rich results from Phase 2 trials and real-world deployments, indicating a potential shift in both early diagnosis and day-to-day treatment paradigms for type 1 and type 2 diabetes.

Among the key developments were two late-breaking presentations leveraging machine learning to identify undiagnosed type 1 diabetes risk up to a year before symptoms arise, a first-of-its-kind patch for lipohypertrophy reversal in insulin users, and new injectable regimens that showed superior efficacy to market leaders like Ozempic® and Tresiba® in glycemic control.

What new evidence suggests that machine learning can detect type 1 diabetes risk before symptoms develop?

Artificial intelligence and machine learning models developed by researchers working with Sanofi were among the most closely watched presentations at ADA 2025. Two independent cohorts were used to train predictive models—one targeting individuals aged 0 to 24 and the other adults over 25—leveraging the NorstellaLinQ database of medical claims and lab results.

The study showed that these AI-driven models were able to predict type 1 diabetes onset up to 12 months in advance with significantly higher precision and fewer false positives than traditional methods. Sensitivity scores were 80% for younger patients and 92% for adults, compared to conventional screening tools that often detect less than 0.3% of true positives in a general screening population.

Laura Wilson, director of health economics outcomes research at Sanofi, emphasized the broader clinical utility of these findings. By embedding predictive models into real-world EHR workflows, she suggested, health systems may be able to offer early lifestyle interventions or continuous glucose monitoring before irreversible pancreatic β-cell damage occurs.

Institutional sentiment around this innovation remains cautiously optimistic. Although large-scale validation is still pending, experts note that combining AI predictions with personalized outreach strategies could form the basis of a new, proactive diabetes care model—especially for children and young adults.

How does BERT-based machine learning improve the odds of identifying presymptomatic type 1 diabetes?

Another Sanofi-supported study presented at ADA 2025 demonstrated the use of Bidirectional Encoder Representations from Transformers (BERT)—a sophisticated natural language processing tool—for clinical risk scoring in diabetes.

Using the Symphony Health Database, which contains claims data on over 75 million U.S. patients, the model was trained to distinguish individuals with early-stage type 1 diabetes from a background cohort of 2.5 million non-diabetics. Among the standout results: the BERT model achieved an odds ratio of 97.27 for correctly identifying future type 1 diabetes cases—far exceeding the 38.01 of more conventional machine learning frameworks.

The model also highlighted a critical challenge in current diagnostic practice: approximately 29% of the individuals who were later confirmed as having type 1 diabetes had initially been misdiagnosed with type 2 or other variants. Experts noted that BERT’s accuracy could reduce the diagnostic lag that often prevents early intervention.

Researchers plan to refine these models using genomics and multimodal data layers, and to validate predictions in real-world, multinational clinical settings. The long-term vision includes a fully integrated digital diagnostic layer within routine pediatric and primary care.

What does clinical data show about the new embrace patch for lipohypertrophy in insulin users?

Fremont, California-based Neodyne Biosciences presented results from a 16-week clinical trial demonstrating the efficacy of its embrace® Active Site Care patch in treating lipohypertrophy (LH), a condition affecting more than 60% of insulin users worldwide. LH is caused by repeated subcutaneous insulin injections that lead to fat accumulation and fibrotic scarring, which in turn disrupts insulin absorption and glycemic control.

Participants self-applied the patch every 10 days for 16 weeks, targeting longstanding LH lesions averaging over five years in duration. Treated lesions showed an average 20% reduction in volume, a 6.97-fold decrease in lipid content, and visibly reduced fibrosis on histological analysis.

Notably, 63% of participants were able to re-utilize previously unusable injection sites. Patient satisfaction was also high, suggesting that ease of use and comfort were aligned with long-term therapy needs.

Neodyne’s president Kelley Lipman framed the product as a foundational change in diabetes care. With CE mark and 510(k) clearance already obtained, the patch could soon become a routine intervention across endocrinology clinics. Experts anticipate institutional trials may soon begin to assess cost-effectiveness and real-world impact on insulin utilization.

How do bofanglutide injections compare to semaglutide in Phase 2 studies for type 2 diabetes patients?

Beijing-based Gan & Lee Pharmaceuticals (603087.SH) presented multiple sets of Phase 2 data around its proprietary GLP-1 receptor agonist bofanglutide (GZR18). In a 24-week Phase 2b trial involving 272 Chinese patients with type 2 diabetes, bofanglutide demonstrated superior HbA1c reductions compared to semaglutide (Ozempic®), with additional improvements in weight loss, fasting glucose, and lipid markers.

Participants receiving weekly 24 mg or biweekly 18 mg doses of bofanglutide saw HbA1c drops of -2.32% and -2.28%, respectively, compared to -1.60% in the semaglutide group. In drug-naïve subpopulations, results were even more dramatic, with some achieving nearly 3% reductions in HbA1c.

In terms of weight loss, bofanglutide-treated patients averaged between 4.26 to 6.54 kg of weight reduction, outperforming semaglutide’s 3.25 kg benchmark.

Gan & Lee noted that gastrointestinal side effects were mild to moderate, consistent with known GLP-1 RA class characteristics. Institutional observers say these results strengthen the Chinese insulin developer’s global standing, particularly as Ozempic® faces market saturation and increasing regulatory scrutiny on pricing.

What did the ADA conference reveal about Gan & Lee’s GZR4 insulin analog for once-weekly administration?

Gan & Lee also shared interim results from a 16-week Phase 2 study of GZR4, its once-weekly basal insulin analog, in comparison to insulin degludec (Tresiba®). Two patient groups were studied: one on oral anti-diabetic drugs alone and another on a combination of OADs and basal insulin.

In the latter cohort, GZR4 led to a significantly greater HbA1c reduction (-1.26% vs. -0.87%; p<0.01) and enabled more patients to reach therapeutic targets below 7% HbA1c.

Importantly, GZR4 achieved comparable glycemic control with only 40–50% of the insulin dosage required by degludec, without needing a loading dose. Safety outcomes were strong, with no severe hypoglycemic events reported.

With Phase 3 programs already underway in China, GZR4 could soon offer an alternative to daily basal injections, a move likely to resonate with both clinicians and healthcare payers focused on improving adherence and reducing long-term treatment costs.

What are the broader implications of ADA 2025 for global diabetes care innovation?

Institutional analysts interpret the ADA 2025 results as a convergence point for digital diagnostics and advanced biologics. The coupling of real-world AI models with pipeline innovations like GZR4 and bofanglutide suggests a future where earlier intervention and personalized therapy may significantly curb diabetes-related complications and costs.

Investors are watching companies like Sanofi, Neodyne Biosciences, and Gan & Lee Pharmaceuticals for potential partnerships, licensing deals, or commercial launches following these promising Phase 2 results. The ADA’s visibility as a global stage ensures that these findings will ripple into international regulatory, commercial, and payer discussions in the coming quarters.

The next phase will be crucial. While early data is promising, large-scale, multi-site trials and health economics studies will be required to validate these technologies’ readiness for broad adoption.