Beyond scalar metrics: functional data analysis of postprandial continuous glucose monitoring in the AEGIS study

Figures

• 

  Fig. 1. Postprandial CGM data over several days in four individuals. Each panel corresponds to one participant. Those in the first column are normoglycemic, and those in the second column have prediabetes. Glycemic status was established according to the American Diabetes Association (ADA) criteria based on glycated hemoglobin (HbA1c) and fasting plasma glucose (FPG) values

• 

  Fig. 2. Pipeline steps for a multilevel postprandial framework

• 

  Fig. 3. Eigenfunctions of the postprandial CGM responses at the subject and meal levels

• 

  Fig. 4. Projections of Raw Data onto Eigen-Functions. Raw Trajectories (Transparent Blue): These trajectories display the original data collected over consecutive days. Estimated Participant Trajectories (Red Curve): Utilizing the scores and eigenfunctions at the individual level, these visualizations illustrate the estimated participant-level trajectories. Projected Trajectory for Each Meal (Green Curves): Constructed from the scores and eigenfunctions at both levels, these curves represent the estimated meals using the parsimonious MFPCA representation

• 

  Fig. 5. Estimated time-varying covariate coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{\beta }_{s}(t)$$\end{document} for each predictor from the multilevel function-on-scalar regression model used to predict postprandial glucose trajectories. Dashed blue curves show the point estimates, grey shading denotes the 95% confidence bands, and the dotted horizontal line at 0 marks the null effect

• 

  Fig. 6. Participant-level and model-level global \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^2$$\end{document} estimates for MFPCA models by number of eigenfunctions K, H at both levels

• 

  Fig. 7. Participant-level and model-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{R}^2$$\end{document} estimates for the chosen MFPCA model with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K = H = 3$$\end{document} eigenfunctions at both levels

• 

  Fig. 8. Marginal and Conditional \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^2$$\end{document} plotted over the course of the postprandial observation window