Introduction: Remembering to deliver insulin before meals represents a significant burden for people with diabetes (PWD) and often results in late or entirely missed meal boluses.1,2 A new Medtronic CGM-driven retrospective meal-detection (MD) tool was developed and used to retrospectively identify missed meal events in a MiniMed™ 780G system (MM780G) pivotal trial that included an unannounced meal evaluation.
Methods: The CGM-driven missed meal detection tool integrates a Machine Learning model trained on CGM and insulin data for meal identification, combined with a local adjustment of the meal start time. Study participants (N=155, aged 7-75 years with T1D) using the MM780G were directed to abstain from announcing one meal each day for a duration of 5 days, while recording both the time and carb amount of these ‘test’ meals in a diary. The new MD algorithm was evaluated on 600 unannounced meals and its performance was characterized by recall rate and delay of missed meals.
Results: The model was able to detect 91% (recall) of test meals with a mean delay of 12.9 (38.4) minutes.
Conclusion: The Medtronic missed meal-detection algorithm can accurately mark missed meal events and support in-depth analysis of glycemic outcomes. Providing healthcare professionals with a more comprehensive understanding of mealtime behavior will enable more informed therapy decisions and the possibility of improved glycemic outcomes.
A. Benedetti: None. A. Mikhno: Employee; Medtronic. A. Roy: Employee; Metronics. J. Shin: None.
