Risk analysis of a closed-loop artificial pancreas based on generalized predictive control

Abstract

An improved generalized predictive control (GPC) algorithm with two adaptive strategies, namely, an adaptive reference glucose trajectory (AT) and an adaptive softening factor (AF), was proposed for artificial pancreas systems (AP) in our previous research. Tests with the UVA/Padova type 1 diabetes mellitus simulator (T1DMS), approved by the US Food and Drug Administration, showed that it realized an effective control of the blood glucose concentrations (BGCs) of adult and adolescent patients with type 1 diabetes. Here, risk analysis was further performed for the GPC controllers with 20 in-silico subjects (10 adults and 10 adolescents). Two indexes provided by the UVA/Padova T1DMS, including low blood glucose index (LBGI) and high blood glucose index (HBGI), were used to analyze the long-term risks for hypoglycemia and hyperglycemia of the GPC controllers. Results showed that both adult and adolescent groups had minimal risks for hypoglycemia and hyperglycemia with our GPC controllers. Moreover, AT strategy played a better role in preventing hypoglycemia and AF strategy played a better role in preventing hyperglycemia. Thus, the GPC+AT+AF controller is effective and safe, and it could be potentially applied in the AP systems.