The sweet spot: fasting glucose, cardiovascular disease, and mortality in older adults with diabetes: a nationwide population-based study

We used data from the National Health Insurance Service-National Health Screening (NHIS-HEALS) cohort in Korea. The insurance system was established by the Korean government and covers approximately 97% of residents. Subjects aged ≥ 40 years are entitled to a general health screening program every 2 years. The screening includes standardized self-reporting questionnaires on medical history and lifestyle behaviors, anthropometric and blood pressure measurements, and routine laboratory tests using blood and urine. Thus, NHIS-HEALS would be nationally representative cohort database consisting of nearly the whole South Korean population. The cohort profile of the NHIS-HEALS has been described previously [8]. Among individuals who underwent health examination provided by the NHIS at least once between January 1, 2009 and December 31, 2010, we excluded those with no prior diagnosis of diabetes (n = 15,819,273). The occurrence of diabetes mellitus was defined according to the following criteria: (i) at least two claims under International Classification of Diseases 10th Revision (ICD-10) codes or (ii) fasting glucose level ≥ 126 mg/dL. We further excluded subjects with a diagnosis of MI or stroke (n = 71,691) prior to enrollment and those aged ≤ 64 years (n = 1168,473). We also excluded those with any missing data on fasting glucose (n = 1953). To minimize bias caused by the single fasting glucose measurement, we used the mean fasting glucose value obtained from the 2009–2010 and 2011–2012 measurements. Thus, subjects who did not participate in the second health screening in 2011–2012 (n = 250,664) were excluded. The remaining 227,938 eligible individuals (110,123 men and 117,815 women) whose fasting glucose were checked twice at baseline, were included in the analyses and followed up until the date of death or until December 31, 2017. The flow chart for the selection process is shown in Additional file 1. This study was approved by the institutional review board of Yonsei University Wonju College of Medicine (CR318331). The requirement for written informed consent was waived by the review board because anonymous and de-identified information was used for analysis.

Information on current smoking and alcohol consumption was obtained using a questionnaire. Heavy alcohol consumption was defined as ≥ 5 or ≥ 4 alcoholic drinks per week for males and females, respectively. Regular exercise was defined as physical activity that was performed at least five times per week. Blood samples for the measurement of serum fasting glucose and cholesterol levels were drawn after an overnight fast. Hospitals wherein these health examinations were performed were certified by the NHIS and subjected to regular quality control. Baseline comorbidities (hypertension, dyslipidemia, congestive heart failure, chronic obstructive pulmonary disease, depression, chronic kidney disease, and dementia) were identified based on the combination of past medical history, ICD-10 Clinical Modification (ICD-10-CM), and prescription codes. To assess the overall comorbidity load, we used the primary care equivalent of the Charlson comorbidity index (CCI) [9]. The CCI provides a way of quantifying this impact in terms of survival and is also used as a prognostic comorbidity index. CCI was developed empirically 26 years ago as a prognostic index of comorbid conditions for patients admitted for general medical service with a variety of medical conditions. Such conditions, alone or in combination, might alter the risk of short-term mortality in patients enrolled in longitudinal studies.

The endpoints of this study were newly diagnosed CV events (MI or ischemic stroke) and all-cause mortality. MI was defined as ICD-10 codes I21 or I22 during hospitalization or when these codes were recorded at least two times. Ischemic stroke was defined as ICD-10 codes I63 or I64 during hospitalization with claims for brain magnetic resonance imaging or brain computerized tomography [10]. Mortality data were extracted from the death statistics from the Korean National Statistical Office. Subjects without MI or stroke during follow-up periods were considered to have completed the study at the date of their death or at the end of follow-up, whichever came first. The study population was followed up from the date of the second fasting glucose measurement in 2011–2012 to the date of death or CV events, or until December 31, 2017, whichever came first.

Data are expressed as means with standard deviations, n (%), or hazard ratios (HRs) with 95% confidence intervals (CIs). We compared means in two groups using the Student’s t test and compared categorical variables using the Chi-square test. Subjects were classified into ten groups according to the fasting glucose levels at increasing intervals of 15 mg/dL. The risk of overall mortality and CV events (MI or stroke) was calculated using Cox proportional hazards model with fasting glucose level as the dependent variable. In addition, to reduce the impact of competing risk bias on the result, we performed competing risk model analysis to assess the risk of CVD, with death considered as a competing event, using subdistribution hazard model by Fine-Gray [11]. The following covariates: age, sex, income, place of living (urban vs. rural), smoking, duration of diabetes, alcohol, exercise, body mass index, systolic blood pressure, CCI, and cholesterol were included in the multivariable model. The incidence rate of outcomes was calculated by dividing the number of incidences by the total follow-up duration (person-years). A restricted cubic spline transformation of fasting glucose was used to evaluate nonlinear associations. In the subgroup analysis, the presence of interaction was assessed in multivariate models by testing the significance of two-way interaction terms.

Statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA), and a P-value < 0.05 indicated statistical significance.

Overall, 227,938 subjects were included in the analysis. The mean age of the study subjects was 72.5 years (Additional file 2). Of the study subjects, males were 48% (n = 110,123), while in about 67% of patients, the diabetes duration was ≥ 5 years. Hypertension and dyslipidemia prevalence were 74% and 46%, respectively. About 61% of the study population had CCI ≥ 3, while almost all the subjects (94%) were receiving pharmacological treatment for diabetes.

Overall, in this cohort, 27,212 deaths occurred in a median of 5 years (mean 5.7 ± 1.1 years) of the follow-up period. The fasting glucose levels and the risk of all-cause mortality showed a J-shaped relationship, with both the too low and high fasting glucose levels showing a higher risk of mortality (Fig. 1a). The optimal glucose level associated with the lowest risk of all-cause mortality was 122 mg/dL in cubic spline analysis. In categorical analysis, the glucose range of 110–124 mg/dL was associated with the lowest risk of all-cause mortality (Table 1).

Multivariable adjusted HRs (95% CI) of mortality associated with fasting glucose levels < 80, 80–95, 95–110, 125–140, 140–155, 155–170, 170–185, 185–200, and > 200 mg/dL were 1.54 (1.38–1.72), 1.25 (1.19–1.32), 1.08 (1.04–1.13), 1.06 (1.02–1.11), 1.16 (1.11–1.21), 1.25 (1.19–1.31), 1.35 (1.27–1.43), 1.60 (1.49–1.71), and 1.89 (1.79–2.0) respectively, compared with the 110–124-mg/dL level.

During the mean 5.6 ± 1.2 years of follow-up, 17,406 newly diagnosed CV events (6398 MI and 11,863 ischemic stroke) occurred. The relationship between fasting glucose and CV events also showed a J-shaped distribution in study subjects. The fasting glucose level associated with the lowest CV event risk was 116 mg/dL in cubic spline analysis (Fig. 1b). Among the different fasting glucose level categories, those ranging between 95 and 124 mg/dL were associated with the lowest risk of CV events. Multivariable adjusted HRs (95% CI) of CV events associated with fasting glucose levels of < 80, 80–95, 95–110, 125–140, 140–155, 155–170, 170–185, 185–200, and > 200 mg/dL were 1.27 (1.10–1.48), 1.14 (1.06–1.22), 0.99 (0.94–1.05), 1.05 (1.01–1.10), 1.12 (1.06–1.18), 1.25 (1.17–1.32), 1.36 (1.27–1.46), 1.47 (1.35–1.60), and 1.76 (1.64–1.88) respectively, compared with the 110–124 mg/dL level (Table 2). Similar pattern was observed in the competing risk model analysis (Additional file 3).

When the analysis was separately performed in subjects with low (CCI ≤ 2) and high (CCI ≥ 3) comorbidity load, J-shaped relationship between fasting glucose and all-cause mortality was observed in both groups (Fig. 2a). The optimal fasting glucose levels associated with the lowest risk of all-cause mortality were 121 mg/dL in subjects with both CCI ≤ 2 and ≥ 3 in cubic spline analysis. Patients with fasting glucose levels of 110–124 mg/dL showed the lowest risk for all-cause mortality among the different fasting glucose level categories in both comorbidity groups (Table 3). There was no significant difference in the trend of mortality risk between the two groups (P for interaction = 0.12).

When assessing the risk of CV events, similar trends were observed. The optimal fasting glucose levels associated with the lowest risk of CV events were 112 mg/dL and 119 mg/dL in subjects with CCI ≤ 2 and ≥ 3, respectively, in cubic spline analysis (Fig. 2b). Among subjects with CCI ≤ 2, those with fasting glucose level below the optimal value (< 112 mg/dL) showed a gentle slope curve with wide CI compared with those with CCI ≥ 3 who showed a steeper slope of HR curve with less wide CI. When assessing the risk of CV events according to the different categories of fasting glucose, the group with a fasting glucose level of 95–124 mg/dL showed the lowest risk of CV events in both comorbidity groups (Table 4).