SUN-536 - Efficacy and Safety of Artificial Intelligence in Automated Insulin Delivery Systems for Diabetes Management: A Systematic Review of Clinical Trials

Background: Artificial intelligence (AI) driven automated insulin delivery (AID) systems are increasingly used in diabetes care; however, quantitative synthesis of their efficacy and safety across clinical settings remains limited. To address this gap, we conducted a systematic review assessing the efficacy and safety of AI-based insulin delivery and decision-support systems in type 1 and type 2 diabetes.

Methods: This systematic review followed PRISMA guidelines. A comprehensive literature search across PubMed, Embase, and ClinicalTrials.gov identified 96 records. Clinical trials evaluating AI-based insulin delivery or decision-support systems in patients with diabetes were eligible. After screening and full-text review, five studies met inclusion criteria. Outcomes extracted included time in range (TIR 70–180 mg/dL) and hypoglycemia ( < 70 mg/dL).

Results: Across five trials, AI-based systems demonstrated comparable or superior glycemic control to standard approaches. In individuals with type 1 diabetes (T1D) during exercise, a hybrid machine learning model predictive control system increased TIR from 69.8% to 75.4% and reduced exercise-associated hypoglycemia. In adults with T1D, an artificial neural network achieved a TIR of 86% compared with 87% using traditional model predictive control, while significantly reducing time below range and requiring sixfold less computation. Overnight glycemic control with a fuzzy logic system improved TIR from 54.8% (control) to 76.2%, with no severe hypoglycemic events reported. In hospitalized patients with type 2 diabetes, AI-guided insulin titration achieved a TIR of 76.4% versus 73.6% with physician-led management, with fewer hypoglycemic episodes. In youth with T1D in real-world settings, AI decision support achieved a TIR of 50.2% compared with 51.6% under specialist care, with equivalent safety and no increase in hypoglycemia.

Conclusions: Quantitative evidence from clinical trials indicates that AI-based AID and decision-support systems achieve TIRs of 50.2% to 86%, are non-inferior or superior to standard care, and consistently reduce or do not increase hypoglycemia. These findings support the clinical efficacy and safety of AI in diabetes management and underscore the need for larger, longer-term trials to confirm durability and real-world impact as current practice integrates hybrid AI closed-loop systems and AI-guided insulin management with clinician and nursing oversight.

*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.*