Glycated haemoglobin and the risk of cardiovascular disease, diabetes and all-cause mortality in the Copenhagen City Heart Study

Authors


Abstract

Objective

Individuals with diabetes mellitus (DM) have a considerably elevated risk of developing serious health problems including cardiovascular disease (CVD). Long-term elevated levels of blood glucose in nondiabetic individuals may also be associated with increased risk of CVD. The aim of this study was to investigate the relationships between glycated haemoglobin A1c (HbA1c) and CVD, DM and all-cause mortality.

Subjects and design

The Copenhagen City Heart Study is a prospective study of individuals from the Danish general population. The cohort was followed for 10 years via national registers with respect to incident CVD, DM and all-cause mortality. Follow-up was 100% complete.

Results

A total of 5127 subjects were included, of whom 597 had DM. In the nondiabetic population, HbA1c was significantly associated with incident CVD events in both univariate [hazard ratio (HR) 1.38, 95% CI 1.11–1.71] and multivariate analyses (HR 1.31, 95% CI 1.05–1.64). In the nondiabetic population, increased levels of HbA1c were correlated with developing DM. There was a threefold increase in risk of incident DM per unit increase in HbA1c with a univariate HR of 3.83 (95% CI 1.96–7.51). This relationship was essentially unchanged after multivariate adjustments (HR 4.19, 95% CI 2.01–8.71). Furthermore, we found that net reclassification improvement for diagnosed DM and CVD was significantly improved with the addition of HbA1c in the analyses. Although not statistically significant, we found a strong trend towards an association between HbA1c and all-cause mortality (HR 1.21, 95% CI 0.99–1.47). We did not find the same associations amongst the population with DM.

Conclusion

In the Danish general population, HbA1c was strongly associated with CVD in individuals without DM.

Introduction

Individuals with diabetes mellitus (DM) are at increased risk of developing numerous serious health problems including cardiovascular disease (CVD) [1]. It is has been proposed that high levels of glycated haemoglobin A1c (HbA1c) are an independent risk factor for CVD in both diabetic and nondiabetic individuals. The relationship between HbA1c and CVD is somewhat complex, and the predictive value of HbA1c is uncertain.

HbA1c reflects the weighted mean plasma glucose concentration during the preceding months [2] and is relatively insensitive to short-term lifestyle changes [3]. There is a log-linear correlation between HbA1c and microvascular complications [4, 5], and HbA1c is thus used ubiquitously as a tool for monitoring glycaemic control and quality of care in patients with diabetes worldwide. Because of this, as well as the completion of the National Glucohemoglobin Standardization Program [6], the World Health Organization (WHO) recently recommended that HbA1c should be used as a diagnostic test for DM [7]. In agreement with this, the American Diabetes Association 2011 guidelines [8] state that an HbA1c level ≥ 6.5% is one of four diagnostic criteria for DM.

In diabetic individuals, the relationship between HbA1c and macrovascular disease is unclear. Amongst nondiabetic individuals, the role of HbA1c is also uncertain. In 2010, Selvin et al. [9] showed an association between HbA1c and CVD in a community-based study of nondiabetic adults in a heterogeneous American population. Other authors have shown similar results, however inconsistently [10, 11].

With new recommendations for the use of HbA1c and equivocal results from clinical studies, it is necessary to clarify the role of HbA1c within a representative population. The Copenhagen City Heart Study is a large-scale, prospective, population-based study of CVD with uniform sample methods, knowledge of medication and comorbidities and complete follow-up.

Using data from the Copenhagen City Heart Study, our aim was to examine the relationship between baseline HbA1c levels and the risk of developing DM, incident CVD and all-cause death in a homogeneous Western European population of diabetic and nondiabetic individuals.

Materials and methods

The Copenhagen City Heart Study is a prospective study comprising a random sample from the Danish general population. It was initiated in 1976, with follow-up examinations in 1981–1983, 1991–1994 and 2001–2003. Overall, 6237 subjects who participated in the final examination in 2001–2003 were included in the present analysis [12].

All subjects underwent a full physical examination, and a self-administered questionnaire provided information regarding medical history, smoking and drinking habits, leisure time physical activity, medication and history of contact with the healthcare system. Blood pressure was measured in a standardized manner using the London School of Hygiene sphygmomanometer. Plasma glucose and cholesterol levels were measured in nonfasting venous blood samples. A 12-lead ECG was recorded at rest in a supine position, and findings were defined according to the Minnesota Code. HbA1c measurement was based on a turbidimetric inhibition immunoassay (Thermo Fisher Scientific, Vantaa, Finland) for haemolysed whole blood collected at the first visit. The HbA1c method was standardized against the approved International Federation of Clinical Chemistry and Laboratory Medicine reference method.

The diabetic population was defined as subjects with HbA1c ≥ 6.5% (thus the upper limit of the normal range was <6.5%) or with nonfasting plasma glucose ≥ 11.1 mmol L−1, or reported use of insulin or other antidiabetic medication.

Subjects were excluded from the analyses if they had missing values of HbA1c (= 290), missing information about blood glucose level or antidiabetic medication (= 49), reported or had a diagnosis of previous coronary heart disease, ischaemic or haemorrhagic stroke, transient ischaemic attack or electrocardiographic evidence of ischaemic heart disease (Minnesota Codes 1-1 and 1-2) (= 771).

Patients were followed until 2011, an event, or censoring (whichever was first) through central registers of the Danish National Board of Health; follow-up was 100% complete. Vital status (dead or alive) was obtained from the national Danish Civil Personal Register system and cause of death from the national Danish Register of Causes of Death. Admission data were extracted from the Danish National Patient Registry and specified according to diagnostic codes of the International Classification of Diseases, 8th and 10th revisions (ICD-8 and ICD-10, respectively). Ischaemic heart disease was defined as ICD-8 410–414 or IDC-10 I20–I25, ischaemic stroke as ICD-8 430–438 or ICD-10 I60–I68 and G45, and incident DM as ICD-10 E11 and E13–E14.

The study was conducted in accordance with the Declaration of Helsinki and was approved by the regional ethics committee.

Statistical analysis

Fisher's exact test was used for categorical variables and anova for continuous variables. The associations between HbA1c, all-cause mortality and CVD events were analysed for the populations with and without DM separately. Multivariate Cox proportional hazards were calculated adjusted for age, sex, body mass index (BMI), systolic blood pressure, alcohol consumption (never, monthly, weekly and everyday drinker), smoking (never, former, moderate and heavy smoker), leisure time physical activity (sedentary, light activity < 2 h per week; moderate, light activity 2–4 h per week; and high, light activity > 4 h per week or high-level activity > 2 h per week) and cholesterol.

The association between HbA1c and incident DM was also analysed in the nondiabetic population using univariate and multivariate Cox models. The assumption of proportionality in the Cox regression models was tested with the score process test. To evaluate the improvement in risk prediction when adding HbA1c to a model including traditional risk factors, we calculated C-statistics and the net reclassification improvement (NRI) index. The NRI index determines the percentage of subjects correctly reclassified when including HbA1c to the model, that is, subjects with an event should be reclassified as at higher risk and subjects without an event at lower risk. P-values below 0.05 were considered statistically significant.

All statistical analyses were carried out using the statistical software R, version 2.13.1 (R Foundation for Statistical Computing, Vienna, Austria).

Results

A total of 5127 subjects were included in the study: 597 with and 4530 without DM at inclusion. During follow-up of up to 10 years, there were 732 deaths, 592 CVD events and 61 cases of incident DM. Demographic characteristics of the participants are shown in Table 1.

Table 1. Baseline demographic characteristics by quartile of HbA1c in subjects with and without DM at inclusion
 HbA1c (%)HbA1c (%)HbA1c (%)HbA1c (%)AllP-value
4.8 < 6.7 (= 211)6.7–6.9 (= 114)6.9–7.4 (= 133)7.4–13.8 (= 139)= 597
  1. SD, standard deviation.

Subjects with DM at inclusion
Age, mean (SD)64 (14)67 (13)68 (12)65 (12)65.8 (13)0.01
Male (%)39.842.135.354.742.70.01
BMI, kg m−2, mean (SD)27.6 (5.3)27.1 (5.0)27.8 (5.2)28.6 (5.2)27.8 (5.2)0.14
Systolic blood pressure, mmHg, mean (SD)143 (22)143.5 (21)146.7 (22)147.4 (21)145 (22)0.21
Current smoking (%)32.240.529.331.232.90.28
Daily drinking (%)15.71419.523180.21
Low physical activity (%)12.810.514.411.612.40.82
Cholesterol, mmol L−1, mean (SD)5.6 (1.1)5.8 (1.1)5.8 (1.1)5.5 (1.3)5.7 (1.2)0.11
HbA1c, %, mean (SD)6.5 (0.3)6.8 (0.0)7.2 (0.1)8.9 (1.5)7.3 (1.2)<0.001
Glucose, mmol L−1, mean (SD)6.7 (2.5)6.4 (2.0)7.3 (3.3)12.0 (5.2)8.0 (4.1)<0.001
All-cause mortality, % (n)23.2 (49)22.8 (26)21.8 (29)26.6 (37)23.6 (141)0.808
CVD, % (n)18.0 (38)15.8 (18)19.6 (26)22.3 (31)18.9 (113)0.599
Subjects without DM at inclusion
 

4.1 < 5.4

(= 1202)

5.4–5.7

(= 1155)

5.7–6.0

(= 1060)

6.0–6.5

(= 1113)

= 4530 
Age, mean (SD)54 (16)56 (17)57 (17)60 (16)56.5 (17)<0.001
Male (%)40.843.641.839.841.50.29
BMI, kg m−2, mean (SD)24.8 (3.7)25.3 (4.1)25.5 (4.2)26.0 (4.5)25.4 (4.1)<0.001
Systolic blood pressure, mmHg, mean (SD)131.8 (21)135.1 (22)136 (22)139.1 (23)135.4 (22)<0.001
Current smoking (%)30.833.535.532.232.90.11
Daily drinking (%)14.914.714.917.115.40.35
Low physical activity (%)7.58.48.27.27.80.69
Cholesterol, mmol L−1, mean (SD)5.4 (1.1)5.4 (1.1)5.5 (1.2)5.6 (1.1)5.5 (1.1)<0.001
HbA1c, %, mean (SD)5.2 (0.2)5.6 (0.1)5.9 (0.1)6.3 (0.1)5.7 (0.4)<0.001
Glucose, mmol L−1, mean (SD)5.3 (0.9)5.5 (1.0)5.5 (1.0)5.7 (1.1)5.5 (1.0)<0.001
All-cause mortality, % (n)9.3 (112)11.9 (137)14.3 (151)17.2 (191)13.1 (591)<0.001
CVD, % (n)8.2 (99)9.7 (112)10.3 (109)14.3 (159)10.6 (479)<0.001
Incident DM, % (n)0.6 (7)0.8 (9)1.4 (15)2.7 (30)1.4 (61)<0.001

HbA1c and all-cause mortality

In the nondiabetic population, there was a trend towards an association between HbA1c and all-cause mortality [univariate hazard ratio (HR) 1.21, 95% CI 0.99–1.47, = 0.06; multivariate HR 1.17, 95% CI 0.96–1.44, = 0.12]. However, this relationship was not statistically significant (Fig. 1). In the diabetic population, HbA1c was not associated with all-cause mortality (univariate HR 1.02, 95% CI 0.88–1.19, = 0.77; multivariate HR 1.0, 95% CI 0.85–1.17, = 0.99). C-statistics without and with HbA1c in the full model were 0.802 (95% CI 0.78–0.82) and 0.803 (95% CI 0.78–0.82), respectively. HbA1c did not improve risk prediction in the NRI analysis (= 0.19).

Figure 1.

Relationships between HbA1c (per 1 unit increase) and all-cause mortality, incident CVD events and incident DM in the Copenhagen City Heart Study, fourth cross-sectional survey.

HbA1c and incident CVD

In the nondiabetic population, HbA1c was significantly associated with incident fatal and nonfatal CVD events in both univariate (HR 1.38, 95% CI 1.11–1.71, = 0.004)) and multivariate analyses (HR 1.31, 95% CI 1.05–1.64, = 0.018). Figure 2 shows the relationship between quartiles of HbA1c and incident CVD in the nondiabetic population.

Figure 2.

Relationship between level of HbA1c and incident CVD in the nondiabetic population. Subjects were divided into quartiles of HbA1c according to baseline values. Log-rank test for equality of hazards amongst quartiles.

In the diabetic population, there was a trend towards an association between HbA1c and incident fatal and nonfatal CVD events but this association was only borderline significant (univariate HR 1.12, 95% CI 0.97–1.30, = 0.13; multivariate HR 1.11, 95% CI 0.95–1.30, = 0.2). C-statistics without and with HbA1c in the full model were 0.75 (95% CI 0.72–0.78) and 0.75 (95% CI 0.73–0.78), respectively. HbA1c significantly improved the risk prediction of CVD in the NRI analysis (13%, = 0.04).

HbA1c and incident DM

In the nondiabetic population, increasing levels of HbA1c within the normal range were highly correlated with development of DM. As shown in Fig. 3, subjects in the highest quartile of HbA1c were at a considerably increased risk of developing DM. This increase in risk was evident during the entire follow-up period. Thus, there was an almost fourfold increase in the risk of incident DM per unit increase in HbA1c (univariate HR 3.83, 95% CI 1.96–7.51, < 0.001). This relationship was essentially unchanged after multivariate adjustments (HR 4.19, 95% CI 2.01–8.71, < 0.01). C-statistics without and with HbA1c in the full model were 0.73 (95% CI 0.66–0.81) and 0.79 (95% CI 0.73–0.86), respectively. Furthermore, HbA1c significantly improved the risk prediction of incident DM in the NRI analysis (42%, = 0.02).

Figure 3.

Relationship between level of HbA1c and incident DM in the nondiabetic population. Subjects were divided into quartiles of HbA1c according to baseline values. Log-rank test for equality of hazards amongst quartiles.

There was no interaction with gender in any of the analyses.

Discussion

In this large-scale population study, we demonstrated that HbA1c is correlated with development of CVD amongst nondiabetic individuals. Thus, individuals without DM with HbA1c levels in the normal reference range had an increased risk of developing CVD. The increased risk was seen in the three highest quartiles of HbA1c. In the highest quartile, the risk was apparent after <2 years of follow-up. No significant relationship between HbA1c and development of CVD was found amongst patients with DM. Additionally, we found a statistically significant association between increased levels of HbA1c and the development of DM. In nondiabetic individuals with HbA1c in the highest quartile (i.e. below the suggested diagnostic level of <6.5% and below the therapeutic goal for diabetic individuals), the future risk of DM was increased. Furthermore, we found that addition of HbA1c to the model significantly improved risk prediction of diagnosed DM and coronary heart disease in the NRI analysis.

The findings are in agreement with WHO recommendations of using HbA1c as a diagnostic test, although the exact threshold has not been clearly defined. HbA1c reflects long-term glycaemic exposure and is intuitively a more reliable marker than single measures of plasma glucose concentration. HbA1c has several advantages compared to fasting plasma glucose measurements, such as lower pre-analytical instability (i.e. due to lack of dependence on the timing of sample collection) and biological variance [13-15]. The role of blood glucose concentrations and HbA1c levels in risk prediction of CVD is controversial. Population-based studies demonstrated significant associations if HbA1c was assessed on a linear basis, although findings were not statistically significant when subjects with prior CVD and DM were excluded [16], or after adjusting for gender and CVD risk factors [17]. In a recent analysis from the Atherosclerosis Risk in Communities (ARIC) study [9], Selvin et al. [9] found that risk of coronary artery disease was increased in nondiabetic individuals with high baseline HbA1c. Furthermore, they found a nonlinear relationship between HbA1c levels and all-cause mortality, with individuals in the lowest HbA1c category (<5.0%) having the highest risk of death. Several differences in the study populations may account for that fact that these results are not comparable to the present findings. The ARIC population had higher average HbA1c and BMI and a higher rate of hypertension. Furthermore, participants in the ARIC study were younger with less variance in age within the categories of HbA1c. This could be related to a greater risk of CVD according to the findings of a recent study by Gerstein et al. [18], in which men < 65 years and women < 55 years of age had an increased risk of myocardial infarction, compared to older participants in similar HbA1c categories. In the present study, we demonstrate that HbA1c is associated with CVD in a low-risk Western European population.

Previous studies examining the role of HbA1c as a predictor of DM have traditionally been performed in high-risk [19-21], small or heterogeneous populations [22, 23]. In the present study, we found a clear relationship between HbA1c levels and future DM in a general population.

Our results did not extend to the finding of statistically significant associations between HbA1c and incident CVD events or all-cause death in diabetic individuals. Furthermore, development of macrovascular complications was not related to HbA1c levels in these subjects. This is in contrast to the results of previous epidemiological cohort studies, which however lacked adjustment for risk factors and standardized HbA1c measurements. Several biological mechanisms (such as the effect of reactions between glucose and different proteins leading to long-term complications and endothelial dysfunction) have been proposed to explain the relation between chronic hyperglycaemia and coronary disease. Nevertheless, interventional studies have shown little or a possible harmful effect of lowering HbA1c to prevent CVD events in diabetic individuals [24-27] underlining that whether or not the relationship between DM and CVD (and atherosclerosis) is causal remains unknown and indeed that HbA1c may be distal to the actual pathological effects of hyperglycaemia.

There are some possible limitations to consider in the present study. First, only baseline HbA1c was measured, which makes it difficult to evaluate changes over time. However, HbA1c reliably reflects glycaemic status over the proceeding weeks and months in diabetic and nondiabetic individuals [28]. By performing a nested case–control study from the cohort data set, changes over time could be taken into account. However, this would raise other concerns such as selection bias. Fasting glucose and oral glucose tolerance tests with 2-h plasma glucose data were not available in our study, thus there is a risk that individuals with unrecognized DM or prediabetes and HbA1c levels in the upper range of normal may have been missed. However, all individuals with HbA1c levels ≥6.5% and those taking antidiabetic medication were classified as having DM. Findings in the diabetic population might suggest that the study was underpowered. Due to the nature of the study design, residual confounding cannot be ruled out completely.

It is noteworthy that despite advantages, HbA1c has important limitations. HbA1c levels differ across ethnic, racial and age groups [29-31]. In the present study, we have a very homogenous Caucasian population, which is an advantage when interpreting our findings in terms of ethnic and racial differences. Furthermore, HbA1c is affected by conditions that alter the lifespan of erythrocytes, such as types of anaemia, and genetic variants including haemoglobinopathies and thalassaemia syndromes. However, in our homogenous study population, these conditions are rare.

Conclusion

In this large-scale population study, we have demonstrated that HbA1c level is highly correlated with development of CVD amongst individuals without DM. Furthermore, we found a strong association between level of HbA1c, even within the normal range and the development of DM. In addition to this, using the NRI index significantly improved risk prediction of future CVD and DM in the nondiabetic population. This confirms the appropriateness of using HbA1c levels for assessing risk in patients. We found no relationship between HbA1c and incident CVD or all-cause mortality in individuals with DM.

Conflict of interest statement

No conflicts of interest to declare.

Acknowledgments

This study was supported by grants from the Danish Heart Foundation, the Research Foundation of Herlev Hospital, the Aase and Ejnar Danielsen Foundation and the Hede-Nielsen Foundation.

Ancillary