CGM technology allows care teams to remotely monitor glucose levels of people with diabetes. Provider availability to monitor and contact people is limited. Individuals respond differently to CGM review calls; some people improve glucose control while others disengage. No mathematical framework exists to optimize population-level care while accounting for provider constraints and differences in people’s responses. We hypothesize that a care team’s call schedule can be optimized using predictive modeling and personalized recommendations.

We designed a recommendation tool that ranks people based on expected improvement in glucose management due to a call, incorporating people’s responsiveness. Individuals’ short- and long-term improvements after simulated previous calls were used to calculate the expected decrease in mean glucose due to a new call.

The tool is designed with 724,672 hours of CGM data from 120 individuals with T1D (63% new diagnosis, 57% ≤ 18 years), through Tidepool. For a daily capacity of 5 calls, the recommendation has a 29.1 mg/dL expected decrease in mean glucose in contacted people, while current practice has a 4.7 mg/dL decrease.

The tool allows for customizable care team constraints and adaptively learned responsiveness parameters. CGM and call data are currently being collected from pediatric patients with newly diagnosed T1D in our clinic, to tune and test the tool in a clinical setting.


J. Vallon: None. A.T. Ward: None. P. Prahalad: None. K.K. Hood: Research Support; Self; Dexcom, Inc. Speaker’s Bureau; Self; LifeScan, Inc., MedIQ. D.M. Maahs: Advisory Panel; Self; Eli Lilly and Company, Insulet Corporation, Medtronic, Novo Nordisk A/S. Consultant; Self; Abbott, Sanofi. Research Support; Self; Bigfoot Biomedical, Dexcom, Inc., Roche Diabetes Care, Tandem Diabetes Care. D. Scheinker: None.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at