Abstract
Context
Previous studies suggest that the complexity of glucose time series may serve as a novel marker of glucose homeostasis.
Objective
We aimed to investigate the relationship between the complexity of glucose time series and all-cause mortality in patients with type 2 diabetes.
Methods
Prospective data of 6000 adult inpatients with type 2 diabetes from a single center were analyzed. The complexity of glucose time series index (CGI) based on continuous glucose monitoring (CGM) was measured at baseline with refined composite multiscale entropy. Participants were stratified by CGI tertiles of: < 2.15, 2.15 to 2.99, and ≥ 3.00. Cox proportional hazards regression models were used to assess the relationship between CGI and all-cause mortality.
Results
During a median follow-up of 9.4 years, 1217 deaths were identified. A significant interaction between glycated hemoglobin A1c (HbA1c) and CGI in relation to all-cause mortality was noted (P for interaction = 0.016). The multivariable-adjusted hazard ratios for all-cause mortality at different CGI levels (≥ 3.00 [reference group], 2.15-2.99, and < 2.15) were 1.00, 0.76 (95% CI, 0.52-1.12), and 1.47 (95% CI, 1.03-2.09) in patients with HbA1c < 7.0%, while the association was nonsignificant in those with HbA1c ≥ 7.0%. The restricted cubic spline regression revealed a nonlinear (P for nonlinearity = 0.041) relationship between CGI and all-cause mortality in subjects with HbA1c < 7.0% only.
Conclusion
Lower CGI is associated with an increased risk of all-cause mortality among patients with type 2 diabetes achieving the HbA1c target. CGI may be a new indicator for the identification of residual risk of death in well-controlled type 2 diabetes.
Keywords: complexity of glucose time series, continuous glucose monitoring (CGM), time series analysis, mortality, glycated hemoglobin A1c (HbA1c), interaction
With the rapid growth of technologies, continuous glucose monitoring (CGM) plays an increasingly important role in glycemic management (1). Compared with conventional glucose monitoring methods, CGM can provide a large amount of glucose data varying with time, which enables more accurate assessment of glycemic fluctuations. Glucose time series data from CGM contains complex and nonlinear information about glycemic fluctuations. However, traditional glycemic variability metrics such as coefficient of variation (CV) and SD only assess the amplitude changes in glycemic fluctuations regardless of the time dimension (2, 3). Furthermore, as these linear methods simplify data from glucose dynamics for the assessment of glycemic fluctuations, they miss abundant nonlinear characteristics of dynamical changes related to glucose regulation, which determine the complexity of glucose time series.
Recently, time series analyses have provided a new approach to evaluating the complexity of physiological time series. Nonlinear methods such as detrended fluctuation analysis (DFA) (4), Poincaré plot (5), and multiscale entropy (MSE) (6) have been developed. Of these, MSE analysis based on entropy is one of the popular time series analyses, which can quantify the complexity of time series data by calculating sample entropy over multiple time scales (6). The glucose homeostasis is maintained by multiple interconnected feedback loops over multiple time scales, involving hormones, diet intake, and muscle activity, among other factors. Therefore, the dynamics of glucose homeostasis system could be regarded as a complex system. In accord with this notion, glucose time series data with higher entropy indicates more irregularity and unpredictability, which means the data is more complex with more diverse patterns of glucose dynamic changes.
Previous studies revealed that patients with diabetes were characterized by a loss of complexity across multiple time scales compared with normal individuals (7-10). More recently, our post hoc analysis of a multicenter CGM study (11) showed that the complexity of glucose time series decreased progressively across the glycemic continuum (from healthy to prediabetes to diabetes) and was significantly associated with measures of insulin sensitivity/secretion, implying that glucose time series complexity may serve as a new marker for assessing glucose homeostasis.
However, the clinical relevance of this metric remains to be determined. Therefore, the current study sought to examine the association between the complexity of glucose time series measured with refined composite multiscale entropy (RCMSE) analysis (12), extended from MSE for shorter time series, and all-cause mortality among patients with type 2 diabetes in a large prospective cohort.
Materials and Methods
Study Population
Data were obtained from the INDices of contInuous Glucose monitoring and adverse Outcomes of diabetes (INDIGO) cohort study (13, 14), an ongoing prospective cohort study recruiting inpatients with type 2 diabetes at the Department of Endocrinology and Metabolism from Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, since January 2005. The present study included inpatients enrolled between January 2005 and December 2015, who met the following criteria: (1) age ≥18 years with the diagnosis of type 2 diabetes; (2) complete CGM data for valid complexity of glucose time series index (CGI) measurement; (3) a citizen of Shanghai, China; and (4) a stable antidiabetic treatment regimen within the past 3 months. Exclusion criteria were other types of diabetes (eg, gestational diabetes or type 1 diabetes), severe and recurrent hypoglycemia within the past 3 months, or admission for diabetic ketoacidosis.
The study protocol was approved by the Ethics Committee of Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine according to the principles of the Helsinki Declaration, and all participants signed informed consent in written form.
CGM and Assessment of CGI
Each subject underwent subcutaneous interstitial glucose monitoring for 72 hours since the first day of admission with a retrospective CGM system (CGMS GOLD, Medtronic Inc., Northridge, CA, USA), which recorded 288 consecutive glucose values daily. CGM measurements were calibrated with at least 4 capillary blood glucose readings per day by a SureStep blood glucose meter (LifeScan, Milpitas, CA, USA). Complete 24-hour data were used to calculate CGM metrics: CGI, CV, mean sensor glucose (MSG), and time in range (TIR). CV was calculated as the measurement of glycemic variability (CV = [SD/MSG] × 100%). TIR was defined as the percentage of time in the target glucose range of 3.9 to 10.0 mmol/L during a 24-hour period. All subjects were instructed to adhere to the original lifestyle and a standard diet during the CGM period, as previously described (15).
Due to the higher precision of RCMSE compared with MSE for time series shorter than 750 points (12), RCMSE was used to quantify the complexity of glucose time series data from CGM in our study. First, the 24-hour series was divided into several equally distributed and nonoverlapping time windows according to the predefined time scale, a process called coarse-graining. The resulting length of the coarse-grained sequence equaled to the original length divided by the time scale, which determined the number of data points in each time window. The time scales in this study were set at 1 to 6, which corresponded to the intervals of 5 to 30 minutes, as the CGM system recorded glucose readings every 5 minutes. For each coarse-grained time series, data points inside segmented windows were averaged to form a new set of time series data. Next, the entropy of each coarse-grained time series was calculated according to the RCMSE algorithm (12). Thus, for each participant, 6 entropy values which corresponded to time scale of 1 to 6 were generated. Finally, the CGI was calculated as the sum of entropy over time scales of 1 to 6. Lower CGI reflects higher regularity and predictability of glycemic fluctuations, and therefore lower complexity of glucose dynamics.
Baseline Measurements
A standardized electronic inpatient medical record data collection form was used to gather information on age, sex, diabetes duration, smoking status (current smoking or not), history of cancer, history of cardiovascular diseases (CVDs; angina, coronary heart disease, or stroke), and medication prescriptions. All subjects received a physical examination, including height, body weight, and blood pressure at admission. Body mass index (BMI) was calculated as weight (kg) divided by height in meters squared (m2). Blood pressure was measured 3 times after 5 minutes of sitting using a standard mercury sphygmomanometer, and the measurements were averaged. After an at least 10-hour overnight fast, venous blood samples were drawn in the next morning after admission. Glycated hemoglobin A1c (HbA1c), total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, and triglycerides were assayed as previously reported (13).
Prospective Follow-up
Information on the cause and time of death in the present study was extracted from the database of the Shanghai Municipal Center for Disease Control and Prevention by personal identification number. Causes of death were identified with the International Classification of Diseases 10th edition (ICD-10) codes. The rate of missing death events in Shanghai was 0.7‰. We used chart review to evaluate the confirmation of death (COD) via the Shanghai adaptation of the Medical Record Audit Form. The evaluation of COD was conducted as previously described in detail (13). All-cause mortality was the major outcome of this study. All subjects were followed up until a death event occurred or until December 31, 2021, whichever occurred first.
Statistical Analysis
RCMSE analysis was performed by MATLAB R2019b (MathWorks Inc., Natick, MA, USA). Statistical analyses were performed by SPSS version 26.0 (SPSS Inc., Chicago, IL, USA) and R version 4.2.0 (RStudio Inc., Boston, MA, USA). All subjects were divided into 3 groups based on the tertiles of baseline CGI (< 2.15, 2.15-2.99, and ≥ 3.00), and CGI ≥ 3.00 was selected as the reference group. The trends across different groups were compared by using the ANOVA (for normally distributed continuous variables), the Jonckheere-Terpstra test (for nonnormally distributed continuous variables), and the Cochran-Armitage trend test (for categorical variables). The correlations among glucose metrics were evaluated by Spearman correlation tests. The survival curves of different groups were compared using the Kaplan–Meier method with a log-rank test. The Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% CIs for all-cause mortality risks across different CGI levels in patients with type 2 diabetes and the restricted cubic spline was used to evaluate the nonlinear association between CGI and all-cause mortality risks. Statistically significant variables with a P value less than 0.15 identified by univariate Cox regression and clinically significant variables were included in the multivariate Cox regression models. CGI was entered into models in 2 ways: as categorical variables and as continuous variables. Three different models were used in the analyses. Model 1 adjusted for age and sex. Model 2 was modified based on Model 1 with the additional adjustment for diabetes duration, BMI, systolic blood pressure, triglycerides, total cholesterol, current smoking status, history of cancer, and history of CVDs. Model 3, the fully adjusted model, added HbA1c and the use of insulin and oral hypoglycemic drugs to the covariates in Model 2. In addition, interactions between CGI and other variables were tested by comparing Cox models with and without interaction terms using a likelihood ratio test. A two-sided P value of <0.05 was considered statistically significant.
Results
Characteristics of Study Individuals
The final analysis comprised 6000 patients with type 2 diabetes in total. During a median follow-up of 9.4 years, 1217 deaths were identified. At baseline, 3272 (54.5%) were men, the mean age was 61.6 years, the mean HbA1c was 8.9%, and the median diabetes duration was 10.0 years. The general characteristics of participants stratified by the tertiles of CGI (< 2.15, 2.15-2.99, and ≥ 3.00) are shown in Table 1. Patients with lower CGI were more likely to be male, had longer diabetes duration, worse glycemic control (higher HbA1c, CV, and MSG with lower TIR), lower BMI and triglycerides, a higher proportion of insulin usage, and a lower proportion of oral hypoglycemic drugs usage (P for trend <0.001) at baseline. Spearman correlation analyses revealed a moderate correlation (rs = −0.468, P < 0.001) between CGI and CV, and a weak correlation of CGI with HbA1c (rs = −0.165, P < 0.001), TIR (rs = 0.198, P < 0.001), and MSG (rs = −0.164, P < 0.001).
Table 1.
Characteristics of participants by tertiles of CGI
| Characteristics | Total | CGI | P for trend | ||
|---|---|---|---|---|---|
| <2.15 | 2.15–2.99 | ≥ 3.00 | |||
| Participants, n | 6000 | 2010 | 1965 | 2025 | |
| Age, years | 61.6 ± 11.9 | 62.2 ± 12.0 | 61.2 ± 11.9 | 61.4 ± 11.8 | 0.032 |
| Men, n (%) | 3272 (54.5) | 1191 (59.3) | 1094 (55.7) | 987 (48.7) | <0.001 |
| BMI, kg/m2 | 24.9 ± 3.5 | 24.6 ± 3.5 | 24.9 ± 3.6 | 25.2 ± 3.5 | <0.001 |
| Diabetes duration, years | 10.0 (4.0, 15.0) | 10.0 (4.0, 15.0) | 10.0 (4.0, 15.0) | 8.0 (3.0, 14.0) | <0.001 |
| Systolic blood pressure, mmHg | 132.9 ± 16.9 | 132.9 ± 17.2 | 132.8 ± 16.7 | 132.9 ± 16.8 | 0.917 |
| Diastolic blood pressure, mmHg | 79.8 ± 9.4 | 79.5 ± 9.4 | 80.1 ± 9.4 | 79.7 ± 9.4 | 0.601 |
| Total cholesterol, mmol/L | 4.7 ± 1.2 | 4.7 ± 1.2 | 4.7 ± 1.2 | 4.8 ± 1.2 | 0.482 |
| Triglycerides, mmol/L | 1.4 (0.9, 2.0) | 1.2 (0.9, 1.9) | 1.4 (1.0, 2.1) | 1.4 (1.0, 2.1) | <0.001 |
| HDL cholesterol, mmol/L | 1.1 ± 0.3 | 1.1 ± 0.3 | 1.1 ± 0.3 | 1.1 ± 0.3 | 0.001 |
| LDL cholesterol, mmol/L | 3.0 ± 1.0 | 3.0 ± 1.0 | 2.9 ± 0.9 | 3.0 ± 0.9 | 0.878 |
| HbA1c, % | 8.9 ± 2.2 | 9.3 ± 2.3 | 8.9 ± 2.2 | 8.5 ± 2.0 | <0.001 |
| CGM metrics | |||||
| ȃCGI | 2.7 ± 1.1 | 1.7 ± 0.3 | 2.6 ± 0.2 | 3.9 ± 0.8 | <0.001 |
| ȃCV, % | 26.5 ± 9.5 | 31.7 ± 9.7 | 26.2 ± 8.4 | 21.7 ± 7.7 | <0.001 |
| ȃTIR, % | 64.7 ± 24.2 | 60.2 ± 21.9 | 64.8 ± 24.1 | 69.2 ± 25.5 | <0.001 |
| ȃMSG, mmol/L | 9.2 ± 1.9 | 9.5 ± 1.9 | 9.2 ± 1.9 | 8.8 ± 1.9 | <0.001 |
| History of CVDs, n (%) | 1273 (21.2) | 440 (21.9) | 400 (20.4) | 433 (21.4) | 0.695 |
| History of cancer, n (%) | 273 (4.6) | 99 (4.9) | 81 (4.1) | 93 (4.6) | 0.614 |
| Current smoker, n (%) | 1431 (23.8) | 498 (24.8) | 494 (25.1) | 439 (21.7) | 0.021 |
| Medication, n (%) | |||||
| ȃInsulin | 4014 (66.9) | 1514 (75.3) | 1343 (68.3) | 1157 (57.1) | <0.001 |
| ȃOral hypoglycemic drugs | 4238 (70.6) | 1312 (65.3) | 1393 (70.9) | 1533 (75.7) | <0.001 |
| ȃAntihypertensive drugs | 3266 (54.4) | 1098 (54.6) | 1028 (52.3) | 1140 (56.3) | 0.285 |
| ȃAspirin | 2837 (47.3) | 966 (48.1) | 911 (46.4) | 960 (47.4) | 0.680 |
| ȃStatins | 2327 (38.8) | 775 (38.6) | 748 (38.1) | 804 (39.7) | 0.454 |
Data shown are mean ± SD, median (interquartile range) or number (percentage) unless otherwise indicated.
Abbreviations: BMI, body mass index; CGI, complexity of glucose time series index; CGM, continuous glucose monitoring; CV, coefficient of variation; CVDs, cardiovascular diseases; HbA1c, glycated hemoglobin A1c; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MSG, mean sensor glucose; TIR, time in range.
CGI and All-Cause Mortality
The Kaplan–Meier survival curves indicated that the lowest tertile of CGI was associated with a worse survival rate (log-rank P < 0.001, Fig. 1). After multivariable adjustment (age, sex, diabetes duration, BMI, systolic blood pressure, triglycerides, total cholesterol, current smoking status, history of cancer, and history of CVDs: Model 2), the risk of all-cause mortality was increased by 19% (95% CI, 1.04-1.36) in patients with the lowest tertile of CGI, compared with the highest tertile of CGI (Table 2). The significance was attenuated after further adjustment for potential confounders, including HbA1c (Model 3).
Figure 1.
Cumulative survival curves of all-cause mortality among patients with type 2 diabetes. Abbreviation: CGI, complexity of glucose time series index.
Table 2.
HRs of all-cause mortality across tertiles of CGI among patients with type 2 diabetes
| All-cause mortality | CGI | ||
|---|---|---|---|
| ≥ 3.00 | 2.15–2.99 | < 2.15 | |
| Participants/deaths, n | 2025/371 | 1965/374 | 2010/472 |
| Person-years | 19524.3 | 18749.7 | 18543.5 |
| Adjusted HRs (95% CIs) | |||
| ȃModel 1 | 1.00 | 1.02 (0.88–1.17) | 1.21 (1.06–1.39)** |
| ȃModel 2 | 1.00 | 1.00 (0.87–1.16) | 1.19 (1.04–1.36)* |
| ȃModel 3 | 1.00 | 0.92 (0.80–1.06) | 1.03 (0.89–1.18) |
Model 1 adjusted for age and sex; Model 2 further adjusted for diabetes duration, BMI, systolic blood pressure, triglycerides, total cholesterol, current smoking status, history of cancer, and history of CVDs; Model 3 adjusted for the covariates in Model 2 plus HbA1c, use of insulin, and oral hypoglycemic drugs; *P < 0.05, **P < 0.01.
Abbreviations: BMI, body mass index; CGI, complexity of glucose time series index; CVDs, cardiovascular diseases; HbA1c, glycated hemoglobin A1c; HR, hazard ratio.
Heterogeneity in the Association of CGI With All-Cause Mortality Across Selected Factors
Subgroup analysis was performed according to the baseline characteristics. A significant interaction between CGI and HbA1c with the risk of all-cause mortality was observed in patients with type 2 diabetes (P for interaction = 0.016). Meanwhile, there were no significant interactions between CGI and age, sex, diabetes duration, follow-up period, current smoking status, history of CVDs, history of cancer, use of insulin, and use of oral hypoglycemic drugs in relation to the risk of all-cause mortality (all P for interaction > 0.05) (Table 3).
Table 3.
HRs for all-cause mortality according to tertiles of CGI among subpopulations
| Subpopulations | CGI | P for interaction | ||
|---|---|---|---|---|
| ≥ 3.00 | 2.15–2.99 | < 2.15 | ||
| Age, years | >0.1 | |||
| ȃ<60.0 | 1.00 | 0.79 (0.54–1.15) | 1.11 (0.78–1.58) | |
| ȃ≥60.0 | 1.00 | 0.97 (0.83–1.14) | 1.05 (0.90–1.23) | |
| Sex | >0.1 | |||
| ȃMen | 1.00 | 1.03 (0.85–1.25) | 1.09 (0.90–1.32) | |
| ȃWomen | 1.00 | 0.78 (0.62–0.97)* | 0.94 (0.76–1.16) | |
| Diabetes duration, years | 0.058 | |||
| ȃ<10.0 | 1.00 | 0.77 (0.61–0.98)* | 1.02 (0.81–1.27) | |
| ȃ≥10.0 | 1.00 | 1.03 (0.85–1.23) | 1.04 (0.87–1.25) | |
| Follow-up period, years | >0.1 | |||
| ȃ<5.0 | 1.00 | 0.88 (0.69–1.13) | 1.01 (0.81–1.27) | |
| ȃ5.0–10.0 | 1.00 | 0.89 (0.72–1.08) | 0.97 (0.80–1.19) | |
| ȃ≥10.0 | 1.00 | 1.45 (0.94–2.24) | 1.12 (0.71–1.77) | |
| HbA1c, % | 0.016 | |||
| ȃ<7.0 | 1.00 | 0.76 (0.52–1.12) | 1.48 (1.04–2.11)* | |
| ȃ≥7.0 | 1.00 | 0.96 (0.82–1.12) | 1.00 (0.86–1.17) | |
| Current smokers | >0.1 | |||
| ȃYes | 1.00 | 0.98 (0.71–1.35) | 1.05 (0.77–1.45) | |
| ȃNo | 1.00 | 0.90 (0.76–1.05) | 1.01 (0.86–1.18) | |
| History of CVDs | >0.1 | |||
| ȃYes | 1.00 | 0.94 (0.74–1.20) | 0.89 (0.70–1.14) | |
| ȃNo | 1.00 | 0.91 (0.76–1.09) | 1.10 (0.92–1.30) | |
| History of cancer | >0.1 | |||
| ȃYes | 1.00 | 0.68 (0.37–1.27) | 0.91 (0.52–1.62) | |
| ȃNo | 1.00 | 0.94 (0.81–1.09) | 1.03 (0.89–1.20) | |
| Use of insulin | 0.093 | |||
| ȃYes | 1.00 | 0.88 (0.75–1.05) | 0.93 (0.79–1.09) | |
| ȃNo | 1.00 | 0.99 (0.73–1.33) | 1.50 (1.13–2.00)** | |
| Use of oral hypoglycemic drugs | >0.1 | |||
| ȃYes | 1.00 | 0.99 (0.83–1.18) | 1.09 (0.92–1.29) | |
| ȃNo | 1.00 | 0.84 (0.65–1.09) | 0.98 (0.77–1.24) | |
Adjusted for age, sex, diabetes duration, BMI, systolic blood pressure, triglycerides, total cholesterol, current smoking status, history of cancer, history of CVDs, HbA1c, use of insulin, and oral hypoglycemic drugs, other than the variable for stratification; *P < 0.05, **P < 0.01.
Abbreviations: BMI, body mass index; CGI, complexity of glucose time series index; CVDs, cardiovascular diseases; HbA1c, glycated hemoglobin A1c; HR, hazard ratio.
Among patients with HbA1c < 7.0%, the fully adjusted HRs (Model 3) associated with different CGI levels (≥ 3.00 [reference group], 2.15-2.99, and < 2.15) were 1.00, 0.76 (95% CI, 0.52-1.12), and 1.47 (95% CI, 1.03-2.09), respectively, for the risk of all-cause mortality (Table 4), while the link between CGI tertiles and all-cause mortality was nonsignificant in subjects with HbA1c ≥ 7%. The restricted cubic spline analysis nested in the fully adjusted Cox regression model (Model 3) revealed similar results (Fig. 2). A nonlinear relationship (P for nonlinearity = 0.041) between CGI and all-cause mortality was observed only in participants with HbA1c < 7%, with CGI < 3.00 associated with a higher risk of death.
Table 4.
Hazard ratios for all-cause mortality according to tertiles of CGI among patients with type 2 diabetes across HbA1c categories
| All-cause mortality | CGI | ||
|---|---|---|---|
| ≥ 3.00 | 2.15–2.99 | < 2.15 | |
| HbA1c < 7.0% | |||
| Participants/deaths, n | 540/69 | 407/43 | 313/64 |
| Person-years | 5355.1 | 4029.8 | 2881.5 |
| Adjusted HRs (95% CIs) | |||
| ȃModel 1 | 1.00 | 0.79 (0.54–1.16) | 1.51 (1.07–2.13)* |
| ȃModel 2 | 1.00 | 0.79 (0.54–1.16) | 1.54 (1.09–2.19)* |
| ȃModel 3 | 1.00 | 0.76 (0.52–1.12) | 1.47 (1.03–2.09)* |
| HbA1c ≥ 7.0% | |||
| Participants/deaths, n | 1485/302 | 1558/331 | 1697/408 |
| Person-years | 14169.2 | 14719.9 | 15662.0 |
| Adjusted HRs (95% CIs) | |||
| ȃModel 1 | 1.00 | 1.03 (0.88–1.20) | 1.11 (0.96–1.29) |
| ȃModel 2 | 1.00 | 1.01 (0.86–1.18) | 1.08 (0.93–1.25) |
| ȃModel 3 | 1.00 | 0.94 (0.80–1.10) | 0.96 (0.82–1.12) |
Model 1 adjusted for age and sex; Model 2 further adjusted for diabetes duration, BMI, systolic blood pressure, triglycerides, total cholesterol, current smoking status, history of cancer, and history of CVDs; Model 3 adjusted for the covariates in Model 2 plus HbA1c, use of insulin, and oral hypoglycemic drugs; *P < 0.05.
Abbreviations: BMI, body mass index; CGI, complexity of glucose time series index; CVDs, cardiovascular diseases; HbA1c, glycated hemoglobin A1c; HR, hazard ratio.
Figure 2.
HRs of all-cause mortality by different levels of CGI among patients with HbA1c < 7.0% (A) and HbA1c ≥ 7.0% (B). CGI of 3.00 was set as the reference. Shading indicates 95% CI. Adjusted variables in the model included age, sex, diabetes duration, BMI, systolic blood pressure, triglycerides, total cholesterol, current smoking status, history of cancer, history of CVDs, HbA1c, use of insulin, and oral hypoglycemic drugs. Abbreviations: BMI, body mass index; CGI, complexity of glucose time series index; CVDs, cardiovascular diseases; HbA1c, glycated hemoglobin A1c; HR, hazard ratio.
Discussion
The large prospective cohort study was the first to evaluate the association between glucose times series complexity and diabetes-related adverse outcomes in type 2 diabetes. Although the relationship between CGI and all-cause mortality did not reach statistical significance in the total population, we observed that CGI continued to be the significant predictor of all-cause mortality in patient with HbA1c < 7%, suggesting that CGI may be a novel marker of glucose homeostasis at an early stage of diabetes.
Glucose homeostasis is a relatively stable state in physiological states controlled by multiple complex interaction mechanisms over multiple time scales (16), which determines the complexity of glycemic fluctuations. Endogenous glucose regulation enables normal individuals to maintain glucose homeostasis even under external stressors (eg, diet and exercise). However, in patients with diabetes, their adaptability to external stressors weakens as their ability of physiological glucose regulation declines progressively, which leads to lower complexity and higher variability of glucose dynamics (17). Thus, complexity is a feature of glycemic fluctuations next to variability; complexity is mainly affected by the intrinsic ability of glucose regulation such as β-cell function, while variability is mainly affected by external stressors (18). Due to the nonlinearity, irregularity, and unpredictability of complex glucose dynamic changes, complexity cannot be measured by simple linear statistical methods (17).
Several previous studies showed that the complexity of glucose time series was significantly decreased in both type 1 (8, 9, 19) and type 2 diabetes (7-9, 19, 20) compared with healthy controls, and that the underlying mechanism might be related to β-cell dysfunction (10). To extend these findings, our recent study (11) included 333 matched individuals (normal glucose regulation, impaired glucose regulation, and newly diagnosed type 2 diabetes) and proved that CGI progressively declined across the glucose continuum. Furthermore, CGI was closely correlated with indices of insulin sensitivity/secretion. In aggregate, these findings suggest that CGI may be a new indicator of glucose homeostasis. However, the evidence linking the complexity of glucose time series with adverse outcomes is still very limited. Two previous studies in critically ill patients with a relatively small sample size found that decreased glucose time series complexity measured by DFA was significantly associated with increased mortality (21, 22). To the best of our knowledge, no study has explored the relationship between the complexity of glucose time series and adverse outcomes among patients with type 2 diabetes by RCMSE analysis.
The main finding of our study was the significant association of CGI with all-cause mortality in patients with seemingly well-controlled diabetes (ie, HbA1c < 7%). To date, HbA1c is generally regarded as the gold standard for assessing glucose control, and not achieving the HbA1c target of < 7.0% was reported to be one of the strongest predictors of excess mortality among patients with type 2 diabetes as compared with the general population (23). Moreover, even those with type 2 diabetes achieving the HbA1c target had almost doubled risk of death compared with the general population (24). Therefore, the results of the study imply that the evaluation of CGI may help in identifying the residual risk of all-cause mortality in patients with well-controlled diabetes and targeting patients who could be candidates for intensified management. Additionally, given the differential association of CGI with all-cause mortality in subjects with HbA1c < 7% and ≥ 7%, it is reasonable to postulate that CGI may be a more useful marker of glucose homeostasis at the early stage of diabetes, or even before the onset of diabetes. In this regard, a previous longitudinal study in 206 individuals with a high risk of diabetes found that glucose time series complexity assessed by DFA was a significant predictor of the development of type 2 diabetes (25). However, glucose tolerance testing was not available in that study, raising the possibility that the presence of prediabetes at baseline may confound the result, and the predictive value of glucose time series complexity for the onset of diabetes needs to be further investigated by well-designed prospective studies.
One potential reason that could explain the significant relationship between CGI and all-cause mortality in patients with HbA1c < 7% was that, in this subset of study sample, higher CGI was found to be significantly correlated with more favorable glycemic metrics including higher TIR, and lower MSG and CV (all P < 0.05, data not shown). This observation was somewhat expected, as glucose complexity measures have been demonstrated to be related to β-cell function (10). During recent years, long-term glucose variability, especially HbA1c variability, has gained increased awareness due to its link to diabetic complications (26-28), and unhealthier lifestyles were reported to be a significant determinant of greater HbA1c variability (29). Since CGI reflects the integrity of the regulatory mechanisms of glucose homeostasis, patients with higher CGI may presumably adapt more effectively to external stimuli or stressors such as diet and physical activity, and therefore present with more stable HbA1c over time, which may be an interesting issue to address in the future.
The large sample size of type 2 diabetes with available CGM data and long-term follow-up supported the reliability of our results. There are also several limitations that need to be noted. First, all subjects only underwent CGM for 3 consecutive days, while the international consensus (30) recommended that more than 70% of valid CGM data from 14 days are needed for reliable and accurate interpretation of CGM metrics. Therefore, longer CGM data collections are needed to verify the result. Second, all subjects included in this study were inpatients with type 2 diabetes receiving a standard diet during CGM. Therefore, it is uncertain whether these results could be generalized to other populations with diabetes. In addition, lifestyle and socioeconomic factors were not available in this study, which could affect the risk of all-cause mortality. In this context, residual confounding could not be excluded. Furthermore, the associations between CGI and risks of mortality for different causes of death were not evaluated. Finally, we only analyzed the impact of baseline CGI on the risk of all-cause mortality, thus the association between variations in CGI during follow-up and mortality risk remains to be determined.
In conclusion, the decrease of CGI was associated with an increased risk of all-cause mortality among patients with type 2 diabetes achieving the HbA1c target, suggesting that CGI may help identify the residual risk of death in patients with seemingly well-controlled diabetes, a finding that warrants further investigation in the future.
Acknowledgments
The authors thank all the involved clinicians, nurses, and technicians of the Shanghai Clinical Center for Diabetes for the completion of this study, as well as all the patients who participated in this study.
Abbreviations
- BMI
body mass index
- CGI
complexity of glucose time series index
- CGM
continuous glucose monitoring
- COD
confirmation of death
- CV
coefficient of variation
- CVD
cardiovascular disease
- DFA
detrended fluctuation analysis
- HbA1c
glycated hemoglobin A1c
- HDL
high-density lipoprotein
- HR
hazard ratio
- ICD-10
International Classification of Diseases 10th edition
- MSE
multiscale entropy
- MSG
mean sensor glucose
- LDL
low-density lipoprotein
- RCMSE
refined composite multiscale entropy
- TIR
time in range
Contributor Information
Jinghao Cai, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Qing Yang, Vital Statistical Department, Institute of Health Information, Shanghai Municipal Center for Disease Control and Prevention, Shanghai 200336, China.
Jingyi Lu, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Yun Shen, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Chunfang Wang, Vital Statistical Department, Institute of Health Information, Shanghai Municipal Center for Disease Control and Prevention, Shanghai 200336, China.
Lei Chen, Vital Statistical Department, Institute of Health Information, Shanghai Municipal Center for Disease Control and Prevention, Shanghai 200336, China.
Lei Zhang, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Wei Lu, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Wei Zhu, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Tian Xia, Vital Statistical Department, Institute of Health Information, Shanghai Municipal Center for Disease Control and Prevention, Shanghai 200336, China.
Jian Zhou, Department of Endocrinology and Metabolism, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai Clinical Center for Diabetes, Shanghai Diabetes Institute, Shanghai Key Laboratory of Diabetes Mellitus, Shanghai 200233, China.
Funding
This work was funded by the Program of Shanghai Academic Research Leader (22XD1402300), the Shanghai “Rising Stars of Medical Talent” Youth Development Program–Outstanding Youth Medical Talents, the National Natural Science Foundation of China (31971485), the Shanghai Municipal Project for Academic Leaders Public Health (GWV-10.2-XD20), and Shanghai Municipal Key Clinical Specialty.
Author Contributions
J.Z. and T.X. conceived and designed the study. J.C., Q.Y., and J.L. contributed to data collection, data analysis, and manuscript preparation. Y.S., C.W., L.C., and L.Z. contributed to data collection and analysis. W.L. and W.Z. contributed to the conduction of the study and data collection. T.X. and J.Z. contributed to the interpretation of the data and revision of the manuscript. J.Z. and T.X. are the guarantors of this work, have full access to all of the data in the study, and take responsibility for the integrity of the data and the accuracy of the data analysis. All authors read and approved the final manuscript.
Disclosure
The authors declare no conflicts of interest relevant to this article.
Data Availability
Restrictions apply to the availability of some or all data generated or analyzed during this study to preserve patient confidentiality or because they were used under license. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.
Prior Presentation
Parts of this study were orally presented at the 82nd Scientific Sessions of the American Diabetes Association, New Orleans, LA, 3-7 June 2022 (abstract in Diabetes 2022; 71 (Supplement_1): 91-OR).
References
- 1. Draznin B, Aroda VR, Bakris G, et al. 6. Glycemic targets: standards of medical care in diabetes-2022. Diabetes Care. 2022;45(Supplement_1):S83‐S96. [DOI] [PubMed] [Google Scholar]
- 2. Cobelli C, Facchinetti A. Yet another glucose variability index: time for a paradigm change? Diabetes Technol Ther. 2018;20(1):1‐3. [DOI] [PubMed] [Google Scholar]
- 3. Danne T, Nimri R, Battelino T, et al. International consensus on use of continuous glucose monitoring. Diabetes Care. 2017;40(12):1631‐1640. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Yamamoto N, Kubo Y, Ishizawa K, et al. Detrended fluctuation analysis is considered to be useful as a new indicator for short-term glucose complexity. Diabetes Technol Ther. 2010;12(10):775‐783. [DOI] [PubMed] [Google Scholar]
- 5. Crenier L. Poincaré plot quantification for assessing glucose variability from continuous glucose monitoring systems and a new risk marker for hypoglycemia: application to type 1 diabetes patients switching to continuous subcutaneous insulin infusion. Diabetes Technol Ther. 2014;16(4):247‐254. [DOI] [PubMed] [Google Scholar]
- 6. Costa M, Goldberger AL, Peng CK. Multiscale entropy analysis of complex physiologic time series. Phys Rev Lett. 2002;89(6):068102. [DOI] [PubMed] [Google Scholar]
- 7. Costa MD, Henriques T, Munshi MN, Segal AR, Goldberger AL. Dynamical glucometry: use of multiscale entropy analysis in diabetes. Chaos. 2014;24(3):033139. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Chen J, Chen P, Wang H. Decreased complexity of glucose dynamics in diabetes: evidence from multiscale entropy analysis of continuous glucose monitoring system data. Am J Physiol Regul Integr Comp Physiol. 2014;307(2):R179‐R183. [DOI] [PubMed] [Google Scholar]
- 9. Kohnert KD, Heinke P, Vogt L, Augstein P, Thomas A, Salzsieder E. Associations of blood glucose dynamics with antihyperglycemic treatment and glycemic variability in type 1 and type 2 diabetes. J Endocrinol Invest. 2017;40(11):1201‐1207. [DOI] [PubMed] [Google Scholar]
- 10. Kohnert KD, Heinke P, Vogt L, Augstein P, Salzsieder E. Applications of variability analysis techniques for continuous glucose monitoring derived time series in diabetic patients. Front Physiol. 2018; 9:1257. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Li C, Ma X, Lu J, et al. Decreasing complexity of glucose time series derived from continuous glucose monitoring is correlated with deteriorating glucose regulation. Front Med. Accepted 20 July2022. 10.1007/s11684-022-0955-9 [DOI] [PubMed] [Google Scholar]
- 12. Wu S, Wu C, Lin S, Lee K, Peng C. Analysis of complex time series using refined composite multiscale entropy. Physics Letters A. 2014;378(20):1369‐1374. [Google Scholar]
- 13. Lu J, Wang C, Shen Y, et al. Time in range in relation to all-cause and cardiovascular mortality in patients with type 2 diabetes: a prospective cohort study. Diabetes Care. 2021;44(2):549‐555. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Lu J, Wang C, Cai J, et al. Association of HbA1c with all-cause mortality across varying degrees of glycemic variability in type 2 diabetes. J Clin Endocrinol Metab. 2021;106(11):3160‐3167. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Lu J, Ma X, Zhou J, et al. Association of time in range, as assessed by continuous glucose monitoring, with diabetic retinopathy in type 2 diabetes. Diabetes Care. 2018;41(11):2370‐2376. [DOI] [PubMed] [Google Scholar]
- 16. Röder PV, Wu B, Liu Y, Han W. Pancreatic regulation of glucose homeostasis. Exp Mol Med. 2016;48(3):e219. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Lipsitz LA, Goldberger AL. Loss of ‘complexity’ and aging. Potential applications of fractals and chaos theory to senescence. JAMA. 1992;267(13):1806‐1809. [PubMed] [Google Scholar]
- 18. van Hooijdonk RTM, Abu-Hanna A, Schultz MJ. Glycemic variability is complex–is glucose complexity variable? Critical Care. 2012;16(6):178. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Ogata H, Tokuyama K, Nagasaka S, et al. The lack of long-range negative correlations in glucose dynamics is associated with worse glucose control in patients with diabetes mellitus. Metab Clin Exp. 2012;61(7):1041‐1050. [DOI] [PubMed] [Google Scholar]
- 20. Churruca J, Vigil L, Luna E, Ruiz-Galiana J, Varela M. The route to diabetes: loss of complexity in the glycemic profile from health through the metabolic syndrome to type 2 diabetes. Diabetes Metab Syndr Obes. 2008;1:3‐11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Lundelin K, Vigil L, Bua S, Gomez-Mestre I, Honrubia T, Varela M. Differences in complexity of glycemic profile in survivors and nonsurvivors in an intensive care unit: a pilot study. Crit Care Med. 2010;38(3):849‐854. [DOI] [PubMed] [Google Scholar]
- 22. Brunner R, Adelsmayr G, Herkner H, Madl C, Holzinger U. Glycemic variability and glucose complexity in critically ill patients: a retrospective analysis of continuous glucose monitoring data. Critical Care. 2012;16(5):R175. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Rawshani A, Rawshani A, Franzén S, et al. Risk factors, mortality, and cardiovascular outcomes in patients with type 2 diabetes. N Engl J Med. 2018;379(7):633‐644. [DOI] [PubMed] [Google Scholar]
- 24. Tancredi M, Rosengren A, Svensson A-M, et al. Excess mortality among persons with type 2 diabetes. N Engl J Med. 2015;373(18):1720‐1732. [DOI] [PubMed] [Google Scholar]
- 25. de Castro C R, Vigil L, Vargas B, et al. Glucose time series complexity as a predictor of type 2 diabetes. Diabetes Metab Res Rev. 2017;33(2):e2831. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Wu T, Su Y, Chen H. Mean HbA1c and HbA1c variability are associated with differing diabetes-related complications in patients with type 2 diabetes mellitus. Diabetes Res Clin Pract. 2022;192:110069. [DOI] [PubMed] [Google Scholar]
- 27. Virk SA, Donaghue KC, Cho YH, et al. Association between HbA1c variability and risk of microvascular complications in adolescents with type 1 diabetes. J Clin Endocrinol Metab. 2016;101(9):3257‐3263. [DOI] [PubMed] [Google Scholar]
- 28. Su J, Zhao L, Zhang X, et al. Hba1c variability and diabetic peripheral neuropathy in type 2 diabetic patients. Cardiovasc Diabetol. 2018;17(1):47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Hsu CC, Chang HY, Huang MC, et al. Hba1c variability is associated with microalbuminuria development in type 2 diabetes: a 7-year prospective cohort study. Diabetologia. 2012;55(12):3163‐3172. [DOI] [PubMed] [Google Scholar]
- 30. Battelino T, Danne T, Bergenstal RM, et al. Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range. Diabetes Care. 2019;42(8):1593‐1603. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Restrictions apply to the availability of some or all data generated or analyzed during this study to preserve patient confidentiality or because they were used under license. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.
