A variable state dimension approach to meal detection and meal size estimation: In silico evaluation through basal-bolus insulin therapy for type 1 diabetes

Objective: This paper aims to develop an algorithm that can detect unannounced meals and estimate meal sizes to achieve a robust glucose control. Methods: A variable state dimension (VSD) algorithm is developed to detect unannounced meals and estimate meal sizes, where a Kalman filter operates on a quiescent state model when no meal is detected, and switches to a maneuvering state model to estimate meal information once the meal-induced glucose variability is statistically significant. Results: Through evaluation using 30 subjects of the UVa/Padova simulator, a basal-bolus (BB) control using the VSD-estimated meal size for each meal can achieve mean blood glucose (BG) of 142 mg/dl with an average 17.7% of time in hypoglycemia. In terms of 20 Monte-Carlo simulations for each subject over a two-day scenario, where each meal/snack has a probability of 0.5 not to be announced, the BB control using VSD for unannounced meals can achieve an average mean BG of 143 mg/dl with 8% of time in hypoglycemia, in contrast to mean BG of 180 mg/dl with 8% of time in hypoglycemia obtained by BB with missing boluses. Additionally, VSD is able to detect a meal within 45 (±14) min since its start with a 76% success rate and 16% false alarm rate. Conclusion: The addition of VSD to the BB control improves glucose control when meal announcements are missed. Significance: The VSD can be used as a complementary tool to detect meal and estimate meal size in absence of a meal announcement.