980-P: Leveraging Machine Learning to Enhance Blood Glucose Management—Insights from Digital Health Platform

by sugaradmin



Introduction and Objective: Integrating digital health and machine learning (ML) is reforming clinical care by offering a personalized, data-driven approach to optimize interventions. This study utilizes Dario’s digital health platform data to gain insights into blood glucose (BG) patterns influenced by clinical and engagement factors, for developing personalized interventions.Methods: Retrospective data analysis was performed on high-risk T2D users (BG >180 mg/dL at month 1, equivalent to eA1c 8.0) who measured their BG since 2020 for >2 months (m). A piecewise linear mixed-effects (LME) model defined time-related fluctuations in monthly average BG (AvgBG) with cutoff at 4m based on a year data visualization. An ML approach was adopted using a piecewise LME decision tree to identify factors moderating BG trajectories such as no. of comorbidities, any insulin usage, self-reported BMI, year of diagnosis, and monthly number of BG measurements.Results: A group of 22414 users was evaluated (Age 57.5 SD ±12.1; 46% females). Significant decrease in the AvgBG was observed in the first 4m (B=-6.8, p<.001) followed by a milder decrease in the next 8m (B=-.3, p<.001). The clinical tree first split based on insulin usage and then on the year since diagnosis, categorizing users into <5 and ≥5 years. In the first 4m, insulin users with <5 years since diagnosis had the strongest decrease in AvgBG (B=-14.3, p<.001). BG monitoring tree segmented into 2 subgroups: ≤12 and >12 monthly measurements. There was a significant decrease in the AvgBG in the first 4m for both groups (≤12: B=-6.4, p<.001; >12: B=-9.0, p<.001), with a stronger decrease for the >12 group. In the following 8m, >12 group demonstrated an additional decrease in the AvgBG (B=-1.3, p<.001) yet the ≤12 experienced no change (p=.5).Conclusion: These findings highlight the importance of using ML to understand clinical and engagement factors enabling the development of personalized interventions on digital health platforms.

Disclosure

Y. Fundoiano-Hershcovitz: Employee; DarioHealth Corp. I. Breuer Asher: None. O. Manejwala: Employee; DarioHealth Corp. Stock/Shareholder; DarioHealth Corp.



Source link