Key points are not available for this paper at this time.
Fully automated insulin delivery (FAID) systems aim to regulate blood glucose levels in individuals with type 1 diabetes with minimal user input. A key challenge to achieving full automation is the need for manual meal announcements. This review focuses on algorithms developed to detect and compensate for unannounced meals. A systematic literature search was conducted across PubMed, Scopus, IEEE Xplore, and Web of Science (2000 – 2025), and the review was reported in accordance with PRISMA guidelines. From 1205 initially retrieved articles, 69 studies met predefined eligibility criteria and were included in the final review. Extracted data included algorithm type, data source (in-silico or clinical), performance metrics, and glycemic outcomes where available. The reviewed algorithms include heuristic rule-based methods, machine learning models, and control systems theory approaches. While varying in complexity and detection strategies, these approaches aim to identify meal events and estimate carbohydrate intake for timely insulin dosing. Median performance across studies included a sensitivity of 88%, precision of 93%, and detection times ranging from 25 to 40 minutes. In-silico evaluations generally reported more false positives than in-vivo. Reported glycemic outcomes demonstrated time-in-range values between 65% and 89%, highlighting the potential of these systems to support fully autonomous glucose regulation. Automated meal detection and compensation algorithms show promise for integration into FAID systems, with encouraging detection performance and glycemic outcomes. However, the evidence remains heterogeneous, and a substantial portion is still based on in-silico studies rather than clinical validation. Challenges remain in minimizing false positives, ensuring generalizability across populations, and validating performance in real-word settings.
Ibrahim et al. (Thu,) studied this question.