Impact of BGM Point Profiles on Glycemic Variability Prediction Accuracies

This study investigates the effect that the number of blood glucose measurements taken per day has on a machine learning model's ability to predict three glycemic variations for the next day: the mean blood glucose value, the occurrence of a hypoglycemia event, and the time-in-range percentage. Blood glucose monitoring (BGM) data from Novo Nordisk clinical trials was used to assess the prediction performances of 1, 4, 7, and 9point profiles (PP) in comparison to continuous glucose monitoring (CGM) data. The results showed that BGM data predictions can perform equivalently to CGM data, most consistently when using a 9-point profile, and competitively at the 7PP. However, a 9PP is not the most viable option, since it would require patients to prick their fingers 8 times per day. Further analysis of the deficiencies in lower point profiles such as 5PP and 6PP is required in order to understand optimal finger prick timings.