. In addition, our proposed framework can be easily expanded to integrate and incorporate continuous streams of other relevant data sources, such as from wearable activity trackers like Fitbit, to discover the impact of daily activity level, sleep, stress, circadian rhythm, and much more, on glycemic control. Our proposed method is intuitive and it provides interpretable and actionable insights that can further guide daily management strategies to improve diabetes outcomes.
This study leverages over 79,000 days of CGM and insulin pump data from 250 subjects for four objectives: 1) define objective features from digital data, 2) model digital biomarkers and glycemic control, 3) classify glycemic control with digital biomarkers, 4) identify the most impactful digital biomarkers associated with good and poor glycemic control and consistent across two distinct datasets, i.e., sensor augmented pump and hybrid closed loop .