EcoSta 2022: Start Registration
View Submission - EcoSta2022
A0918
Title: Function-on-function quantile regression model for predicting glucose levels and excursions Authors:  Jeong Hoon Jang - Yonsei University (Korea, South) [presenting]
Abstract: A hypoglycemic event is characterized by a low excursion of blood glucose level (less than 70 mg/dL) that can lead to serious problems, including seizures or unconsciousness. The goal is to build a functional regression model that leverages glucose curve data (real-time continuous glucose measurements) collected by a wearable glucose monitor to enable accurate and timely prediction of future glucose levels and hypoglycemic events, ultimately providing patients with sufficient time to take necessary preemptive actions. One challenge is that hypoglycemic events denote low glucose excursions that may not be well captured and predicted by the means. Hence, we develop a function-on-function quantile regression model that can reveal how the entire distribution of the functional response (future glucose curve) varies with the functional predictor (current and recent glucose curve) in ways that might not be captured by mean regression. The model incorporates random effects to account for subject-specific glucose patterns and also allows the model parameters to depend on latent classes to capture characteristics of distinct unobserved subgroups of the population. An efficient Bayesian estimation scheme based on asymmetric Laplace likelihood is presented. The predictive performance of the proposed method is examined through simulations and real patient data collected by Indiana University Hospital.