Model-based sensor-augmented pump therapy

Abstract

Background: In insulin pump therapy, optimization of bolus and basal insulin dose settings is a challenge. We introduce a new algorithm that provides individualized basal rates and new carbohydrate ratio and correction factor recommendations. The algorithm utilizes a mathematical model of blood glucose (BG) as a function of carbohydrate intake and delivered insulin, which includes individualized parameters derived from sensor BG and insulin delivery data downloaded from a patient's pump.

Methods: A mathematical model of BG as a function of carbohydrate intake and delivered insulin was developed. The model includes fixed parameters and several individualized parameters derived from the subject's BG measurements and pump data. Performance of the new algorithm was assessed using n = 4 diabetic canine experiments over a 32 h duration. In addition, 10 in silico adults from the University of Virginia/Padova type 1 diabetes mellitus metabolic simulator were tested.

Results: The percentage of time in glucose range 80-180 mg/dl was 86%, 85%, 61%, and 30% using model-based therapy and [78%, 100%] (brackets denote multiple experiments conducted under the same therapy and animal model), [75%, 67%], 47%, and 86% for the control experiments for dogs 1 to 4, respectively. The BG measurements obtained in the simulation using our individualized algorithm were in 61-231 mg/dl min-max envelope, whereas use of the simulator's default treatment resulted in BG measurements 90-210 mg/dl min-max envelope.

Conclusions: The study results demonstrate the potential of this method, which could serve as a platform for improving, facilitating, and standardizing insulin pump therapy based on a single download of data.

Figure 1

The conceptual model of the glucose dynamics. The insulin pharmacodynamics is constructed by a known pharmacokinetics model and an individualized second-order part (shaded rectangular). The overall BG response is a superposition between the insulin and meal parts.

Figure 2

A single bolus followed by a postprandial square bolus. I_Basal is the optimal basal rate, and SqB is the postprandial bolus factor

Figure 3

Dog 1: Three nominal therapy days, the model-based insulin therapy experiment result, and the theoretical model-predicted therapy (black, orange, and blue curves, respectively). The grey curve shows the unexpected elevation in BG in the first control day.

Figure 4

Dog 2: Three nominal therapy days, two model-based insulin therapy experiment results, and the theoretical model-predicted therapy (black, orange, and blue curves, respectively). The grey curves show the unexpected elevation in BG in one of the control days and one of the model-based therapy experiments.

Figure 5

Dog 3: Three nominal therapy days, the model-based insulin therapy experiment result, and the theoretical model-predicted therapy (black, orange, and blue curves, respectively). The grey curves show the unexpected elevation in BG in two of the control days.

Figure 6

Dog 4: Three nominal therapy days, the model-based insulin therapy experiment result, and the theoretical model-predicted therapy (black, orange, and blue curves, respectively). The grey curves show the unexpected elevation in BG in two of the control days.

Figure 7

Control variability grid analysis plot comparing the UVa/Padova simulator’s default therapy and the model-based insulin therapy (pink rectangles and blue circles, respectively).