Population and Individual Level Meal Response Patterns in Continuous Glucose Data

Diabetes research has changed with the introduction of wearables that are able to continuously collect physiological data (e.g., blood glucose levels), which has allowed for data-driven solutions. In this context, patients are still expected to self-record events tied to their daily routines (e.g., meals). Since self-recording is prone to errors, automatic detection of meal events could improve the quality of event data and reduce registration burden. In this paper, we investigate the feasibility of meal detection from continuous glucose data by using population level data compared to individual data. We discuss the advantages and disadvantages of both approaches based on a method to identify patterns in time series that can be used to map the characteristics of a glucose signal response to a meal event. Event responses, i.e., subsequences that come right after a recorded event, are identified and fuzzy clustering is used to group different types of them. Our results indicate that both population and individual data give comparable results, which suggests that both could be used interchangeably to develop event identification models.