Abstract
Disclosure: R.A. Fraser: None. L.E. Egede: None. R.J. Walker: None. J.A. Campbell: None. O.I. Ekwunife: None.
Background: Traditional analyses of continuous glucose monitoring (CGM) data often focus on summary metrics, such as percent time in range, which may not capture the temporal dynamics critical for understanding glucose variability and patient behavior. This analysis leverages Functional Data Analysis (FDA) to incorporate the temporal structure of CGM data, enabling deeper insights into glucose variability and patient profiles that support personalized diabetes management. Methods: FDA techniques were applied to 10 days of 5-minute CGM data from 1,067 participants aged 40 and older in the AI-READI (Version 2.0.0) dataset. The cohort was 57% female and racially/ethnically diverse (29% White, 27% Black, 23% Asian, 21% Hispanic), spanning the full glycemic spectrum from normoglycemia to insulin-treated type 2 diabetes. Analyses were conducted at both individual and population levels to capture within-person trends and between-group differences. Temporal glucose profiles were generated with 95% confidence intervals adjusted for correlation and multiplicity. Clustering based on Functional Principal Component Analysis (FPCA) scores identified dominant variability patterns. Longitudinal changes in FPCA scores were examined to monitor improvements or deteriorations in glycemic control. Additionally, with glucose levels as the outcome, we demonstrate how FDA can be used to compare groups with CGM data while adjusting for covariates. Results: FDA uncovered distinct temporal profiles of glucose variability linked to behavioral rhythms, such as meal timing and physical activity. At the individual level, for example, FPCA revealed three dominant components explaining 83% of the variability. Individuals clustered by FPCA scores showed clinically relevant groupings, including frequent overnight hyperglycemia and rapid post-meal spikes. At the population level, glucose levels varied significantly across study groups (p < 0.001), with distinct and highly nonlinear 24-hour rhythms observed in each group, while age (p < 0.001) remained significant after adjustment. The model explained 17.1% of the variability in glucose levels. Conclusion: FDA provides an interpretable, clinically relevant approach to identifying meaningful temporal patterns from CGM data that can guide personalized care, optimize treatments, and improve outcomes. By incorporating FPCA and derivative analyses, this method highlights critical periods for targeted intervention, supports longitudinal monitoring of treatment response, and helps characterize heterogeneous patient trajectories. These insights can complement standard CGM metrics and may be translated into algorithmic tools embedded in CGM software or EHR systems—offering real-time feedback for clinicians and empowering patients to optimize self-management.
Presentation: Monday, July 14, 2025