AI may facilitate fully closed-loop insulin delivery
medwireNews: A closed-loop system with an artificial intelligence (AI) component that rapidly detects when the user is eating a meal may offer a route towards fully closed-loop insulin delivery, shows a small crossover trial in adolescents and young adults.
Presenting the findings at the virtual 2021 ATTD conference, José García-Tirado (University of Virginia, Charlottesville, USA) noted that adolescent users are particularly prone to skipping meal boluses, and in-development systems with the ability to detect meals usually do so too late to avoid significant postprandial hyperglycemia.
He explained that the closed-loop system used in this study, known as Rocket AP, contains an AI bolus priming system that detects disturbances in continuous glucose monitoring data and calculates the probability of them being caused by a meal. It then delivers an insulin bolus as a percentage of total daily insulin need, with the percentage delivered increasing with rising probability of the detected glucose disturbances being caused by a meal.
The system performed well when tested in a randomized crossover design against a research version of a commercially available hybrid closed-loop control algorithm in 19 adolescents and young adults, aged between 13 and 20 years, who were experienced insulin pump users.
The two systems were equivalent during a meal when the users took an insulin bolus, but during one when they did not, the Rocket AP system reduced postprandial glucose levels significantly faster than the control algorithm.
The median time in range after a bolused meal was 100% with the Rocket AP system and 93% with the control system, but was a corresponding 83% versus 53% after a meal without a bolus.
The AI system detected the probable meal around 10 to 15 minutes after the user began eating, as compared with the “typical” 40 to 50 minutes achieved by most previously reported meal-detection systems, said Garcia-Tirado.
He noted that the rate of hypoglycemic events was higher than expected with both systems, with no clear contributing factor.
The next steps, said the presenter, are to test the system with the addition of a meal-announcement feature, and also with a “faster experimental insulin analog.”
medwireNews is an independent medical news service provided by Springer Healthcare Ltd. © 2021 Springer Healthcare Ltd, part of the Springer Nature Group