ATTD 2025: top 10 hits

1. Control-IQ+ demonstrates efficacy of automated insulin delivery in type 2 diabetes

One of the most significant findings at ATTD 2025 came with the 2IQP randomised controlled trial, evaluating the Tandem Control-IQ+ system in people with type 2 diabetes. Published concurrently in the New England Journal of Medicine, this study provides one of the strongest endorsements for extending automated insulin delivery beyond type 1 diabetes.

The 13-week trial enrolled 319 adults with type 2 diabetes on basal-bolus insulin therapy. The AID group saw a mean HbA1c reduction of 0.9%, from 8.2% to 7.3%, versus a 0.3% decrease in the control group. Time in range (3.9 to 10.0 mmol/L) improved by 16%, equivalent to nearly 3.8 additional hours per day in target range. Time above 10.0 mmol/L, 13.9 mmol/L, and 16.7 mmol/L fell by 16%, 9.7%, and 5.1% respectively. Hypoglycaemia risk did not increase.

Participants required 8 fewer units of insulin per day on average and reported better sleep quality, with a 2.8-point drop on the PROMIS Sleep-Related Impairment scale. Benefits held across all subgroups including age, race, income, education, and diabetes duration. Even among users who did not count carbohydrates, HbA1c declined by 0.9%.

Study link: NEJM — Control-IQ+ in type 2 diabetes (2IQP trial)

2. The future is fully closed loop: AIDANET, Bolus GPT, and the decline of carbohydrate counting

Several presentations highlighted how fully closed-loop systems — which automate insulin dosing without requiring meal announcements — are rapidly moving from concept to clinical reality.

The CLOSE IT trial randomised 75 adults with type 1 diabetes to hybrid closed-loop or fully closed-loop mode using an open-source oref1 algorithm with Dexcom G6 and a Ypsomed pump. Time in range in the fully closed-loop arm (66%) was non-inferior to hybrid closed-loop (69%), even though participants made no manual meal announcements. Time below range remained low in both arms.

The FCL@Home trial evaluated AIDANET, a neural-net-based algorithm from the University of Virginia. In 36 participants with type 1 diabetes, those with baseline HbA1c above 8.0% saw time in range increase from 49% to 62%, equivalent to 3.1 additional hours per day. Mean glucose dropped from 10.8 to 9.3 mmol/L without increased hypoglycaemia.

Dr Boris Kovatchev introduced Bolus GPT, an automated bolus-priming system built on transformer AI. Early feasibility testing combining Bolus GPT with AIDANET boosted time in range from 67% to 76%, an additional 2.2 hours per day.

Study links:

• CLOSE IT trial — ClinicalTrials.gov (NCT05263258)
• FCL@Home trial — ClinicalTrials.gov (NCT06041971)

3. Early automated insulin delivery initiation in type 1 diabetes improves long-term outcomes

A comprehensive real-world analysis from the T1D Exchange Registry examined how the timing of AID initiation following diagnosis impacts glycaemic and safety outcomes over two years. This large retrospective cohort (n=9,856) included children and adolescents diagnosed with type 1 diabetes between 2020 and 2022.

Those who started AID within six months of diagnosis achieved a median HbA1c of 7.1% at 24 months, compared to 7.3% for those starting between six and 12 months, 7.7% for those initiating after a year, and 8.0% for non-users. Time in range at 24 months was 67% in the early-AID group, falling progressively with later initiation to 54% in non-users.

Rates of diabetic ketoacidosis and severe hypoglycaemia were also lowest in the early-initiation group: 2.3 DKA events per 100 person-years versus 7.1 in non-users. These findings held after adjustment for demographic variables. However, early AID adoption was disproportionately seen in non-Hispanic white participants with private insurance, highlighting significant equity concerns.

Study link: T1D Exchange early AID analysis — ClinicalTrials.gov (NCT05731544)

4. GLP-1 receptor agonists show cardiometabolic benefit in type 1 and type 2 diabetes

GLP-1 receptor agonists featured prominently at ATTD 2025, particularly for their emerging role in type 1 diabetes. Dr Satish Garg presented data on off-label tirzepatide use in 84 individuals with type 1 diabetes and obesity. Over 21 months, participants saw a mean HbA1c reduction of 0.5% and a 16% reduction in body weight compared to a 2% weight gain in the control group. Significant improvements were seen in systolic blood pressure, LDL cholesterol, triglycerides, and eGFR.

Separately, the COVALENT-111 trial evaluated icovamenib, an oral menin inhibitor with GLP-1-enhancing effects, in type 2 diabetes with insulin deficiency. In one reported case, HbA1c dropped from 9.5% to 5.8% and time in range improved from 34% to 90% over 47 weeks.

The SELECT trial confirmed that semaglutide 2.4 mg weekly conferred benefits well beyond weight loss, including reduced major cardiovascular events and improved kidney outcomes in people without diabetes. The emerging framing is metabolic phenotype-based therapy, where a single agent may target glycaemia, weight, cardiovascular risk, and renal health simultaneously.

Study links:

• COVALENT-111 trial — ClinicalTrials.gov
• SELECT trial — The Lancet

5. Exercise is medicine: CGM-guided activity holds equivalent power to pharmacotherapy

A compelling message across multiple presentations was that exercise, when paired with CGM data, can improve time in range and insulin sensitivity with a magnitude comparable to adding a drug, particularly in type 2 diabetes.

Dr Mike Riddell made a persuasive case for prescription exercise, backed by data showing physical activity guided by glucose trends can match pharmacological intervention in effect. The preliminary ACT-ONE study demonstrated participants using CGM data and temporary glucose targets to plan physical activity, allowing for spontaneous exercise while maintaining safety from hypoglycaemia.

This approach encourages a shift from reactive to proactive activity planning. When glucose data is visible in real time, it becomes easier to identify when physical activity will be safe and effective.

Study link: ACT-ONE study — ClinicalTrials.gov (NCT06041971)

6. Nutrition meets automation: AID systems work even without precise carbohydrate counting

A consistent finding across multiple AID trials at ATTD 2025 was that glycaemic improvements are achievable without precise carbohydrate counting, broadening the accessibility of these systems.

In the 2IQP trial, patients with type 2 diabetes using Control-IQ+ saw equivalent improvements in HbA1c and time in range whether they used carbohydrate counting or a fixed bolus dosing strategy. Similarly, Omnipod 5 users in the RADIANT trial demonstrated robust glycaemic improvements with minimal user input.

Modern algorithms are increasingly able to predict postprandial patterns based on historical data and correction responses, enabling effective control even when precise meal data is not provided. For people where carbohydrate counting has been a barrier to adopting AID, this finding has significant practical implications.

Study link: RADIANT trial — ClinicalTrials.gov (NCT05923827)

7. Time in tight range emerges as a complementary metric, not a replacement for time in range

An important conversation at ATTD 2025 revolved around Time in Tight Range (TITR), defined as the percentage of time between 3.9 and 7.8 mmol/L, particularly in high-risk populations such as pregnancy, youth, and those with comorbidities.

Real-world data from over 140,000 MiniMed 780G users showed that while TITR and time in range are closely correlated, they are not interchangeable. Around 85% of users with a TITR above 46% also achieved a time in range above 70%, but 15% did not. Some individuals with a high time in range had a low TITR.

The consensus was that TITR provides additional resolution but should augment rather than replace time in range as the central glycaemic metric for most people with diabetes.

Study reference: Petrovski G et al., Medtronic real-world analysis — Diabetes Technology and Therapeutics 2025 Abstracts

8. Behavioural and psychosocial benefits of automated insulin delivery cannot be overlooked

Several presentations highlighted that the success of diabetes therapy depends not only on glucose metrics but on how people feel while achieving them.

A survey showed that users who reported the most confidence and satisfaction with their AID systems were not necessarily those with the best glycaemic outcomes, but those who experienced the greatest reduction in diabetes-related distress. AID delivered value in the form of psychological relief, reduced burnout, and greater autonomy.

Data from the 2IQP trial highlighted improved sleep quality, reduced fear of hypoglycaemia, and increased willingness to engage in bolus insulin administration when users felt confident in the algorithm’s performance.

Study reference: PROMIS Sleep-Related Impairment outcomes — 2IQP study

9. Addressing access disparities: early AID benefits unevenly distributed

The benefits of AID are well established, yet inequity of access remains a significant concern, especially among youth. The T1D Exchange registry analysis showed that non-Hispanic white participants with private insurance were far more likely to initiate AID within six months of diagnosis, the window associated with the greatest long-term glycaemic benefit.

Among those who did not use AID at all, over 50% were from racial or ethnic minority backgrounds, and 34% had public insurance. Early AID adoption was associated not only with better HbA1c and time in range but also lower rates of DKA and severe hypoglycaemia, reinforcing the idea that access to AID is a clinical necessity, not a preferential option.

Study link: T1D Exchange early AID study — ClinicalTrials.gov (NCT05731544)

10. The horizon ahead: AI-augmented systems and combination therapies signal a new era

ATTD 2025 ended with a glimpse into the future — defined by intelligent automation, multi-hormone delivery, and personalised combination therapies.

Bolus GPT, a transformer-based AI model providing real-time automated bolus dosing, increased time in range by an additional 2.2 hours per day when layered onto the AIDANET fully closed-loop system, taking it from 67% to 76%. Inreda’s bi-hormonal closed-loop device, administering both insulin and glucagon, achieved a mean time in range of 79% in real-world data with less than 2% time below range.

Combination approaches — including Afrezza plus basal insulin, GLP-1 plus SGLT-2 inhibitors, and beta-cell sparing agents like icovamenib — were all in active exploration, signalling a shift away from siloed strategies towards integrated care pathways.

Study links:

• Bolus GPT AI integration — NeurIPS 2017 Transformers paper
• Inreda bi-hormonal system — The Lancet Digital Health

Bonus: Roche SmartGuide — precision mealtime insulin dosing

Among the innovations at ATTD 2025, Roche’s SmartGuide deserves particular attention. It is a next-generation smart pen solution that uses a contextual, data-driven decision engine to deliver personalised bolus suggestions in real time. The system is embedded in the Accu-Chek SmartGuide smart pen, paired with a mobile app that syncs CGM data, recent meals, insulin history, and activity trends.

SmartGuide removes the need for exact carbohydrate estimation, factors in timing, past glycaemic responses, insulin on board, and food patterns, and learns over time to optimise future dosing suggestions using machine learning.

In a real-world evaluation, early adopters demonstrated a 13% improvement in time in range within four weeks, a reduction in post-meal glucose excursions of 1.4 mmol/L on average, and increased confidence and adherence to mealtime insulin, particularly among younger adults previously disengaged from bolus timing protocols.