Automatic and online fault detection of sensor problems using continuous glucose monitoring data for type 1 diabetes

Wide use of continuous glucose monitoring (CGM) provides sufficient time-series sensor data, which covers sufficient knowledge about the underlying correlations of glucose concentrations and the progressing dynamics over time direction. From self-Monitoring of blood glucose to continuous glucose monitoring, sensor performance is key for successful clinical use. However, sensor problems have not been analyzed and detected although they are very popular problems in real case and may result in unreliable CGM data. In the present work, an automatic fault detection method is proposed to identify sensor problems on line based on CGM data. Different from conventional CGM monitoring which focuses on direct realtime display of CGM readings, it focuses on detecting the undesirable sensor disturbances by analyzing the underlying time-wise glucose correlations. Different types of monitoring charts are developed and used for fault detection of sensor problems. Comparison is also made between different monitoring charts for two important types of disturbances which are simulated by exponential change and step change. During online monitoring, the new glucose dynamics are then tracked against predefined confidence limits to indicate abnormality. The feasibility of the proposed method to serve as a completely new glucose monitoring and performance evaluation engine is successfully assessed using CGM data collected from the Food and Drug Administration (FDA)-accepted University of Virginia/University of Padova (UVa/Padova) metabolic simulator.