Modeling glucose and subcutaneous insulin dynamics in critical care

Model-based decision support could be used to tailor insulin treatment to patients suffering from stress hyperglycemia, while avoiding hypoglycemia. This work combines a previously published glucose and insulin model with a subcutaneous insulin delivery model, herein simplified using Markov Chain Monte Carlo optimization and Kullback-Liebler distance, to capture fast-acting and regular insulin using two shared and one type-specific fitted parameter. Glucose data from a critical care population (N=48) receiving subcutaneous insulin are fit to within finger stick glucose measurement error of 5% using a regularized, time-varying parameter. The resulting virtual patient cohort provides a basis on which automated insulin delivery systems can be tested.