Association between continuous glucose monitoring-derived metrics and coronary plaque vulnerability: A retrospective exploratory analysis
- Department of Biochemistry and Molecular Biology, Graduate School of Medicine, The University of Tokyo, Japan
- Department of Biological Sciences, Graduate School of Science, The University of Tokyo, Japan
- Division of Diabetes and Endocrinology, Department of Internal Medicine, Kobe University Graduate School of Medicine, Japan
- Division of Cardiovascular Medicine, Department of Internal Medicine, Kobe University Graduate School of Medicine, Japan
Figures
Figure 1 with 1 supplement
Relations among clinical measures.
(A) Multiple regression analysis between necrotic core (%NC) and continuous glucose monitoring (CGM)_Mean, CGM_Std, and autocorrelation (AC)_Var. Scatter plots for predicted %NC versus measured %NC (the left). Each point corresponds to the values for a single individual. Bars represent the 95% confidence intervals (CIs) of the regression coefficients (the right). (B) Multiple regression analysis between %NC and fasting blood glucose (FBG), HbA1c, and PG120. (C) Linear regression analysis between %NC and the Framingham Risk Score for Hard Coronary Heart Disease. (D) A spring layout of the correlation network involving %NC (black), 14 CGM-derived indices (red), three blood glucose level-related indices (magenta), three insulin sensitivity or secretion-related indices (blue), and six other clinical indices (green) obtained from a single blood test or physical measurement. Connections denote relationships with Q<0.05. The width of the edges is proportional to the corresponding correlation coefficient. (E) The absolute values of Spearman’s correlation coefficients between clinical parameters and %NC. Bars represent the 95% CIs. (F) Scatter plots for AC_Var versus CGM_Mean (the left), and AC_Var versus CGM_Std (the right). Each point corresponds to a single individual’s values. Individuals are colored based on their %NC value. R is Spearman’s correlation coefficient, and the value in parentheses is 95% CI. (G, H) Variance inflation factor (VIF) of each variable remaining after removing the variables with the highest VIF one by one until the VIFs of all variables are less than 10. The input variables include the following 26 variables: BMI, SBP, DBP, TG, LDL-C, HDL-C, FBG, HbA1c, PG120, I.I., composite index, Oral disposition index (DI), CGM_Mean, CGM_Std, CONGA, LI, JINDEX, HBGI, GRADE, MODD, MAGE, ADRR, MVALUE, MAG, AC_Mean, and AC_Var. The input variables of (H) included only the CGM-derived measures from the above 26 indices. (I) Scatter plot of the VIF of the indices measured in this study (VIFt) versus that of indices measured in a previous study (VIFp1) (Sugimoto et al., 2025). R is Spearman’s correlation coefficient, and the value in parentheses is the 95% CI. (J) Scatter plot of the VIF of the indices measured in this study (VIFt) versus that of the indices measured in a previous study (Hall et al., 2018) (VIFp2).
Figure 1—figure supplement 1
Relationship among clinical parameters.
(A) Scatter plot of the autocorrelation (AC)_Var calculated from 15 min-intervals and that calculated from 5 min-intervals. R is Spearman’s correlation coefficient, and the value in parentheses is the 95% CI. (B) Multiple regression analysis between necrotic core (%NC) and CGM_Mean, Standard deviation (Std) of glucose levels (CGM_Std), and AC_Var. These CGM-derived indices were calculated from CGM data of 15 min intervals. Scatter plots for predicted %NC versus measured %NC (the left). Each point corresponds to the values for a single individual. Bars represent the 95% CIs of the coefficients of the regression models (the right). (C) Scatter plot for systolic blood pressure (SBP) and low-density lipoprotein cholesterol (LDL-C). Gray shaded areas indicate the range of values for high SBP (>140 mmHg) or high LDL-C (>120 mg/dL). Each point corresponds to the values for a single participant. Green circles, red cross marks, and blue squares indicate normal glucose tolerance (NGT), impaired glucose tolerance (IGT), and type 2 DM (T2DM), respectively. (D) Scatter plots and fitted linear regression lines for each clinical index versus %NC. Each point corresponds to the values for a single individual. Gray shaded area indicates the 95% confidence interval (CI).
Figure 2 with 1 supplement
Least Absolute Shrinkage and Selection Operator (LASSO) and Partial Least Squares (PLS) regression analyses for predicting necrotic core (%NC).
(A) Relationship between regularization coefficients (lambda) and the mean-squared error (MSE) based on the leave-one-out cross-validation in predicting %NC. A dotted vertical line indicates the optimal lambda, which provides the lowest MSE. The optimal lambda was 0.849. (B) LASSO regularization paths along the lambda in predicting %NC. Cyan, magenta, and gray lines indicate the estimated coefficients of autocorrelation (AC)_Mean, AC_Var, and the other input variables, respectively. A dotted vertical line indicates the optimal lambda. (C) Estimated coefficients with the optimal lambda. Only variables with non-zero coefficients are shown. Input variables include the following 21 variables: body mass index (BMI), fasting blood glucose (FBG), HbA1c, PG120, I.I., Composite index, Oral DI, continuous glucose monitoring (CGM)_Mean, CGM_Std, CONGA, LI, JINDEX, HBGI, GRADE, MODD, MAGE, ADRR, MVALUE, MAG, AC_Mean, and AC_Var. (D) Variable Importance in Projection (VIP) generated from the PLS regression predicting %NC. Variables with a VIP ≥ 1 (the dotted line) were considered to significantly contribute to the prediction.
Figure 2—figure supplement 1
Least Absolute Shrinkage and Selection Operator (LASSO) and Partial Least Squares (PLS) regression analyses for predicting necrotic core (%NC) including systolic blood pressure (SBP), diastolic blood pressure (DBP), triglycerides (TGs), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C).
(A) Relationship between regularization coefficients (lambda) and the mean-squared error (MSE) based on the leave-one-out cross-validation in predicting %NC. A dotted vertical line indicates the optimal lambda that provides the least MSE. The optimal lambda was 0.849. (B) LASSO regularization paths along the lambda in predicting %NC. Cyan, magenta, and gray lines indicate the estimated coefficients of autocorrelation (AC)_Mean, AC_Var, and the other input variables, respectively. A dotted vertical line indicates the optimal lambda. (C) Estimated coefficients with the optimal lambda. Only variables with non-zero coefficients are shown. Input variables include the following 26 variables: body mass index (BMI), SBP, DBP, TGs, LDL-C, HDL-C, FBG, HbA1c, PG120, I.I., composite index, oral disposition index (DI), continuous glucose monitoring (CGM)_Mean, CGM_Std, CONGA, LI, JINDEX, HBGI, GRADE, MODD, MAGE, ADRR, MVALUE, MAG, AC_Mean, and AC_Var. (D) Variable Importance in Projection (VIP) generated from the PLS regression predicting %NC. Variables with a VIP ≥ 1 (the dotted line) were considered to significantly contribute to the prediction.
Figure 3 with 4 supplements
Factor analysis of the clinical parameters.
(A) Factor analysis after orthogonal rotation. The values and colors are based on the factor loadings. The columns represent each factor. The rows represent input indices. (B) Cronbach’s α for each factor. Bars represent the 95% confidence interval (CI). (C) Scatter plots and fitted linear regression lines for factor scores versus necrotic core (%NC). Each point corresponds to the values for a single individual. R is Spearman’s correlation coefficient, and the value in parentheses is the 95% CI.
Figure 3—figure supplement 1
Factor analyses of the clinical parameters.
Factor analyses after orthogonal rotation. The values were based on the factor loadings. The columns represent each factor. The rows represent input indices. The analyses with the 21 variables (A) and the 26 variables (B).
Figure 3—figure supplement 2
Hierarchical clustering analysis of metabolic syndrome-related indices.
(A) Relationship between the number of clusters and the silhouette coefficient. A dotted vertical line indicates the optimal number of clusters, which provides the best silhouette coefficient. (B) Hierarchical clustering analysis of the standardized metabolic syndrome-related indices using Euclidean distance as a metric with the Ward method. The columns represent the standardized value of each index. The rows represent individuals. The indices are grouped and sorted according to their degree of relatedness.
Figure 3—figure supplement 3
Factor analyses of the clinical parameters using the previously reported datasets.
Factor analyses after orthogonal rotation. The values were based on the factor loadings. The columns represent each factor. The rows represent input variables. The analyses used the Japanese data from a previous study (Sugimoto et al., 2025) (A), and the American data from a previous study (Hall et al., 2018) (B).
Figure 3—figure supplement 4
Factor analysis of continuous glucose monitoring (CGM)-derived indices using a previously reported dataset.
(A) Factor analysis of the CGM-derived indices. The heat map shows the factor loadings, with columns representing each factor and rows representing the input variables. The analysis used data from 100 Chinese individuals from a previous study (Zhao et al., 2023). (B) Box plots comparing factors 1 (Mean), 2 (Variance), and 3 (Autocorrelation) between individuals without (-) and with (+) diabetic macrovascular complications. Each point corresponds to an individual. The boxes represent the interquartile range, with the median shown as a horizontal line. Mann-Whitney U tests were used to assess differences between groups, with p-values p-values <0.05 considered statistically significant.
Figure 4 with 2 supplements
Overview of the three components of glucose dynamics.
(A) 240 min simulated glucose concentration. The colors of the lines are based on the mean (Mean), Std, and autocorrelation (AC)_Var of the simulated blood glucose. Red and gray dotted horizontal lines indicate the minimum and maximum blood glucose values, respectively. (B) Previously reported patterns of blood glucose during the oral glucose tolerance test (OGTT) (Hulman et al., 2018). Green, class 1; light blue, class 2; dark blue, class 3; red, class 4. (C) Mean, Std, and AC_Var of the glucose during the OGTT. Colors are based on the classes shown in (B).
Figure 4—figure supplement 1
Infused glucose and parameters used in simulating glucose dynamics.
(A) The amount of the external input of glucose (see Methods). (B) The parameters used in the simulation. The values of these parameters shown in red and gray correspond to the color of the simulated glucose dynamics shown in Figure 4A. Bars indicate the range of values for normal glucose tolerance (NGT) individuals (De Gaetano and Arino, 2000).
Figure 4—figure supplement 2
Relation between continuous glucose monitoring (CGM)-derived measures and clinical measures.
(A) Scatter plots and fitted linear regression lines for each CGM-derived measure versus age. Each point corresponds to the values for a single individual. The gray shaded area indicates the 95% confidence interval (CI). (B) Box plots comparing CGM_Mean, CGM_Std, and autocorrelation (AC)_Var for females and males. Each point corresponds to an individual. The boxes represent the interquartile range, with the median shown as a horizontal line. Mann-Whitney U tests were used to assess differences between groups, with p-values p-values <0.05 considered statistically significant. (C) Comparison of predicted Time in range (TIR) versus measured TIR using multiple regression analysis between TIR and factor scores in Figure 3. In this analysis, TIR was the dependent variable, and the factor scores corresponding to the first three latent components (factor 1 representing the mean, factor 2 representing the variance, and factor 3 representing the autocorrelation) were the independent variables. Each point corresponds to the values for a single individual. (D) Bars represent the 95% CIs of the coefficients of the regression model in (C).
Tables
Key resources table
| Reagent type (species) or resource | Designation | Source or reference | Identifiers | Additional information |
|---|---|---|---|---|
| Software, algorithm | SciPy v1.10.1 | Virtanen et al., 2020 | RRID:SCR_008058 | |
| Software, algorithm | scikit-learn v1.0.2 | https://scikit-learn.org/stable/ | RRID:SCR_002577 |
Additional files
-
MDAR checklist
- https://cdn.elifesciences.org/articles/102860/elife-102860-mdarchecklist1-v1.pdf
-
Supplementary file 1
Calculating formulae of the continuous glucose monitoring-derived indices.
- https://cdn.elifesciences.org/articles/102860/elife-102860-supp1-v1.docx
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Association between continuous glucose monitoring-derived metrics and coronary plaque vulnerability: A retrospective exploratory analysis
eLife 14:RP102860.
https://doi.org/10.7554/eLife.102860.3
