1945-LB: Evaluation of Proactive Targeting Models Used to Identify People with Diabetes at Risk of Being Uncontrolled during Participation in a Digital Diabetes Management Program (DDMP)

Introduction & Objective: Predictive machine learning models were developed for each month during DDMP to predict individual risk of having an estimated A1c (eA1c) ≥7.5% at 12 months for personalized Health Nudges to be deployed for improved fingerstick blood glucose monitoring (FBGM) and program engagement associated with improved glycemic control. This study evaluated short-term impact on FBGM and reduction in eA1c after 9 months on program.

Methods: Deployment of models and personalized Health Nudges launched in May 2023 to a population allowing for comparison to two control groups of DDMP participants without (C1) and randomly assigned Health Nudges (C2). Participants were selected for treatment group if they had a self-reported A1c at program initiation and adequate BG checks to eA1c at 9 months using a modified Nathan Method developed by Teladoc Health. Repeated measures ANOVA analysis was used to evaluate differences between treatment and control groups adjusting for diabetes type.

Results: Treatment (n=526) and control groups (C1 n=73, C2 n=55) were similar in age [years: 55 (11.7), 56 (8.8) and 56 (12.5)], gender [female: 48%, 44%, 51%], and diabetes type [type 2: 91.2%, 86%, 91.8%], respectively. At Month-9, average 30-day FBGM checks were 32, 30, and 28, for treatment, C1, and C2. Treatment group’s eA1c at Month 9 was 7.8% while C1 and C2 were 8.2%. Statistical analysis showed personalized Health Nudges led to significant reduction in eA1C levels from baseline (P < 0.001). C2 group showed reduction as well, although not as pronounced as the treatment group (P = 0.038). Conversely, the C1 group did not demonstrate a significant reduction in eA1C levels (P = 0.070).

Conclusion: Preliminary evaluation of machine learning models supports the ability to target DDMP users in early months on program with Health Nudges aimed at increasing program engagement that was associated with improved clinical outcomes.