The handle http://hdl.handle.net/1887/39795 holds various files of this Leiden University dissertation
Author: Gast, Karin
Title: Insulin resistance and atherosclerosis : the role of visceral fat
Issue Date: 2016-06-01
Insulin resistance and atherosclerosis:
the role of visceral fat
Karin B. Gast
PhD Thesis, Leiden University Medical Center, Leiden
Cover, Layout, and printing: Optima Grafische Communicatie, Rotterdam ISBN/EAN: 978-94-6169-871-1
© 2016, K.B. Gast
The copyright of the articles that have been published has been transferred to the respective
journals. All rights reserved. No part of this book may be reproduced, stored in a retrieval sys-
tem, or transmitted in any form or by any means without prior written permission of the author.
Proefschrift
ter verkrijging van
de graad van Doctor aan de Universiteit Leiden, op gezag van Rector Magnificus prof. mr. C.J.J.M. Stolker,
volgens besluit van het College voor Promoties te verdedigen op woensdag 1 juni 2016
klokke 10.00 uur
door
Karin Bianca Gast geboren te Leiden
in 1984
Prof. dr F.R. Rosendaal
Prof. dr J.W.A. Smit (Radboud Universiteit Nijmegen)
Copromotor:
Dr ir R. de Mutsert
Leden promotiecommissie:
Dr ir J.W.J. Beulens (Vrije Universiteit Amsterdam)
Prof. dr M. den Heijer (Univeristeit Leiden en Vrije Universiteit Amsterdam) Prof. dr S. Middeldorp (Universiteit van Amsterdam)
Prof. dr H. Pijl
Financial support by the Dutch Heart Foundation for the publication of this thesis is gratefully
acknowledged.
Chapter 1 General introduction 9
Chapter 2 Insulin resistance and risk of incident cardiovascular events in adults without diabetes: meta-analysis
23
Chapter 3 Abdominal adiposity largely explains associations between insulin resistance, hyperglycemia and subclinical atherosclerosis: the NEO study
37
Chapter 4 The association of abdominal subcutaneous and visceral fat with insulin resistance and insulin secretion: the NEO study
51
Chapter 5 Individual contributions of total body fat and visceral fat to subclinical atherosclerosis: the NEO study
69
Chapter 6 Reproducibility of carotid intima-media thickness
measurement in overweight and obese adults: the NEO study
85
Chapter 7 General discussion and summary 99
Appendices 115
Nederlandse samenvatting 143
References 151
Dankwoord 175
Curriculum vitae 179
Chapter 1
General introduction
1
Worldwide, the prevalence of overweight, defined as a body mass index (BMI) of 25 kg/m 2 or higher, has reached pandemic proportions. In 2013, an estimated 2.1 billion adults worldwide were overweight and among them were 671 million adults with obesity, defined as a BMI of 30 kg/m 2 or higher [1]. These numbers correspond to an age-standardized global prevalence of overweight of 36.9 % in men and 38.0% in women with the standard population based on the World Health Organization standard age structure [2]. Globally, the age-standardized prevalence of obesity was 10.1% in men and 13.9% in women, with a high variation between countries. Whereas the prevalence of obesity in the United States was 31.6% in men and 33.9%
in women, in the Netherlands, in 2013, the prevalence of obese men and women was estimated to be around 9.2% and 11.2% [3]. The global prevalence of overweight is expected to increase further and by 2030, 3.3 billion adults could be either overweight or obese (57.8%) [4].
This rising prevalence of overweight is a major threat to global health, as overweight is a well- established risk factor for type 2 diabetes and cardiovascular disease. Cardiovascular disease is worldwide the leading cause of death, accounting for 14.1 million deaths in 2012 [5], of which 38,371 were in the Netherlands [6]. In 2014, an estimated 387 million adults (8.3%) worldwide had diabetes, of whom 887 thousand (national prevalence: 7.2%) were in the Netherlands [7]. Above a BMI of 25 kg/m 2 , each 5 kg/m 2 higher BMI was associated with 40% increase in cardiovascular mortality, and more than 100% increase in diabetes-related mortality [8]. As a result, projections estimate that in the United States, when the number of obese adults will have risen with 65 million by 2030, this will add 5.4-6.8 million cases of coronary heart disease or stroke and 5.5-6.8 million cases of diabetes [9]. Therefore, the expected worldwide increase in the prevalence of overweight will have a huge impact on the incidence of type 2 diabetes and cardiovascular disease. This rise will not only affect mortality and morbidity, but will also impact on the increase of years lived with disability [10, 11].
GLuCose metaBoLIsm and atherosCLerosIs
Large observational studies have shown that persons with type 2 diabetes have a 2- to 3-fold
increased risk of cardiovascular disease [12, 13]. Overt cardiovascular disease is the end stage
of atherosclerosis progression in which the arterial wall thickens gradually and eventually a
plaque or thrombus is formed that obstructs the artery [14]. Likewise, type 2 diabetes is the
result of insulin resistance progression. Insulin resistant cells have a diminished ability to trans-
port glucose from the bloodstream into the cell, but normal plasma glucose concentrations
can be maintained by increasing insulin secretion by beta-cells in the pancreas. Once insulin
secretion cannot adequately compensate for tissue insulin resistance (i.e., beta-cell failure),
hyperglycemia develops and this may progress to type 2 diabetes. A follow-up study in 3,145
individuals has shown that insulin resistance is present many years before diabetes is diagnosed
[15], thereby already increasing insulin and glucose concentrations. Therefore, it is important
to know whether insulin resistance, hyperinsulinemia, and hyperglycemia are associated with atherosclerosis and cardiovascular disease before type 2 diabetes is diagnosed.
Animal studies have shown that high plasma glucose and insulin concentrations, as well as insulin resistance can induce atherosclerosis [16-18]. Insulin resistance may be involved in the development of early and advanced atherosclerosis via various mechanisms, including dyslipidemia and inflammation [16]. Hyperglycemia appears to be exclusively involved in early atherosclerosis and contributes to endothelial dysfunction by increased oxidative stress and production of advanced glycation endproducts [17]. In addition, insulin resistance appears to modify the effect of insulin on the vascular wall; insulin is anti-atherogenic in the insulin sensi- tive state and pro-atherogenic in the insulin resistant state [18]. Unfortunately, it is not clear to what extent these mechanisms contribute to the development of atherosclerosis in humans.
Results from randomized controlled trials suggest that the contribution of hyperglycemia to the development of atherosclerosis are limited, since intensive glycemic control (i.e. lowering HbA 1C concentrations) did not result in a reduction of the incidence of cardiovascular disease in persons with pre-diabetes or type 2 diabetes.
Recent meta-analyses, however, have shown that high plasma glucose concentrations, and to a lesser extent, high insulin concentrations in persons without diabetes were associated with an increased risk of cardiovascular disease [13, 19]. Also on a subclinical level of atherosclerosis, previous studies reported that measures of insulin resistance [20, 21] and hyperglycemia [21- 23] were associated with carotid intima-media thickness (cIMT), a marker of atherosclerosis, in persons without diabetes. However, insulin resistance precedes hyperglycemia and these observational studies have not distinguished their individual contributions to subclinical atherosclerosis. Therefore, atherosclerosis and cardiovascular disease may be caused by insulin resistance rather than being a consequence of the toxic effects of elevated glucose concentra- tions. This distinction is essential to understand the development of atherosclerosis and can thereby guide specific therapeutic interventions.
Furthermore, when studying associations between insulin resistance, hyperglycemia and atherosclerosis, it is important to take adiposity into account. Adiposity can lead to insulin resistance, and consequential hyperglycemia, and can contribute to atherosclerosis through various mechanisms which will be discussed in more detail in the following paragraphs. Hence, adiposity can be considered as a common cause of both insulin resistance and atherosclerosis [24] and, may be responsible for the observed associations between insulin resistance and atherosclerosis (Figure 1).
aBdomInaL oBesIty
During the past fifty years, it has emerged that not only the degree of adiposity is a risk factor for
type 2 diabetes and cardiovascular disease, but also the way the adipose tissue is distributed in
1
the body. Already in 1956 it was observed that persons with an accumulation of adipose tissue around the abdomen more often had type 2 diabetes and cardiovascular disease than those with adipose tissue elsewhere [25]. From 1980s onwards, large cohort studies have shown that anthropometric measures of abdominal obesity such as waist circumference and waist-to-hip ratio are associated with an increased risk of type 2 diabetes and cardiovascular disease, even after adjustment for overall adiposity [26-33]. Waist circumference and waist-to-hip ratio, how- ever, cannot distinguish between abdominal subcutaneous and visceral adipose tissue. With the development of more advanced imaging techniques, such as computed tomography (CT) and magnetic resonance imaging (MRI), subcutaneous and visceral adipose tissue depots could be directly quantified. Small cross-sectional studies showed that the anthropometric measures of abdominal adiposity were more strongly associated with the amount of intra-abdominal adipose tissue (i.e., visceral adipose tissue) than with subcutaneous adipose tissue [34-36]. In 1987 it was first shown that in obese persons visceral fat accumulation was associated with dis- turbances in glucose and lipid metabolism [37]. These results ultimately led to the hypothesis that the excess risk of type 2 diabetes and cardiovascular disease associated with abdominal obesity is due to increased amounts of visceral adipose tissue [38]. Several mechanisms have been proposed to explain how excess adiposity and visceral fat accumulation specifically lead to type 2 diabetes and cardiovascular disease. Adipose tissue dysfunction may be the central mechanism linking obesity and visceral fat accumulation to insulin resistance and atheroscle- rosis.
Total body fat Visceral fat Insulin resistance
L
Atherosclerosis Hyperglycemia
Ch 5 Ch 3
Ch 3 Ch 4
Figure 1. Hypothesis Path diagram
Figure represents disease progression from fat accumulation to atherosclerosis.
Ch, Chapter; L, known confounding factors including age, sex, ethnicity, education, tobacco smoking, al- cohol consumption, and physical activity, but there may be unknown or unmeasured confounding factors.
───▶: evidence from literature
┄┄┄▶: pathway under study in this thesis
adIPose tIssue dysFunCtIon
Usually, excess energy is stored as triglycerides in subcutaneous adipose tissue. Subcutaneous adipose tissue on average accounts for 80 to 90% of total body fat and visceral fat for 6 to 20% [39] The ability to store excess energy predominantly in subcutaneous fat varies between individuals and probably has a multifactorial origin [39, 40]. Whereas on average women have more total body fat than men, men have more visceral fat than women [39]. Studies in identical twin pairs suggest that there is a strong genetic component of body fat distribution [41-43]
and genome wide association studies have identified loci associated with visceral fat accumula- tion. Other factors that are associated with visceral fat accumulation are: age, ethnicity, sex hormones, increased concentrations of glucocorticoids, and physical activity [44]. The main mechanism of adipose tissue to expand is by increasing the size of adipocytes and probably to a lesser extent adipocyte number [45]. Hypertrophic adipocytes (cell diameter > 150 µm) release various stress factors and could induce hypoxia. Hypoxia in adipose tissue may trigger mono- cytes to migrate into adipose tissue and differentiate into macrophages [46]. The accumulation of macrophages in adipose tissue may lead to adipose tissue dysfunction. Adipocyte tissue dysfunction is characterized by hypertrophic adipocytes, impaired adipogenesis, an increased secretion of non-esterified fatty acids (NEFAs) and pro-inflammatory cytokines (adipokines) [46, 47], and a decreased ability to further store lipids in subcutaneous adipose tissue [44]. As a result, triglycerides will be stored viscerally and in and around organs such as the liver, heart, kidney, skeletal muscle, and pancreas. This is also referred to as the lipid overflow hypothesis (Figure 2) [38, 44, 48, 49].
Dysfunctional adipose tissue is thought to promote insulin resistance and atherosclerosis mainly via secretion of NEFAs and pro-inflammatory adipokines [46, 47, 49]. Increased con- centrations of NEFAs induce fat accumulation in skeletal muscle and liver cells [50, 51]. This interferes with insulin signaling, leading to a decreased glucose uptake by skeletal muscle cells and an increased glucose production by the liver [50, 51]. Beta-cells are also susceptible to the toxic effect of NEFAs, which may induce beta-cell apoptosis [52]. Pro-inflammatory adipokines such as TNF-α and IL-6, and the hormone leptin may further suppress insulin signaling [53].
These pro-inflammatory adipokines may also be involved in several stages in the development
of atherosclerosis. In the early stage, they can promote endothelial dysfunction and monocyte
recruitment and adhesion [54, 55]. In the more advanced stage, they may stimulate smooth
muscle cell proliferation and induce a pro-coagulant state, leading to plaque and thrombus
formation [54, 55]. High concentrations of NEFAs promotes the formation of foam cells [55]. In
addition, as described previously, dysfunctional adipose tissue may also lead to atherosclerosis
through insulin resistance and consequential hyperglycemia [16].
1
Atherosclerosis Normal artery
Insulin sensitive
liver en muscles Insulin resistant
liver en muscles
Beta-cell apoptosis Normal insulin secretion
HIGH RISK OF CARDIOVASCULAR DISEASE
AND TYPE 2 DIABETES LOW RISK OF
CARDIOVASCULAR DISEASE AND TYPE 2 DIABETES
Figure 2. Adipose tissue dysfunction and its complications
Adopted from Després et al. [38] Used with permission. Copyright © 2006 Nature Publishing Group. All
rights reserved.
sPeCIFIC roLe For vIsCeraL Fat In the deveLoPment oF dIsease?
Recent studies that directly assessed visceral fat amount by CT or MRI have shown that visceral fat accumulation is associated with measures of insulin resistance [56-62] and cIMT [57, 63-68].
However, whether visceral fat is specifically involved in the development of insulin resistance and atherosclerosis remains unclear.
One the one hand, visceral adipose tissue has a higher secretion of NEFAs [69] and pro- inflammatory adipokines [70] per mg tissue than subcutaneous adipose tissue. Furthermore, visceral adipose tissue products directly drain in the portal vein, exposing the liver to elevated concentrations of NEFAs and adipokines which may lead to hepatic insulin resistance [71, 72].
This is referred to as the portal vein hypothesis and may explain why specifically visceral fat seems related to insulin resistance and atherosclerosis. On the other hand, subcutaneous adi- pose tissue represents on average 80 to 90% of total body fat [73] and may therefore contribute to a larger extent to the pool of circulating NEFAs and adipokines than visceral adipose tissue [69, 74].
Visceral fat is strongly associated with total body fat [75]. Therefore, results of studies investi- gating relationships between visceral fat, insulin resistance, hyperglycemia and atherosclerosis should among others, be adjusted for total body fat (Figure 2) [76]. Most previous studies did not adjust for total body fat and consistently observed that an increased amount of visceral fat was associated with insulin resistance [56-58] and cIMT [57, 63, 65-67]. Studies that did adjust for total body fat observed that visceral fat remained associated with insulin resistance [59-62], but inconsistent results were reported for cIMT [64, 68]. In one study visceral fat contributed to cIMT above total body fat [64], whereas in another study visceral fat did not contribute and total body fat was therefore more important in the association with cIMT [68]. Therefore, the specific contribution of visceral fat to the development of insulin resistance and atherosclerosis remain unclear.
oBjeCtIve and outLIne oF thIs thesIs
The main objective of this thesis was to unravel relationships between obesity, insulin resis-
tance, hyperglycemia and atherosclerosis (Figure 1). We investigated whether insulin resistance
and hyperglycemia are associated with subclinical atherosclerosis and incident cardiovascular
disease, and to what extent the observed associations are explained by body fat. We further
aimed to study the specific contribution of visceral fat accumulation to the development of
insulin resistance and atherosclerosis, using direct assessment of visceral and subcutaneous
adipose tissue while adjusting for total body fat.
1
In Chapter 2, we performed a meta-analysis of cohort studies on the associations of hypergly- cemia, hyperinsulinemia, and insulin resistance with incident cardiovascular disease in persons without diabetes mellitus. We hypothesized that insulin resistance is most strongly associated with incident cardiovascular disease.
The studies included in this meta-analysis had not distinguished the contributions of hyperglycemia and insulin resistance in the development of incident cardiovascular disease.
Furthermore, not all adjusted for adiposity. Therefore, in Chapter 3, we investigated the relative contributions of insulin resistance and hyperglycemia to subclinical atherosclerosis, and stud- ied to what extent associations between insulin resistance, hyperglycemia and atherosclerosis could be explained by adiposity. For this purpose, we used BMI as a measure of overall adiposity and waist circumference as a measure of abdominal adiposity.
The anthropometric measure waist circumference does not distinguish between visceral and subcutaneous adipose tissue within the abdomen. We therefore used MRI techniques to directly assess abdominal subcutaneous and visceral adipose tissue and we investigated in Chapter 4 the association of abdominal subcutaneous and visceral adipose tissue with insulin resistance and insulin secretion in men and women. In Chapter 5 we investigated the rela- tive contributions of abdominal subcutaneous and visceral adipose tissue, and their ratio to subclinical atherosclerosis, taking total body fat into account. Throughout this thesis we used the cIMT as a marker of subclinical atherosclerosis and we studied its reproducibility in persons with overweight in Chapter 6. Finally, Chapter 7 summarizes the results of this thesis and discusses its strengths, limitations and implications.
the neo study
To answer the research questions of chapters three to six, we used the baseline measurements
of the Netherlands Epidemiology of Obesity (NEO) study. The NEO study is a population-based
prospective cohort study in 6,673 individuals aged between 45 and 65 years, with an overs-
ampling of persons with a BMI of 27 kg/m 2 or higher, who were recruited in the greater area of
Leiden (in the West of The Netherlands) and included between September 2008 and October
2012. All participants underwent an extensive physical examination including anthropometric
measurements, blood sampling, and a cIMT measurement. In a random subset of 2,581 par-
ticipants without contraindications for undergoing MRI, abdominal subcutaneous and visceral
adipose tissue were assessed by MRI [77].
assessment oF Body Fat
As mentioned earlier, several anthropometric and imaging techniques are available to assess total and regional body fat. The main methods to assess total body, subcutaneous and visceral fat are summarized in table 1. The most accurate and precise methods to measure total body fat are: densitometry, hydrometry, neutron activation analysis and measurement of total body potassium. These methods are considered the “gold standards” for measurement of total body composition; however, they are expensive, time-consuming, and some expose persons to ra- diation. Therefore, these methods cannot be used in large population studies. In epidemiology, BMI is most frequently used as a measure of overall adiposity, but it cannot distinguish between fat mass and fat free mass and will, for example, misclassify particularly muscular persons as overweight, or even persons with large amounts of visceral fat but little subcutaneous fat as lean [78]. Furthermore, most methods that are used to assess total body fat do not indicate the location of the adipose tissue. Waist circumference and waist-to-hip ratio are frequently used
table 1. Different methods to asses total body fat or fat distribution: advantages, disadvantages, validity, and feasibility in large studies
Method Advantage (+)/
Disadvantage (−)
Validity Feasibility in large studies Whole body measurement
Total body potassium
A small proportion of total body potassium is radioactive (
40K) and found in known proportions in fat-free mass. Gamma rays emitted by
40K can be counted and fat-free mass can be estimated.
+ accurate and precise (gold standard)
− limited availability, very expensive, difficult
very high very low
Neutron activation analysis
A neutron stream is projected on the body, resulting in a radioactive isotope of the element of interest and the gamma rays emitted by this element can be counted. Specific elements are used to estimate total body fat.
+ accurate and precise (gold standard)
− limited availability, very expensive, difficult, radiation dose
very high very low
Hydrometry (Isotope dilution)
An isotopic tracer, such as deuterium, is ingested and distributed in body water. The dilution concentration of this isotope in excreted fluids (e.g. urine) can be measured and is used to estimate total body water and fat.
+ accurate and precise (gold standard)
− expensive, time- consuming, dependent on hydration status
very high very low
Densitometry
This method determines body volume by the volume of water (underwater weighting) or air (Whole-Body Air Displacement Plethysmography) that is displaced when a person is respectively immersed in water or is sitting in a closed chamber. The body density is calculated from body volume and weight and subsequently body fat is estimated using standard equations.
+ accurate and precise (gold standard)
− limited availability, expensive, requires specific maneuvers, equations assume that densities of fat-free mass are constant [91]
very high low
1
table 1. (continued)
Method Advantage (+)/
Disadvantage (−)
Validity Feasibility in large studies Bioelectrical Impendence Analysis
The electrical impedance is measured between two electrodes located at both feet (foot-to-foot) or hand and foot (hand-to-foot). This impedance is used to estimate total body water, and hence fat mass.
+ non-invasive, quick
− measured between two extremities, affected by hydration status
high high
Body Mass Index
BMI is calculated by dividing weight in kilograms by the height in meters squared.
+ non-invasive, quick, inexpensive
− no distinction between fat mass and fat free mass
intermediate very high
Whole body and regional measurement Computed Tomography
X-rays are projected on the body and the different ability of tissues to block these X-rays are used to create cross-sectional images of the body.
+ able to discriminate between visceral and subcutaneous fat
− radiation dose, expensive
high low
Magnetic Resonance Imaging
The body is exposed to a magnetic field and radiofrequency waves are projected on the body. A radiofrequency signal emitted by activated atomic nuclei is used to create cross-sectional images. The volume of adipose tissue can be calculated from these images.
+ able to discriminate between visceral and subcutaneous fat, non- invasive,
− expensive
high low
Dual Energy X-ray Absorptiometry
The body is scanned with two X-ray beams of different energy levels. Differences in attenuation of these X-rays by soft tissue and bone are used to create an image and to calculate body fat.
− small radiation dose, small errors (<1%) with hydration changes [92]
high intermediate
Regional measurement Ultrasound
An ultrasound beam is projected on the skin.
Differences in reflection of ultrasound waves are used to create an image and the thickness of the adipose layer of interest can be measured.
+ able to discriminate between visceral and subcutaneous fat, non- invasive, quick
high Intermediate
Waist and hip circumference
Waist circumference is usually measured midway between the border of the lower costa margin and the iliac crest, above the umbilicus. Hip circumference is measured around the largest circumference of hips or buttocks.
+ non-invasive, quick, inexpensive
− moderate correlation with total body and visceral fat in cadavers [93]
intermediate very high
Skinfold thickness
Skinfold thickness can be measured at different sites and represents subcutaneous adipose tissue.
+ non-invasive, quick, inexpensive
− high inter-observer variability [94]
intermediate very high
References are listed in the reference list.
BMI, Body Mass Index
in epidemiological studies to reflect accumulation of adipose tissue around the abdomen [26, 27, 30, 32]. Imaging modalities such as echography, MRI, and CT, allow direct quantification of visceral and subcutaneous adipose tissue depots. In the NEO study, total body fat was assessed by bioelectrical impedance analysis in all participants and by Dual Energy X-ray Absorptiometry in a random subset of 916 participants. BMI, waist and hip circumference were measured and waist circumference was used as a measure of abdominal adiposity. Abdominal subcutaneous and visceral adipose tissue depots were quantified with MRI.
assessment oF GLuCose metaBoLIsm
Insulin resistance and impaired insulin secretion, the two key features of type 2 diabetes, can be
measured using direct and indirect methods. Direct methods to assess insulin sensitivity are the
insulin suppression test and the hyperinsulinemic euglygemic glucose clamp. This euglycemic
clamp is considered the gold standard for the direct assessment of insulin sensitivity. In this
technique, insulin is infused at a constant rate and to maintain euglycemia, glucose is infused at
a variable rate. Insulin sensitivity is estimated using the ratio of the mean glucose infusion to the
mean insulin concentrations over the last 20 to 30 minutes of the clamp [79]. Its validity relies on
several assumptions, namely achieving a steady state condition and complete suppression of
hepatic gluconeogenesis. The glucose clamp technique is also time-consuming, expensive, and
difficult to assess. Consequently, direct assessment of insulin sensitivity, is not feasible in large
population studies. Therefore, in such studies insulin sensitivity is often assed by non-invasive,
inexpensive indirect methods such as the Quantitative Insulin Sensitivity Check Index (QUICKI),
Matsuda-index or the Homeostasis Model Assessment of insulin resistance (HOMA-IR). HOMA
is a mathematical model which can be used to estimate insulin resistance and insulin secretion
(HOMA-B) from fasting plasma glucose and insulin concentrations [80]. HOMA estimates of
insulin resistance and insulin secretion correspond well to estimates derived from euglycemic
and hyperglycemic glucose clamps (gold standards) [81, 82]. A limitation of HOMA is that
insulin resistance and insulin secretion are estimated solely from fasting plasma glucose and
insulin concentrations and can therefore not account for disturbances in non-fasting condi-
tions. The Matsuda Insulin Sensitivity Index is calculated using fasting and non-fasting plasma
glucose and insulin concentrations and therefore also accounts for disturbances in non-fasting
conditions [83]. The Insulinogenic Index is the ratio of the increment in insulin concentration
to glucose concentration 30 minutes after an oral glucose tolerance or mixed meal test. The
Insulinogenic Index reflects first-phase insulin secretion and loss of this response may be the
earliest sign of impaired insulin secretion in patients with type 2 diabetes [82]. In the NEO study,
the Insulinogenic Index and Matsuda Insulin Sensitivity Index were calculated after an mixed
meal and used as estimates of insulin secretion and insulin sensitivity respectively. In addition,
both Indices were used to calculate the Disposition Index (Matsuda Insulin Sensitivity Index
1
* Insulinogenic Index) which is a measure of insulin secretion that accounts for variations in whole body insulin sensitivity [84]. HOMA was used to estimate both insulin sensitivity (HOMA- IR) and insulin secretion (HOMA-B).
assessment oF atherosCLerosIs
Several non-invasive imaging techniques can be used to assess atherosclerosis, each reflecting different stages in the development of atherosclerosis. Traditionally, angiography is considered the gold standard for the detection of arterial stenosis. Disadvantages of angiography are its in- vasive nature and that it provides no information regarding the thickness or composition of the arterial wall. The thickness of the intima and media layer in the carotid artery can be visualized and measured by ultrasound. This so-called carotid intima-media thickness (cIMT) is considered a marker of early atherosclerosis [85]. Ultrasound can also be used to assess the presence of plaques in the carotid artery, which is considered a marker of more advanced atherosclerosis [85]. This can also be done with more expensive methods such as CT and MRI, and recently positron emission tomography (PET) imaging has been proposed as a method to measure arterial inflammation [86]. Other methods to assess early atherosclerosis are flow mediated dilatation or pulse wave velocity, but these are more time-consuming and thereby less feasible in large population studies. The carotid intima-media thickness corresponds to histology of the arterial wall [87] and is strongly associated with future risk of cardiovascular disease [88, 89]. For this reason, cIMT is frequently used as a marker of atherosclerosis to investigate the effect of potential cardiovascular risk factors on atherosclerosis.
Precision of the cIMT measurement is important, because observed variance of cIMT should
reflect ‘true differences’ between individuals and not measurement error. Previous studies have
shown that the reproducibility of the cIMT varies according to the measured artery, the equip-
ment, number of assessors, and study population [90]. In overweight individuals excess fat in
the neck region may render cIMT difficult to assess, though its reproducibility in a population
of adults with overweight is unknown. In the NEO study, we studied the reproducibility of the
cIMT in persons with overweight and used the cIMT as a measure of subclinical atherosclerosis.
Chapter 2
Insulin resistance and risk of incident cardiovascular events in adults without diabetes: meta-analysis
K.B. Gast N. Tjeerdema T. Stijnen J.W.A. Smit O.M. Dekkers
PLoS One 2012; 7: e52036. doi:10.1371/journal.pone.0052036
aBstraCt
Background
Glucose, insulin and Homeostasis Model Assessment Insulin Resistance (HOMA-IR) are markers of insulin resistance. The objective of this study is to compare fasting glucose, fasting insulin concentrations and HOMA-IR in strength of association with incident cardiovascular disease.
Methods
We searched the PubMed, MEDLINE, EMBASE, Web of Science, ScienceDirect and Cochrane Library databases from inception to March, 2011, and screened reference lists. Cohort studies or nested case-control studies that investigated the association between fasting glucose, fast- ing insulin or HOMA-IR and incident cardiovascular disease, were eligible. Two investigators independently performed the article selection, data extraction and risk of bias assessment. Car- diovascular endpoints were coronary heart disease (CHD), stroke or combined cardiovascular disease. We used fixed and random-effect meta-analyses to calculate the pooled relative risk for CHD, stroke and combined cardiovascular disease, comparing high to low concentrations of glucose, insulin or HOMA-IR. Study heterogeneity was calculated with the I 2 statistic. To en- able a comparison between cardiovascular disease risks for glucose, insulin and HOMA-IR, we calculated pooled relative risks per increase of one standard deviation.
Results
We included 65 studies (involving 516,325 participants) in this meta-analysis. In a random-effect meta-analysis the pooled relative risk of CHD (95% CI; I 2 ) comparing high to low concentrations was 1.52 (1.31, 1.76; 62.4%) for glucose, 1.12 (0.92, 1.37; 41.0%) for insulin and 1.64 (1.35, 2.00;
0%) for HOMA-IR. The pooled relative risk of CHD per one standard deviation increase was 1.21 (1.13, 1.30; 64.9%) for glucose, 1.04 (0.96, 1.12; 43.0%) for insulin and 1.46 (1.26, 1.69; 0.0%) for HOMA-IR.
Conclusions
The relative risk of cardiovascular disease was higher for an increase of one standard deviation
in HOMA-IR compared to an increase of one standard deviation in fasting glucose or fasting in-
sulin concentration. It may be useful to add HOMA-IR to a cardiovascular risk prediction model.
2
IntroduCtIon
Cardiovascular disease is worldwide the leading cause of death [5]. Type 2 diabetes contributes importantly to cardiovascular disease, because it is highly prevalent and doubles cardiovascular disease risk [12, 13]. Before type 2 diabetes is diagnosed, insulin resistance can be present for years, thereby increasing insulin and glucose concentrations [15, 95].
Recent meta-analyses have shown that elevated insulin and glucose concentrations in persons without diabetes were associated with an increased cardiovascular disease risk [13, 19]. In accordance, mechanistic studies have shown that elevated glucose and insulin con- centrations can be pro-atherogenic [17, 18]. Elevated insulin and glucose concentrations are direct consequences of insulin resistance. Insulin resistance can promote the development of atherosclerosis through elevated glucose and insulin concentrations, but also through mechanisms that involve dyslipidemia, hypertension, and inflammation [16, 17]. Therefore, cardiovascular disease may be caused by insulin resistance rather than being a consequence of the toxic effects of elevated insulin or glucose concentrations. A validated and frequently used marker of insulin resistance is the Homeostasis Model Assessment Insulin Resistance (HOMA-IR). Since, HOMA-IR incorporates both glucose and insulin concentrations and repre- sents insulin resistance, which can promote atherosclerosis trough several mechanisms [16, 17], it might be more strongly associated with cardiovascular disease than individual glucose or insulin concentrations. No meta-analysis thus far, has compared the strength of association between HOMA-IR and cardiovascular disease to associations between fasting glucose, fasting insulin and cardiovascular disease.
Our aim was to perform a systematic review and meta-analysis on the association between fasting glucose, fasting insulin, HOMA-IR and incident cardiovascular disease in individuals without diabetes. Our second aim was to compare fasting glucose, fasting insulin and HOMA-IR in strength of association with incident cardiovascular disease. We hypothesized that HOMA-IR is more strongly associated with incident cardiovascular disease than fasting glucose or fasting insulin.
methods
Data Sources and Searches
We searched the following databases from their inception to February 23, 2010: PubMed,
MEDLINE, EMBASE, Web of Science, ScienceDirect and Cochrane Library. We updated the search
to February 29th, 2011 for the MEDLINE and PubMed databases. The search strategy was opti-
mized for all consulted databases, taking into account the differences of the various controlled
vocabularies as well as the differences of database-specific technical variations (e.g. the use of
quotation marks). The reference lists of all potentially relevant articles were screened for ad-
ditional publications. Detailed and database specific information about the search strategy is shown in appendix table 1.
Study Selection
The aim of our meta-analysis was to investigate the association between fasting glucose, fasting insulin, HOMA-IR and incident cardiovascular disease in individuals without diabetes at baseline. Cohort studies that measured glucose, insulin or HOMA-IR and reported original data on their association with cardiovascular disease, were eligible. We considered only cohort studies or nested case-control studies that measured glucose or insulin concentrations prior to the assessment of cardiovascular disease with a subsequent follow-up of minimally one year.
No cross-sectional studies were eligible. In addition, articles in other languages than English were not eligible.
Since anti-diabetic drugs influence insulin and glucose concentrations, study populations should preferably have excluded participants with overt diabetes at baseline. However, population based studies that did not exclude participants with overt diabetes at baseline were eligible for inclusion. We excluded studies performed in populations exclusively consisting of persons with known diabetes or cohorts restricted to specific populations such as intensive care or transplant patients.
Studies that measured glucose or insulin concentrations in the fasting state were eligible for inclusion. Unfortunately, no uniform definition of fasting exists and many different defini- tions are being used [96]. Concentrations were considered to be fasting if study participants abstained from food for at least eight hours. Studies that reported the glucose or insulin con- centrations to be fasting or measured after an overnight fast, but did not report the time span of fasting, were not excluded.
Studies reporting on at least one of the following endpoints were eligible: myocardial infarc- tion, angina pectoris, stroke (ischemic or hemorrhagic), arrhythmias, congestive heart failure or sudden cardiac death separately or combinations. Studies that combined these endpoints with peripheral arterial disease, arterial aneurysm or arterial dissection in a composite endpoint were not excluded.
Furthermore, to be included studies should (1) report the association by comparing catego-
ries (percentiles or cut-off values), (2) express the association as relative risks (hazard ratios,
rate ratios, risk ratios or odds ratios) with corresponding standard errors, confidence intervals
or exact p-values and (3) adjust effect estimates at least for age and sex. In case of multiple
publications arising from the same study population we included the study with the highest
number of participants or the longest follow-up.
2
Data Extraction and Quality Assessment
Two investigators (K.G. and N.T.) independently performed the article selection based on titles and abstracts, data extraction and risk of bias assessment using a standard data sheet.
Disagreement was resolved by consensus or by a third party (O.D.).
If necessary, glucose and insulin concentrations were recalculated to the international system of units (i.e. mmol/L for glucose and pmol/L for insulin) [97]. Values for HOMA-IR were based on values provided by the authors of included studies. In general, HOMA-IR is calculated by the formula: (fasting insulin x fasting glucose)/ 22.5 or by the more recently updated computer model [81]. We recalculated HOMA-IR values for studies that reported HOMA insulin sensitivity, which is the reciprocal of HOMA-IR.
We categorized study endpoints as (fatal or non-fatal): (1) coronary heart disease (CHD), (2) stroke and as (3) combined cardiovascular disease outcome (CVD), including studies contribut- ing to 1 or 2. CHD was defined as myocardial infarction or angina pectoris; stroke consisted of hemorrhagic or ischemic stroke and CVD consisted of myocardial infarction, angina pectoris, hemorrhagic stroke, ischemic stroke, arrhythmias, congestive heart failure or sudden cardiac death.
Risk of bias assessment was based on design elements of cohort studies and nested case- control studies that could potentially bias the association between fasting glucose, fasting insulin, HOMA-IR and cardiovascular disease. Potential sources of bias were assessed by using a predefined assessment form. Dimensions considered for both cohort and nested case-control studies were (1) presence of overt diabetes at baseline, (2) presence of cardiovascular disease at baseline, (3) adequacy of exposure measurement, (4) missing glucose, insulin or HOMA-IR data, (5) adequacy of endpoint ascertainment. Bias was considered to be likely present when:
(1) study populations had overt diabetes prevalence of twice their country specific diabetes prevalence estimates of 2011 [98]; indicating that studies have selected their study population based on high glucose concentrations (selection bias), (2) persons with prevalent cardiovas- cular disease according to their outcome definition were not excluded; (3) the time span of fasting was not reported, (4) ≥10% missing data of the exposure except when data was missing completely at random (e.g. in the case of later introduction of the measurement), (5) outcome classification was based on self- or family reports, (6) there was ≥10% loss to follow-up. Reliable methods of outcome assessment were assessment by medical records, death certificates or hospital discharge records. Diagnosis of myocardial infarction was considered reliable when WHO MONICA criteria or Minnesota coding of electrocardiograms during follow-up visits were used [99-101].
Data Synthesis and Analysis
Hazard ratios, rate ratios, risk ratios or odds ratios (relative risks) of cardiovascular disease comparing high to low concentrations of glucose, insulin or HOMA-IR values were extracted.
If necessary, we recalculated these relative risks in a way that the lowest category (percentile
or cut-off value) comprised the reference category. Our first aim was to estimate the pooled relative risk for cardiovascular disease, when comparing categories (based on either percentiles or cut-offs) of high concentrations of glucose, insulin or HOMA-IR to categories of lower con- centrations. We pooled maximally adjusted effect measures of studies with corresponding 95%
confidence intervals (CI). For all analyses, both a fixed and a random-effect meta-analysis were performed. Study heterogeneity was calculated with the I 2 statistic. Elements of the risk of bias assessment were used to explore potential heterogeneity in sensitivity analyses. We assessed the presence of funnel plot asymmetry by calculating Egger’s test [102].
Our second aim was to compare fasting glucose, fasting insulin and HOMA-IR in strength of association with cardiovascular disease by comparing pooled standardized relative risks (i.e.
risk increase per increase of one standard deviation). First, we calculated the standard deviation per exposure by pooling reported standard deviations with a weight factor based on study size.
Secondly, we applied the method of Hartemink et al. [103] to calculate an overall relative risk per one unit increase of the exposure. Then, we multiplied the logarithm of the relative risks by the pooled standard deviation of the exposure. In short, the method of Hartemink et al. [103]
assumes a log-linear relation between the risk and the exposure. The input of the algorithm consists of the means and variances of the exposure within each category of the exposure, the log relative risks of the categories with respect to a reference category, and the number of cases within each category. To determine the category means and variances we applied vari- ous methods, depending on the kind of data reported in the article. We assumed a lognormal distribution for the exposures. Finally, we tested differences in pooled relative risks between the three exposures by using multivariate meta-analysis. Relative risks obtained from the same study (i.e. for studies that reported relative risks for more than one exposure) are likely to be correlated and this correlation is taken into account by multivariate meta-analysis.
We investigated sex differences in studies that presented sex-specific relative risks of car- diovascular disease by performing meta-analyses stratified by sex. Statistical analyses were performed with STATA Statistical Software (Statacorp, College Station, Texas, USA), version 11.2 and SAS software (SAS Institute Inc., Cary, NC, USA), version 9.2.
resuLts
Search Results
We identified 4,792 unique publications by database search (MEDLINE: n = 2,095, PubMed:
n = 1,480, EMBASE n = 852, Cochrane: n = 112, ScienceDirect: n = 103, Web of Science: n =
86) and by screening reference lists of potentially relevant articles (n = 64). After exclusion of
4,469 publications by screening title and abstract, 323 publications were retrieved for detailed
assessment of which 184 fulfilled inclusion criteria and were assessed in duplicate. To avoid
multiple inclusions of the same study participants, we excluded 32 publications originating
2
from the same study populations and included the publication with the largest population or the longest follow-up. Sixty-five studies (from 64 publications) were included. Forty-five studies presented data on fasting glucose, 16 studies presented data on fasting insulin and 17 studies presented data on HOMA-IR (Figure 1).
4,792 Unique publications retrieved
4,469 Publications excluded by screening title and abstract 323 Publications retrieved for detailed assessment 139 Publications did not fulfill
inclusion criteria:
84 no relevant or original data 17 no relative risk
15 cross-sectional
13 specific study population 8 duplicates
2 not English 184 Publications assessed in duplicate
Glucose:140 publications excluded:
90 no fasting glucose data 22 no categories 21 duplicates 5 no relative risk
1 reference category unclear 1 not adjusted for age or sex
HOMA-IR: 17 studies included 4,728 Publications
retrieved through database searching
Insulin: 168 publications excluded:
145 no fasting insulin data 16 no categories 6 duplicates 1 no relative risk
HOMA-IR:167 publications exlcuded:
157 no HOMA-IR data 5 no categories 5 duplicates 65 Unique studies a included
Insulin: 16 studies included Glucose: 45 studies included
64 Publications indentified through reference lists
searching
Figure 1. Summary of search results
a
One publication consisted of two studies
HOMA-IR, Homeostasis Model Assessment insulin resistance; RR, relative risk
Study characteristics
Study characteristics of the included studies are summarized in table 1. Sixty-four cohort stud- ies and 1 nested case-control study were included. The controls in this case-cohort study were matched on time and therefore the odds ratio corresponds to a rate ratio [104]. Fifty-six studies presented a hazard ratio and nine studies presented an odds ratio. Most study populations consisted of both men and women. Individual study characteristics of included studies are shown in appendix table 2.
table 1. Study characteristics of the included studies summarized for three exposures
Characteristic
Exposure Glucose
(45 studies)
Insulin (16 studies)
HOMA-IR (17 studies)
Total participants 450,487 46,236 51,161
Participants per study (range) 541-63,443 541-13,446 839-6,942
Year of publication 1983-2010 1992-2010 2001-2010
Mean follow-up (years, range) 3.2-23.5 5.0-22.3
a2.2-30
Study design
Cohort 45 15 17
Nested case-control 0 1 0
CHD endpoint
Number of studies 23 9 7
Events per study 23-4,490
b16-677 33-169
bTotal events 10,884
b2,149 441
bStroke endpoint
Number of studies 14 2 4
Events per study 13-405
c25-70 23-70
bTotal events 1,936
c95 164
bCombined CVD endpoint
Number of studies 45 16 17
Events per study 23-4,490
b16-492 58-340
Total events 19,993
b3,329 3,035
Data are presented as number or range.
a
Three studies did not report follow-up time
b
Two studies did not report the number of participants who encountered the outcome of interest.
c
One study did not report the number of participants who encountered the outcome of interest.
HOMA-IR, Homeostasis Model Assessment Insulin Resistance; CVD, cardiovascular disease; CHD, coronary heart disease
Risk of bias
The risk of bias assessment is summarized in appendix table 3 and shown per study in appen-
dix table 4. Most studies excluded persons with overt diabetes at baseline. One study included
2
persons with prevalent cardiovascular disease and this was unclear in 20 studies. Twenty-two studies did not specify the time span of fasting or whether participants had an overnight fast.
Five studies had more than 10% missing data for glucose, four studies for insulin and three studies for HOMA-IR which was not reported to be completely at random. In 13 studies we considered bias to be likely present due to inadequate outcome assessment. The percentage of participants that were loss to follow-up ranged from 0% to 42%. Seven studies had a loss to follow-up of more than 10% and this was unclear in most studies. The p-values of Egger’s test were 0.08 for glucose, <0.01 for insulin and <0.01 for HOMA-IR.
Comparison between glucose, insulin and HOMA-IR
In a random-effect meta-analysis the pooled relative risk of CHD comparing the highest versus the lowest category was 1.52 (95% CI: 1.31, 1.76; I 2 : 62.4%) for glucose, 1.12 (95% CI: 0.92, 1.37;
I 2 : 41.0%) for insulin and 1.64 (95% CI: 1.35, 2.00; I 2 : 0%) for HOMA-IR (Figure 2 and appendix Figure). The pooled relative risks for the association with stroke and CVD, and meta-analyses stratified by sex for studies that provided sex-specific relative risks are summarized in appen- dix Figure.
To enable a direct comparison between CHD and CVD risks for glucose, insulin and HOMA-IR we calculated pooled relative risks for an increase of one standard deviation [18]. We did not in- vestigate the endpoint stroke, because only two studies investigated the association between insulin and stroke. The relative risks per increase of one standard deviation for glucose (1.05 mmol/L), insulin (43.53 pmol/L) and HOMA-IR (2.23 units) are shown in Figure 3. The pooled relative risk of CHD per one standard deviation increase was 1.21 (95% CI: 1.13, 1.30; I 2 : 64.9%) for glucose, 1.04 (95% CI: 0.96, 1.12; I 2 : 43.0%) for insulin and 1.46 (95% CI: 1.26, 1.69; I 2 : 0.0%) for HOMA-IR. The pooled relative risks of CHD for glucose, insulin, and HOMA-IR were all statisti- cally different from each other (p-values: <0.05). The pooled relative risks of CVD for glucose, insulin, and HOMA-IR were not statistically different (p-value: 0.27).
Thirty-three studies provided sex-specific relative risks of CVD. Few studies provided relative risks of CHD or stroke for women and therefore we only investigated sex differences for incident CVD. Women had higher relative risks of CVD per one standard deviation increase of glucose (1.25 (95% CI: 1.11, 1.41; I 2 : 65.0%) versus 1.13 (95% CI: 1.08, 1.18; I 2 : 29.3%); p-value: 0.01) and insulin (1.24 (95% CI: 1.08, 1.44; I 2 : 18.5%) versus 1.06 (95% CI: 0.97, 1.16; I 2 : 60.4%); p-value:
0.03) and lower relative risk of CVD per one standard deviation increase of HOMA-IR (1.37 (95%
CI: 1.05, 1.80; I 2 33.6%) versus 1.41 (95% CI: 1.12, 1.77; I 2 66.5%); p-value: 0.73) (Figure 3). In
sensitivity analyses we excluded studies which had a high risk of bias based on items of the risk
of bias assessment. The results of the meta-analyses were materially unchanged.
.
.
.
Glucose (mmol/L) Baba 2007 Balkau 1998
bBalkau 1998
cBarrett-Connor 1984
dBarrett-Connor 1984
eBrunner 2010 Doi 2010
dDoi 2010
eFord 2004 Girman 2004 Hailpern 2006
fHailpern 2006
gHwang 2009
dHwang 2009
eKhang 2010 Kokubo 2010 Lapidus 1985 Liu 2007 Marin 2006 Preiss 2010 Sarwar 2010 Selvin 2010 Simons 2000
dSimons 2000
eTai 2004 Wang 2007 Wilson 2005 Yarnell 1998
Subtotal (I-squared = 62.4%, p = 0.000) Insulin (pmol/L)
Folsom 1997
dFolsom 1997
eJeppesen 2010 Liu 1992 Nakamura 2010 Nilsson 2003 Orchard 1994 St-Pierre 2005 Wang 2007 Yarnell 1998
Subtotal (I-squared = 41.0%, p = 0.084) HOMA-IR
Hanley 2002 Hedblad 2002 Hwang 2009
dHwang 2009
eJeppesen 2010 Nakamura 2010 Onat 2006 Rundek 2010
Subtotal (I-squared = 0.0%, p = 0.700) Author
≥6.1
avs <6.1 6.9-7.0 vs ≤6.0 6.0-7.0 vs ≤5.3 6.1-6.9 vs 3.9-6.0 6.1-6.9 vs 3.9-6.0
≥7.0 vs <6.1
≥7.0 vs <5.6
≥7.0 vs <5.6
≥6.1 vs <6.1
≥6.1 vs <6.1
≥6.1 vs <6.1
≥6.1 vs <6.1
≥7.0
avs <7.0
≥7.0
avs <7.0
≥5.6 vs <5.6
≥7.0
avs <5.5
≥5.5 vs <5.5
≥7.0
avs ≤5.5
≥7.0 vs <7.0 5.2-6.9 vs ≤4.3
≥7.0 vs <7.0
≥7.0 vs <5.6 5.3-6.0 vs 3.2-4.5 5.3-6.0 vs 3.2-4.5 6.1-6.9 vs ≤5.5
≥6.1 vs <6.1 5.6-6.9 vs <5.6 5.3-7.7 vs ≤4.4
≥100 vs <34
≥100 vs <34
≥49/41 vs <49/41 Not specified 49-507 vs 7-21 146-972 vs ≤139 140-508 vs 12-72
≥85.2 vs <85.2 Not specified
≥71 vs ≤21
4.8-41.7 vs 0.0-1.0
>2.12/1.80 vs ≤2.12/1.80 Not specified
Not specified
≥1.53/1.26 vs <1.53/1.26 1.52-18.73 vs 0.18-0.66
≥2.2455 vs <2.2455 Not specified Comparison
1.59 (0.76, 3.33) 1.42 (0.81, 2.49) 4.48 (1.68, 11.95) 2.60 (1.50, 4.51) 1.00 (0.50, 2.00) 1.29 (0.64, 2.60) 1.29 (0.65, 2.56) 3.83 (1.59, 9.23) 1.17 (0.73, 1.88) 1.22 (0.64, 2.33) 1.23 (0.79, 1.92) 0.66 (0.20, 2.18) 3.58 (1.87, 6.85) 6.99 (1.41, 34.65) 1.24 (0.86, 1.79) 2.28 (1.34, 3.88) 1.80 (0.20, 16.20) 1.81 (1.20, 2.73) 1.52 (1.00, 2.31) 0.87 (0.70, 1.08) 2.37 (1.79, 3.14) 1.29 (1.04, 1.60) 1.08 (0.77, 1.51) 1.52 (1.08, 2.14) 1.51 (0.79, 2.89) 1.25 (0.82, 1.91) 1.60 (1.00, 2.56) 1.39 (1.02, 1.89) 1.52 (1.31, 1.76) 0.49 (0.26, 0.92) 2.06 (0.52, 8.16) 1.44 (1.05, 1.97) 2.30 (0.95, 5.57) 1.85 (0.57, 6.00) 1.03 (0.81, 1.31) 0.94 (0.55, 1.61) 1.21 (0.90, 1.63) 0.89 (0.56, 1.41) 1.27 (0.75, 2.15) 1.12 (0.92, 1.37) 1.54 (0.91, 2.61) 2.18 (1.22, 3.90) 4.33 (1.26, 14.88) 4.04 (0.28, 58.29) 1.50 (1.09, 2.06) 2.03 (0.61, 6.76) 1.43 (0.95, 2.16) 1.77 (0.88, 3.56) 1.64 (1.35, 2.00) Relative Risk (95% CI) 1.59 (0.76, 3.33) 1.42 (0.81, 2.49) 4.48 (1.68, 11.95) 2.60 (1.50, 4.51) 1.00 (0.50, 2.00) 1.29 (0.64, 2.60) 1.29 (0.65, 2.56) 3.83 (1.59, 9.23) 1.17 (0.73, 1.88) 1.22 (0.64, 2.33) 1.23 (0.79, 1.92) 0.66 (0.20, 2.18) 3.58 (1.87, 6.85) 6.99 (1.41, 34.65) 1.24 (0.86, 1.79) 2.28 (1.34, 3.88) 1.80 (0.20, 16.20) 1.81 (1.20, 2.73) 1.52 (1.00, 2.31) 0.87 (0.70, 1.08) 2.37 (1.79, 3.14) 1.29 (1.04, 1.60) 1.08 (0.77, 1.51) 1.52 (1.08, 2.14) 1.51 (0.79, 2.89) 1.25 (0.82, 1.91) 1.60 (1.00, 2.56) 1.39 (1.02, 1.89) 1.52 (1.31, 1.76) 0.49 (0.26, 0.92) 2.06 (0.52, 8.16) 1.44 (1.05, 1.97) 2.30 (0.95, 5.57) 1.85 (0.57, 6.00) 1.03 (0.81, 1.31) 0.94 (0.55, 1.61) 1.21 (0.90, 1.63) 0.89 (0.56, 1.41) 1.27 (0.75, 2.15) 1.12 (0.92, 1.37) 1.54 (0.91, 2.61) 2.18 (1.22, 3.90) 4.33 (1.26, 14.88) 4.04 (0.28, 58.29) 1.50 (1.09, 2.06) 2.03 (0.61, 6.76) 1.43 (0.95, 2.16) 1.77 (0.88, 3.56) 1.64 (1.35, 2.00) Relative Risk (95% CI)
1
.25 .5 1 2 4 8
Figure 2. Random-effect meta-analyses of coronary heart disease risk for the highest category of glucose, insulin or HOMA-IR compared to the lowest category
aor known diabetes was used to define the highest category
b
Paris Prospective Study
c
Helsinki Policemen Study
d
Men
e
Women
f
Glomerular Filtration Rate ≥ 60 ml/min/1.73 m
2g
Glomerular Filtration Rate < 60 ml/min/1.73 m
295% CI, 95% confidence interval; vs, versus; I
2, measure of heterogeneity; HOMA-IR, Homeostasis Model
Assessment Insulin Resistance
2
dIsCussIon
The present meta-analyses showed that fasting glucose, fasting insulin and HOMA-IR were all associated with incident cardiovascular disease in individuals without diabetes. In a standard- ized meta-analysis we found that coronary heart disease risk increased with 46% for an increase of one standard deviation in HOMA-IR concentration compared to an increase of 21% for fast- ing glucose concentration and an increase of 4% for fasting insulin concentration.
To our knowledge, this was the first meta-analysis that directly compared fasting glucose, fasting insulin and HOMA-IR in strength of association with cardiovascular disease.
A number of previous meta-analyses have investigated the association between fasting glucose, fasting insulin or HOMA-IR concentrations and cardiovascular disease by comparing high to low concentrations. Our pooled relative risks of cardiovascular disease (glucose: 1.44, insulin: 1.28, HOMA-IR: 1.44) are within the range of pooled relative risks reported in previous meta-analyses [19, 105-107]. Differences in pooled relative risks between meta-analyses may be, for a large part attributed to different cut-off levels of the exposure, leading to different causal contrasts. Further, differences in design aspects of meta-analyses may explain different pooled relative risks. For example, including studies with only fatal events versus studies with fatal and non-fatal events can result in different pooled RR for glucose, since diabetes seems
Figure 3. Results of random-effect meta-analyses comparing cardiovascular disease risk for an increase of one standard deviation
a
