Development and evaluation of pid controllers for glucose control in people with type 1 diabetes mellitus

The aim of this research is to develop and evaluate a novel, model-based PID control design method to achieve normo-glycemia with an artificial pancreas. A key objective is to utilize a simple, control-relevant model that can be personalized for individual subjects based only on clinical information that is readily available, such as total daily insulin (TDI). The proposed PID design method is based on the Internal Model Control (IMC) approach and a simple dynamic model that has a single adjustable parameter that is personalized based on the subject’s TDI. The proposed controller design method was evaluated in an in silico study using the UVa/Padova metabolic simulator. Ten simulated subjects were used to determine a conservative value of the single IMC design parameter. This value provided good post-prandial responses and a reasonable degree of robustness for changes of+/- 50% in the insulin sensitivity. Then ten additional subjects were simulated in a validation study that included three meals (50, 40 and 50 g CHO) during a 30 hours. The PID controllers based on the personalized models performed better than controllers based on a fixed model. For the validation study, the average amount of time that the glucose concentration was in the desired range (70-180 mg/dL) was 88% for the personalized models and 83% for the fixed models. Similarly, the average values for the 180-250 mg/dL range were 10% and 16% for the personalized and fixed models, respectively. Neither design method resulted in hypoglycemia (<60 mg/dL). Simulation studies have demonstrated that the proposed controller design method based on personalized models is practical and superior to controllers based on a fixed model