Multi-dimensional Quantile Regression Using Polynomial Function Fitting for Insulin Sensitivity Forecasting

A significant portion of patients in the intensive care unit experience hyperglycemia, meaning abnormally high blood sugar levels. Prolonged hyperglycemia poses numerous risks, leading to organ damage in the medium-term and potentially life-threatening conditions. Therefore, insulin dosing is essential to regulate the blood sugar levels of hyperglycemic patients. The STAR protocol is an insulin dosing support, a so-called tight glycemic control protocol that uses insulin sensitivity among the patient's physiological parameters to characterize their current state. Estimating the patient's future SI is crucial for determining optimal treatment. Various methods exist for this estimation, and different metrics are available for evaluating these methods. The challenge in the estimation task is to determine not the expected value of future SI but its 90% confidence interval, which is necessary for implementing clinical protocols. The research presented in the article developed SI estimation methods using quantile regression, polynomial fitting, and neural network-based approaches. Multiple previous insulin sensitivity values were used as input parameters to improve prediction accuracy. Copyright (C) 2024 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)