A Control-Relevant Model of Subcutaneous Insulin Absorption

Abstract

Listen

Translate article

Critically ill patients suffer from “stress hyperglycemia,” a diabetes-like condition of elevated glucose concentrations. The outcomes from controlling glucose levels are mixed; some trials have shown significant reductions in morbidity and mortality rates for patients, while the NICE-SUGAR trial debates this result. The current state of critical care practice is a conservative approach to glucose control, where physicians maintain glucose levels via strategic administration of insulin, and the result is mild-to-moderate hyperglycemia for patients. Automating this insulin delivery process can improve glucose control, while mitigating hypoglycemia, using mathematical model-based tools. Key to clinical implementability and performance is a subcutaneous insulin delivery model. The proposed model is a reduction of an extended Wilinska model (Wilinska et al. (2005)) that captures plasma insulin dynamics after insulin administration. The proposed model holds for regular and rapid-acting insulins administered via bolus injection or continuous infusion. The model parameters were fit to clinical data from insulin-dependent diabetics and healthy patients administered insulin. Regular insulin data were fit simultaneously across three dose levels by adjusting the rate parameters. Fast acting insulin data were simultaneously fit separately from regular insulin, and similarities and differences between fast-acting insulin analogues were observed. The subcutaneous model, integrated with our previously-published whole body model (Roy and Parker (2006)) is able to accurately capture plasma glucose levels of patients published in the literature. Our overall objective is to couple this system-level model to a model-based control algorithm to facilitate clinical decision-making for glucose control and insulin delivery in critical care.