Key points are not available for this paper at this time.
Introduction: Understanding factors that influence glycemic response to meals can inform strategies for diabetes management. Continuous glucose monitoring (CGM) facilitates analyzing glycemic regulation, but concurrent meal timing, required to study meal-induced excursions, is rarely collected. Objective: We develop a framework for extracting meal-related glucose excursions from CGM time series, modeling the excursions using interpretable features, and estimating feature variability explained by patient characteristics and instance-specific patterns. Methods: We used CGM data (Abbott Freestyle Pro) from N = 175 clinical trial participants with type 2 diabetes not using insulin (HYPNOS Trial: mean age 60; 49% female; 36% Black adults). To find excursions, we marked local maxima points in the CGM time series. We discarded points with glucose increases <50 mg/dL in a symmetric window around the point. Remaining point-windows became excursions ( Fig. 1A) . We summarized excursions using relevant features: baseline, height, and shape quantified by splines ( Fig. 1B ).We then assumed each feature could be modelled as the sum of patient-specific effects, both measured as covariates and unmeasured, and instance-specific effects ( Fig. 1C ). Covariates included demographic, biometric, and lifestyle characteristics. Using linear mixed-effects models, we estimated the effects and corresponding variance explained for each feature. Results: For baseline glucose, measured covariate effects notably explain 61.2% of variability. For both height and shape, the summed patient-specific effects explain negligible variation, 17.8% and 10% respectively ( Fig. 1D ). Conclusion: Using our novel framework, we found that instance-specific information explains most variability in glucose excursions. As diet and adjacent physical activity are the instance-specific features primarily impacting glucose metabolism, it follows that healthy lifestyle choices impact postprandial glucose patterns more than inherent traits.
Sartini et al. (Tue,) studied this question.