Simulation-based methodology for improving the safety and efficacy of insulin therapy in critically ill patients

Abstract

Critically ill patients often have hormonal stress responses due to trauma and intensive therapy. As a result, they spend an excessive amount of time with abnormally high blood glucose (BG) levels. Computerized protocols, such as the adaptive proportional feedback (APF) protocol, provide computer-directed insulin delivery to treat stress-induced hyperglycemia and its complications, while minimizing the risk of hypoglycemia. While better than paper-based protocols, computer-based protocols are sometimes insensitive to patients with rapidly decreasing BG levels. To improve the performance of APF insulin protocols, a simulation-based methodology was developed to optimize the protocol parameters. The simulated data was gathered by modeling the time-varying BG responses of in silico patients over time that correspond to a particular selection of APF protocol parameters. Analysis began with splitting the data into test and validation sets and identifying Pareto efficient parameters settings for both sets, which allowed for visualization of the tradeoffs between the time spent in hyper- and hypoglycemia. This yielded a Pareto frontier of potential APF parameters settings that translate to treatment options that improve upon current insulin therapy. The potential improvements can be vital in ensuring safe insulin therapy in clinical settings, and the overall methodology can be applied in a variety of healthcare situations.