Meal estimation from Continuous Glucose Monitor data using Kalman filtering and hypothesis testing

Abstract

A method for estimating meal inputs from Continuous Glucose Monitoring (CGM) data is presented. The method is based on Kalman filtering and hypothesis testing, and provides estimates of the time the meal was initiated and the carbohydrate content of the meal. The sensitivity to model correctness is evaluated, and suggestions for how the method can be tuned and extended are given. The method is tested on synthetic data from two simple, individualisable models of glucose dynamics as well as on real CGM data. The method has potential as a meal detector and estimator in a data cleaning settings as well as in a real-time, artificial pancreas (closedloop glucose control) setting. Further research is needed to determine its performance on larger data sets and compare it to other methods.