Are Metabolic Risk Factors One Unified Syndrome? Modeling the

American Journal of Epidemiology
Copyright © 2003 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved
Vol. 157, No. 8
Printed in U.S.A.
DOI: 10.1093/aje/kwg045
Are Metabolic Risk Factors One Unified Syndrome? Modeling the Structure of the
Metabolic Syndrome X
Biing-Jiun Shen1, John F. Todaro1, Raymond Niaura1, Jeanne M. McCaffery1, Jianping
Zhang1, Avron Spiro III2,3, and Kenneth D. Ward4
1
Center for Behavioral and Preventive Medicine, Brown University and the Miriam Hospital, Providence, RI.
Boston University School of Public Health, Boston, MA.
3 Massachusetts Veterans Epidemiology Research and Information Center, Boston Veterans Administration Healthcare
System, Boston, MA.
4 University of Memphis Center for Community Health, Memphis, TN.
2
Received for publication June 24, 2002; accepted for publication November 13, 2002.
The metabolic syndrome, manifested by insulin resistance, obesity, dyslipidemia, and hypertension, is
conceived to increase the risk for coronary heart disease and type II diabetes. Several studies have used factor
analysis to explore its underlying structure among related risk variables but reported different results. Taking a
hypothesis-testing approach, this study used confirmatory factor analysis to specify and test the factor structure
of the metabolic syndrome. A hierarchical four-factor model, with an overarching metabolic syndrome factor
uniting the insulin resistance, obesity, lipid, and blood pressure factors, was proposed and tested with 847 men
who participated in the Normative Aging Study between 1987 and 1991. Simultaneous multigroup analyses were
also conducted to test the stability of the proposed model across younger and older participants and across
individuals with and without cardiovascular disease. The findings demonstrated that the proposed structure was
well supported (comparative fit index = 0.97, root mean square error approximation = 0.06) and stable across
subgroups. The metabolic syndrome was represented primarily by the insulin resistance and obesity factors,
followed by the lipid factor, and, to a lesser extent, the blood pressure factor. This study provides an empirical
foundation for conceptualizing and measuring the metabolic syndrome that unites four related components
(insulin resistance, obesity, lipids, and blood pressure).
blood pressure; body weight; factor analysis, statistical; insulin resistance; lipoproteins
Abbreviation: SD, standard deviation.
The metabolic syndrome is conceived as a clustering of
metabolic risk factors, including visceral obesity, insulin
resistance, hyperglycemia, high blood pressure, and dyslipidemia, that increases the risk for both cardiovascular disease
and type II diabetes (1, 2). Despite evidence for increased
risk among individuals with these characteristics (1, 3–6), it
remains unclear whether this “metabolic syndrome” represents a unified construct reflecting one underlying pathogenic pathway. Further knowledge about the nature of the
metabolic syndrome and the degree of association among its
components will help to inform implementations of preventive and intervention efforts for individuals at risk for cardiovascular disease and type II diabetes.
Several previous studies used factor analysis, mainly principal component analysis, to examine the clustering among
metabolic risk variables (7–15) and mostly yielded a threeor four-factor structure underlying the defining risk
measures (9, 11, 15). These studies have shown that specific
risk variables tend to aggregate together, such that serum
insulin and glucose concentrations (8, 9, 11, 13, 15), blood
lipids (8–11, 13–15), blood pressure (9, 10, 13–15), and
obesity measures (body mass index and waist/hip ratio) (8,
10, 15) form separate factors.
Despite some consensus (9, 12, 13, 15), discrepancies
have also been observed in the number of underlying factors,
which ranged from two (10) to seven (15), and in the combi-
Correspondence to Dr. Biing-Jiun Shen, Department of Psychology, University of Miami, P.O. Box 248185, Coral Gables, FL 33124 (e-mail:
[email protected]).
701
Am J Epidemiol 2003;157:701–711
702 Shen et al.
nations of variables loading on each factor (8). These
discrepancies have posed challenges to interpretation of the
factor structure underlying the metabolic syndrome. The
apparent inconsistencies in past findings may well stem from
the use of principal component factor analysis and its exploratory and subjective nature (16). For example, different findings may arise on the basis of sample selection, variables
included in analysis, number of factors extracted, and rotation method used (17, 18). Another issue in past studies
involves the use of orthogonal rotation, such as varimax, to
obtain “independent” or “uncorrelated” factors (17).
However, applying orthogonal rotation may unwarrantedly
restrict correlations between factors and derive statistically
independent but artificial factors that do not correspond to
actual physiologic processes (19). Finally, the results of
multiple uncorrelated factors comprising the metabolic
syndrome produced by principal component factor analysis
studies also seem contradictory to the concept of a syndrome
that is thought to unite related symptoms under one disorder
entity.
Because of these observed differences and concerns raised
by researchers, it has been suggested to use confirmatory
factor analysis to examine the structure of metabolic
syndrome (17). The confirmatory factor analysis approach
provides some advantages and is complementary to the use
of principal component factor analysis (20, 21). First, with
confirmatory factor analysis, a hypothesized model detailing
the relations among variables and factors is specified and
subjected to examination, thus permitting a hypothesistesting rather than a data-driven approach. Past studies, in
fact, provide excellent exploratory knowledge on which
models can be built and tested with confirmatory factor analysis. Second, questions such as “if there exists a unified
syndrome underlying various metabolic disorders” and “if
components of this syndrome are correlated” can be empirically examined by testing corresponding models. Third, the
stability of the hypothesized factor structure can be tested
across subgroups with different characteristics (e.g., age or
diagnosis) to ensure its generalizability and practical use in
diverse populations (20).
In this study, confirmatory factor analysis was used to
evaluate hypothesized models that delineate the clustering
among variables characterizing the metabolic syndrome. On
the basis of theoretical conceptualization and empirical
evidence provided by previous principal component factor
analysis studies, the main analysis proposed and tested a
hierarchical four-factor model consisting of measures of
insulin resistance, obesity, lipids, and blood pressure. We
also examined whether empirical data support the hypothesis
that there exists a single unifying factor representing the
common pathway underlying all four metabolic factors.
Lastly, to investigate the robustness of the proposed factor
structure, simultaneous subgroup analyses were carried out
to detect possible structural differences across younger
versus older groups and across individuals with and without
cardiovascular disease.
MATERIALS AND METHODS
Participants
Participants were 847 men who were recruited into the
Normative Aging Study, a longitudinal study designed to
examine biomedical and psychosocial changes in the normal
aging process. This study followed an initial cohort of 2,280
men living in the greater Boston, Massachusetts, area who
were between 21 and 80 years old at enrollment between
1961 and 1970. Volunteers received regular examinations
every 3 (aged 52 years or older) or 5 (aged less than 52)
years. Additional details of the sampling procedure and
admission criteria have been described elsewhere (22). Only
1,081 participants who were examined between the years of
1987 and 1991, during which time measurements of serum
insulin and waist/hip ratio were taken, were included in this
study. Among them, 847 provided complete data on the variables examined in this study and were included in analyses.
Procedures
Participants were instructed to refrain from eating or
drinking after midnight and to avoid smoking after 8 p.m. the
night before examination. The examination included blood
pressure measurement, blood work, an anthropometric evaluation, and assessment of health behaviors by standardized
questionnaires. The first blood sample was drawn at 8 a.m. to
measure fasting insulin and glucose. Another blood sample
was taken 2 hours after the participant orally consumed a
100-g glucose load in order to obtain postchallenge insulin
and glucose concentrations.
Measures
Ten measures representing the metabolic syndrome were
obtained, including fasting insulin, postchallenge insulin,
fasting glucose, postchallenge glucose, body mass index,
waist/hip ratio, high density lipoprotein cholesterol, triglycerides, systolic blood pressure, and diastolic blood pressure.
Blood lipids. Total cholesterol, high density lipoprotein
cholesterol, low density lipoprotein cholesterol, and triglycerides were obtained from analysis of serum samples. Serum
cholesterol was assayed enzymatically (SCALVO Diagnostics, Wayne, New Jersey). The high density lipoprotein
cholesterol fraction was measured in the supernatant after
precipitation of the low density lipoprotein cholesterol and
very low density lipoprotein fractions with dextran sulfate
and magnesium, using the Abbott Biochromatic Analyzer
100 (Abbott Laboratories, South Pasadena, California). The
triglyceride concentration was measured with a Dupont
ACA discrete clinical analyzer (Biomedical Products
Department, Dupont Company, Wilmington, Delaware).
Fasting and postchallenge serum insulin and glucose.
Two blood samples were analyzed to obtain fasting and
postchallenge insulin and glucose values. The serum insulin
concentration was determined by a solid-phase 125I-labeled
radioimmunoassay (Diagnostic Products Corporation, Los
Angeles, California). The serum glucose concentration was
measured in duplicate on an autoanalyzer by the hexokinase
method (23).
Am J Epidemiol 2003;157:701–711
Structure of Metabolic Syndrome 703
FIGURE 1. Proposed hierarchical four-factor structure of the metabolic syndrome, with χ2 = 102.10 (df = 25, n = 847, p < 0.01), comparative fit
index = 0.97, average absolute standardized residual = 0.02, and root mean square error approximation = 0.06. Standardized parameter estimates representing factor loadings are shown on paths. All coefficients are significant at p = 0.01 for the two-tailed test. To maintain presentation
clarity, residual terms are not shown. PC insulin, postchallenge insulin; PC glucose, postchallenge glucose; HDL, high density lipoprotein; BP,
blood pressure.
Blood pressure. Systolic blood pressure and fifth-phase
diastolic blood pressure were measured to the nearest 2
mmHg with a standard mercury sphygmomanometer with a
14-cm cuff. Both left and right arm pressures were recorded
with the participant in a sitting position, followed by right
arm pressures taken in a supine position and then followed
30 seconds later by a second reading of right arm pressures
taken in a standing position. The palpatory method was used
to check auscultatory systolic readings. The means of all
systolic and diastolic readings were used in analyses.
Am J Epidemiol 2003;157:701–711
Anthropometric measures. Weight was measured to the
nearest 0.5 pound (0.2268 kg) on a standard hospital scale
with the participant dressed in undershorts and socks. It was
later converted to kilograms. Height was measured to the
nearest 0.1 inch (0.254 cm) with the participant standing in
bare feet against a wall, and this value was then converted to
meters. Body mass index was computed as kg/m2. Abdomen
circumference was measured in centimeters at the level of
the umbilicus with the participant standing. Hip circumference was measured in centimeters at the greatest protrusion
704 Shen et al.
FIGURE 2. An alternative factor structure of the metabolic syndrome, with χ2 = 631.44 (df = 32, n = 847, p < 0.01), comparative fit index = 0.74,
and root mean square error approximation = 0.15. All coefficients are significant at p = 0.01 for the two-tailed test. PC insulin, postchallenge
insulin; PC glucose, postchallenge glucose; HDL, high density lipoprotein; BP, blood pressure.
of the buttocks. The waist/hip ratio was calculated by
dividing the abdomen circumference by the hip circumference.
Formulation of the factor structure of the metabolic
syndrome
On the basis of theoretical conceptualization, measures
belonging to the same physiologic sources, such as glucose
metabolism, obesity, lipids, or blood pressure, should share
more commonality than do measures from different
processes. As reviewed previously, prior research did largely
support a three- to four-factor solution with measures
divided by their respective categories (7–15). Therefore, we
proposed a hierarchical four-factor model with four firstorder factors comprising the major components of the metabolic syndrome (insulin resistance, obesity, lipid, and blood
pressure) and a second-order factor reflecting the syndrome
itself (figure 1). Accordingly, the insulin resistance factor
was defined by fasting insulin, postchallenge insulin,
glucose, and postchallenge glucose (8, 11, 13, 15), the
obesity factor manifested by body mass index and waist/hip
ratio (10), the lipid factor composed of high density lipoprotein cholesterol and triglycerides (8), and the blood pressure
factor represented by systolic blood pressure and diastolic
blood pressure (9, 10, 13, 15). A higher order factor, the
metabolic syndrome, was created to reflect the common
process underlying all these risk factors. Finally, residual
variance terms specific to each of the measures were specified in the model, which represented the variance not
accounted for by the common factors, including measurement errors.
Preliminary analysis found that two pairs of residual variances were highly correlated, one between glucose and
Am J Epidemiol 2003;157:701–711
Structure of Metabolic Syndrome 705
FIGURE 3. An alternative factor structure of the metabolic syndrome, with χ2 = 101.75 (df = 27, n = 847, p < 0.01), comparative fit index = 0.97,
and root mean square error approximation = 0.06. All coefficients are significant at p = 0.01 for the two-tailed test. PC insulin, postchallenge
insulin; PC glucose, postchallenge glucose; HDL, high density lipoprotein; BP, blood pressure.
postchallenge glucose and the other between postchallenge
insulin and postchallenge glucose. Correlated residuals indicate the presence of “small factors” or common variances
shared by some but not all measures under a factor (24). The
addition of these two correlations was well justified because
glucose and postchallenge glucose were both measures of
glucose concentration and because postchallenge insulin and
postchallenge glucose were assessed at the same time.
In summary, there were five latent factors and 10
measured variables in this model (figure 1). The model
contained 30 parameters to be estimated, including 14 path
coefficients (loadings) between factors and measures and
between the first-order and second-order factors, 14 residual
Am J Epidemiol 2003;157:701–711
variances, and two covariances between residuals. There
was an approximate 85:1 case:variable ratio, and there was
an approximate 30:1 case:parameter ratio. Both were excellent for conducting confirmatory factor analysis.
We also tested two alternative metabolic syndrome factor
structures to evaluate their goodness of fit. One assumed a
common factor that directly underlies all measured metabolic risk variables without the presence of separate insulin
resistance, obesity, lipid, and blood pressure factors (figure
2). The other postulated that the four major metabolic factors
were correlated, with the interfactor correlations representing the common association among the components of
metabolic syndrome (figure 3).
706 Shen et al.
Data analysis
The model in figure 1 was subjected to confirmatory factor
analysis based on Bentler and Weeks’ model (25) using the
EQS program (20). Prior to analysis, the normality assumptions of the data were examined, and variables with high
skewness or kurtosis were transformed with a natural log
function. The maximum likelihood method with robust
statistics (Satorra-Bentler scaled test statistics and robust
standard errors) was used for parameter estimations and
significance tests. Tests of significance were set at p = 0.01
for two-tailed tests.
The model was evaluated in three ways. First, the χ2 test
was used to evaluate the congruency between the hypothesized model and empirical data from the sample. Second,
because χ2 tests are highly sensitive to sample size and often
result in statistically significant but empirically trivial differences, three model fit indices were also used to evaluate the
model. These were comparative fit index, average absolute
standardized residuals, and root mean square error of
approximation. Finally, the residual matrix was inspected for
any large deviation from zero that was indicative of discrepancy between parameter estimates and empirical data.
RESULTS
Sample characteristics
The demographic and medical backgrounds of the participants are presented in table 1. Participants in this study were
men between 42 and 83 years of age with a mean of 60.70
(standard deviation (SD) = 7.79) years. The majority had at
least a high school education, and 28.8 percent had
completed college or graduate school. With regard to
medical history, 44 percent had hypertension, and 19.6
TABLE 1. Participants’ demographic characteristics and
medical history (n = 847), Normative Aging Study, 1987–1991
Characteristic
%
Education
Less than high school
9.1
High school
23.9
Some college
38.2
College or postgraduate study
28.8
Medical history
Myocardial infarction
10.4
Ischemic heart disease
11.3
Angina pectoris
13.1
Coronary heart disease
19.6
Hypertension
44.4
percent had documented coronary heart disease, including
myocardial infarction, angina pectoris, or chronic ischemia.
The descriptive statistics of the 10 metabolic risk variables
are shown in table 2. Participants, in general, were somewhat
overweight with an average body mass index of 26.6 (SD =
3.4) and a waist/hip ratio of 0.98 (SD = 0.05). On average,
their blood pressure was within the normal range, with mean
systolic blood pressure = 129.0 (SD = 15.9) mmHg and diastolic blood pressure = 78.3 (SD = 8.7) mmHg. The average
high density lipoprotein cholesterol, triglycerides, and
fasting insulin appeared normal and were measured at 48.0
(SD = 12.0) mg/dl, 145.9 (SD = 77.4) mg/dl, and 11.25 (SD =
6.8) milliunits/ml, respectively. Table 3 presents the Pearson
correlations between each pair of the variables in analysis.
High correlations were observed within clusters of variables,
such as among insulin and glucose measures, between lipids,
between obesity indices, and between blood pressures.
TABLE 2. Physiologic and anthropometric characteristics of the participants (n = 847), Normative
Aging Study, 1987–1991
Original value
Measure
Fasting insulin (mU†/liter)
Postchallenge insulin (mU/liter)
Transformed value*
Mean
Standard
deviation
Mean
Standard
deviation
11.3
6.8
0.99
0.23
66.7
65.1
1.69
0.34
Fasting glucose (mg/dl)
101.9
16.0
4.61
0.13
Postchallenge glucose (mg/dl)
113.4
44.1
4.68
0.31
26.6
3.4
2.11
0.21
Body mass index (kg/m2)
Waist/hip ratio
High density lipoprotein (mg/dl)
0.98
0.05
48.0
12.0
Triglycerides (mg/dl)
145.9
77.4
Systolic blood pressure (mmHg)
129.0
15.9
Diastolic blood pressure (mmHg)
78.3
8.7
* Fasting insulin, postchallenge insulin, glucose, postchallenge glucose, and triglycerides were natural log
transformed in analysis.
† mU, milliunits.
Am J Epidemiol 2003;157:701–711
Structure of Metabolic Syndrome 707
TABLE 3. Pearson’s correlations between the metabolic risk variables examined in the study (n = 847), Normative Aging Study,
1987–1991
Measure
INS*
PCINS*
GLU*
PCGLU*
BMI*
WHR*
HDL*
TRG*
SBP*
INS†
Correlation coefficient
p value
PCINS†
Correlation coefficient
0.65
p value
0.000
GLU†
Correlation coefficient
0.27
0.26
p value
0.000
0.000
Correlation coefficient
0.29
0.55
0.53
p value
0.000
0.000
0.000
Correlation coefficient
0.45
0.37
0.21
0.21
p value
0.000
0.000
0.000
0.000
Correlation coefficient
0.33
0.29
0.11
0.15
0.46
p value
0.000
0.000
0.001
0.000
0.000
PCGLU†
BMI
WHR
HDL
Correlation coefficient
p value
–0.22
–0.21
–0.13
–0.10
–0.27
–0.20
0.000
0.000
0.000
0.004
0.000
0.000
Correlation coefficient
0.29
0.36
0.16
0.22
0.24
0.26
p value
0.000
0.000
0.000
0.000
0.000
0.000
Correlation coefficient
0.17
0.23
0.10
0.21
0.10
0.12
p value
0.000
0.000
0.005
0.000
0.002
0.000
TRG†
–0.47
0.000
SBP
–0.02
0.482
0.15
0.000
DBP*
Correlation coefficient
0.17
0.17
p value
0.000
0.000
–0.01
0.787
0.09
0.19
0.11
0.008
0.000
0.001
–0.01
0.785
0.12
0.57
0.000
0.000
* INS, fasting insulin; PCINS, postchallenge insulin; GLU, fasting glucose; PCGLU, postchallenge glucose; BMI, body mass index; WHR,
waist/hip ratio; HDL, high density lipoprotein; TRG, triglycerides; SBP, systolic blood pressure; DBP, diastolic blood pressure.
† Transformed with a natural log function.
Model estimation and evaluation
Factor structure of the metabolic syndrome
The testing of the proposed model demonstrated favorable
results. Although the χ2 test (χ2 = 102.10, df = 25, n = 847,
p < 0.01) indicated a difference between the estimated and
observed variance-covariance matrices, this was likely due
to the large sample size. The comparative fit index of the
model was 0.97, with the average absolute standardized
residual = 0.02, the root mean square error approximation =
0.06, and the elements in the residual variance-covariance
matrix being small and distributed around zero, all indicating
a high goodness of fit of the factor structure. In summary, the
findings demonstrated that empirical data provided strong
support for the proposed metabolic syndrome factor structure. The model with parameter estimates is presented in
figure 1.
The results showed that the metabolic syndrome could be
summarized by an overarching factor’s subsuming four
component factors, each sharing varying degrees of associations with the overall common factor. Each component
factor was further defined by its respective metabolic risk
variables. As illustrated in figure 1, the metabolic syndrome
factor was strongly represented by the insulin resistance
factor (standardized coefficient as factor loading = 0.83, p <
0.01) and the obesity factor (loading = 0.80, p < 0.01),
moderately by the lipid factor (loading = 0.59, p < 0.01), and
only modestly by the blood pressure factor (loading = 0.33,
p < 0.01). In other words, the metabolic syndrome factor
explained 69 percent of the variance in the insulin resistance
Am J Epidemiol 2003;157:701–711
708 Shen et al.
factor, 64 percent for the obesity factor, 35 percent for the
lipid factor, and 11 percent for the blood pressure factor.
As for each component factor, the insulin resistance
factor was defined mainly by fasting insulin and postchallenge insulin and, to a lesser degree, by glucose and postchallenge glucose (loadings = 0.83, 0.78, 0.34, and 0.37,
respectively; all p < 0.01). The obesity factor was well
represented by the body mass index and the waist/hip ratio
(loadings = 0.77 and 0.60; both p < 0.01). The lipid factor
was reflected by triglycerides and high density lipoprotein
cholesterol (loadings = 0.78 and –0.60; both p < 0.01).
Finally, the blood pressure factor was equally manifested
by systolic blood pressure and diastolic blood pressure
(loadings = 0.75 and 0.76; both p < 0.01).
Alternative models
The test of single-factor structure (figure 2) revealed that
this model performed poorly and received little empirical
support, with χ2 = 631.44 (df = 32, p < 0.0001), the comparative fit index = 0.74, and the root mean square error approximation = 0.15. On the other hand, the correlated four-factor
model appeared comparable with the first model tested, with
χ2 = 101.75 (df = 27, p < 0.01), the comparative fit index =
0.97, and the root mean square error approximation = 0.06.
This result was not surprising because the common association among factors in this model was represented by six
paired correlations instead of one second-order factor. In this
model, varying degrees of interfactor correlations were
observed. The insulin resistance factor was strongly correlated with the obesity factor (r = 0.66) and moderately with
the lipid factor (r = 0.49). The obesity factor also had a
moderately high association with the lipid factor (r = 0.48).
The blood pressure factor, although significantly correlated
with others, appeared to have weaker relations with the
insulin resistance (r = 0.28), obesity (r = 0.26), and lipid (r =
0.17) factors.
The use of correlated factors is similar to creating a
second-order factor to represent the commonality among
factors. Constructing a second-order factor, however, allows
the testing of the single-factor hypothesis directly. In fact,
the interfactor correlations in figure 3 are almost identical to
the products of the two loadings of corresponding first-order
factors on the second-order factor. For example, the interfactor correlation between the insulin resistance and obesity
factors in figure 3 is 0.66, which is the same as the product of
loadings of the insulin resistance factor = 0.83 and the
obesity factor = 0.80 on the metabolic syndrome factor in
figure 1.
Subgroup analyses
Simultaneous multigroup analyses (20) were conducted to
test the stability and generalizability of the proposed hierarchical four-factor structure and to explore possible differences across subgroups. First, the structure was examined
across groups of individuals with or without documented
cardiovascular disease, with cardiovascular disease defined
by diagnosed hypertension, angina pectoris, ischemia, coronary heart disease, and myocardial infarction. Second, the
model was examined across younger (less than 60 years of
age) and older participants.
In a comparison of individuals with and without cardiovascular disease, the factor structure was identical except for
the strength of the path between the body mass index and the
obesity factor (χ2diff = 6.71, df = 1, p < 0.01; final model χ2 =
141.00, df = 63, n = 847, p < 0.01, comparative fit index =
0.97, root mean square error approximation = 0.04). It was
found that body mass index had a slightly stronger association with the obesity factor among individuals with cardiovascular disease (loading = 0.79, p < 0.01) than among those
without cardiovascular disease (loading = 0.75, p < 0.01).
Analyses for age groups also revealed only a small difference (χ2diff = 6.58, df = 1, p < 0.01; final model χ2 = 134.46,
df = 62, n = 847, p < 0.01, comparative fit index = 0.97, root
mean square error approximation = 0.04), such that the path
between triglycerides and the lipid factor was stronger for
younger (loading = 0.84, p < 0.01) than for older (loading =
0.71, p < 0.01) individuals. No subgroup differences were
found in the paths between the metabolic syndrome factor
and its component factors, indicating that proposed structure
was highly stable across subgroups.
DISCUSSION
This study examined the factor structure of metabolic
syndrome using confirmatory factor analysis in a large
epidemiologic data set from the Normative Aging Study.
Unlike the principal component factor analysis techniques
used by previous studies, confirmatory factor analysis
allowed for the examination of an a priori, hypothetical
factor structure based on past findings or theoretical rationales. The results of these analyses supported a hierarchical,
four-factor structure in which the four factors, insulin resistance, obesity, lipid, and blood pressure, were united by a
higher-order common factor representing the metabolic
syndrome. Moreover, when factors were permitted to correlate, significant associations did emerge between factors.
Our analyses showed that the source of these associations
appeared to stem from the common metabolic syndrome
factor. They suggested that the “uncorrelated” factors
reported in earlier studies were likely products of the rotation
techniques applied. Furthermore, the observed hierarchical
factor structure was shown to be stable across relevant
subgroups, including older versus younger individuals and
participants with and without cardiovascular disease. To our
knowledge, no studies have examined the hypothesis of a
unified syndrome directly or attempted to validate its structural stability across various subgroups.
Varying degrees of association were observed between the
metabolic syndrome factor and its four component factors.
Insulin resistance and obesity appeared to be the two most
essential features of the syndrome, followed by hyperlipidemia, and, to a lesser extent, hypertension. The strong association between insulin resistance and obesity is consistent
with the observation in several previous studies in which
insulin resistance and obesity loaded on the same factor (8–
10, 12, 13, 15). The consistency of this relation across
studies is compelling and supports the notion of a synergistic
physiologic process or common pathway underlying
Am J Epidemiol 2003;157:701–711
Structure of Metabolic Syndrome 709
impaired glucose metabolism and obesity. There is evidence
suggesting that visceral obesity may be linked to decreased
glucose tolerance with hyperinsulinemia, leading to a reduction in insulin-mediated glucose uptake (2, 26, 27). It has
also been suggested that adipose cells, via their tendency to
increase the production of free fatty acids, may slow down
glucose uptake, stimulate hepatic glucose production, and
suppress pancreatic insulin secretion (28). These interrelated
mechanisms may explain the strong association between
insulin resistance and obesity and may constitute the basic
pathogenetic pathways of the metabolic syndrome for coronary heart disease and diabetes.
Lipid measures had a moderate contribution to the metabolic syndrome factor and clear associations with other
components of the syndrome including insulin resistance
and obesity. The moderate associations suggest that mechanisms relating lipids to insulin resistance and obesity are
more complex and may involve more intermediate
processes. It is well known that free fatty acids play an
important role in lipid metabolism and are a prominent
source of triglycerides stored in adipose tissue (2, 29). In
patients exhibiting metabolic disorders, particularly those
with insulin resistance, the suppression of free fatty acid
release from adipose tissue is impaired, providing an important ingredient for the liver to synthesize triglycerides and
very low density lipoprotein (30). In turn, increased lipid
production by the liver is associated with an overall increase
in total glucose production and may place additional stress
on beta-cell function and insulin production. These interrelated mechanisms among insulin resistance, lipids, and
glucose production, therefore, provide support for our
model, which indicates that metabolic syndrome risk factors
work in synergy and, ultimately, come together to influence
the etiology of chronic diseases. The mechanisms linking
lipid metabolism, obesity, and insulin dependence, nevertheless, are still controversial and require further research (2).
Although blood pressure was evidenced to be an element
of the metabolic syndrome in our analyses, the strength of
this relation was weaker than with other components. This is
consistent with prior studies reporting tenuous relations
between blood pressure and other risk variables (10, 18).
Research has suggested that insulin-related metabolic abnormality (insulin resistance and hyperinsulinemia) may give
rise to the pathogenesis of hypertension via a number of
mechanisms, including increased renal sodium and water
retention, plasma noradrenaline, and sympathetic nervous
system activity (31–33). The association between hypertension and the metabolic syndrome, however, has been challenged (2, 14, 34), and such an association is not always
evident in African Americans and Pima Indians (35). Our
results showed that the majority of the variance in blood
pressure was not attributable to the metabolic syndrome,
indicating that blood pressure may be related to the
syndrome only secondarily to other mechanisms.
Furthermore, the results are largely consistent with the
National Cholesterol Education Program’s guideline for the
metabolic syndrome (36). Our analyses further suggest that
the body mass index may be a criterion comparable with
waist circumference and that blood pressure shows a smaller
Am J Epidemiol 2003;157:701–711
association with the syndrome than others. Triglycerides and
high density lipoprotein cholesterol are represented as part of
the same risk component. Thus, the inclusion of both as
criteria may be redundant. For determination of the exact
number of risk factors required for diagnosing the metabolic
syndrome, more prospective research is needed to examine
the associations between each component with diabetes and
cardiovascular disease endpoints. Nonetheless, our findings
strongly indicate that insulin resistance, obesity, and an unfavorable lipid profile are essential features and should be
represented in the diagnostic guideline.
There are some limitations to the present study that may
illuminate directions for future research. First, participants in
this study were primarily older, Caucasian males, which may
limit the generalizability of the findings. Given that the prevalence and course of cardiovascular disease differ across
genders and ethnic groups (37), it is plausible that the factor
structure of metabolic syndrome may manifest, at least
partially, as a function of sample characteristics. Future
research should corroborate if the same structure can be
replicated among women, youths, and minority members.
Furthermore, in this study, we included only measures that
are most often associated with the metabolic syndrome.
Recent studies have expanded the concept of the metabolic
syndrome to include other physiologic variables such as uric
acid, inflammation, procoagulation, and vitamin K-dependent protein (10, 15). Future studies should investigate the
influence of these elements on the factor structure of the
syndrome.
Another limitation involved the use of a cross-sectional
design to examine the relations among metabolic risk variables and factors. Although this is similar to most previous
studies examining the factor structure of the metabolic
syndrome, cross-sectional analyses represent only a snapshot of a complicated physiologic system at a single point in
time. These risk variables and factors, nevertheless, are
likely to be linked by a series of progressive and reciprocal
mechanisms. Future research should examine the longitudinal relations among metabolic risk factors to advance our
understanding of the development of coronary heart disease
and diabetes. For example, it would be interesting to investigate whether obesity precedes hyperinsulinemia, which then
leads to dyslipidemia, hypertension, and eventually coronary
heart disease, or if some other etiology exists. Applying a
modeling approach with a longitudinal design can be instrumental to testing these hypotheses.
Finally, there is a growing body of literature examining the
relation between the metabolic syndrome and psychosocial
characteristics (38, 39). Another future direction is to
explore the impact of psychosocial characteristics (e.g.,
socioeconomic status, psychologic traits, lifestyle behaviors)
on the syndrome. To better plan for targeted prevention and
treatment in the general population, a better understanding of
which psychosocial groups are at higher risk for the development and exacerbation of the metabolic syndrome will
contribute to the development of innovative biopsychosocial
approaches that maximize the effectiveness of public health
interventions.
710 Shen et al.
ACKNOWLEDGMENTS
The Veterans Administration Normative Aging Study is
supported by the Cooperative Studies Program/ERIC, US
Department of Veterans Affairs, and is a research component of the Massachusetts Veterans Epidemiology Research
and Information Center. Some of the data analyzed in this
project were obtained with support provided by grants
HL37871 and AG02287.
REFERENCES
1. Kissebah AH, Krakower GR. Regional adiposity and morbidity. Physiol Rev 1994;74:761–811.
2. Timar O, Sestier F, Levy E. Metabolic syndrome X: a review.
Can J Cardiol 2000;16:779–89.
3. Bjorntorp P. “Portal” adipose tissue as a generator of risk factors for cardiovascular disease and diabetes. Arteriosclerosis
1990;10:493–6.
4. Bjorntorp P. Visceral fat accumulation: the missing link
between psychosocial factors and cardiovascular disease? J
Intern Med 1991;230:195–201.
5. DeFronzo RA, Ferrannini E. Insulin resistance. A multifaceted syndrome responsible for NIDDM, obesity, hypertension, dyslipidemia, and atherosclerotic cardiovascular
disease. Diabetes Care 1991;14:173–94.
6. Wing RR, Matthews KA, Kuller LH, et al. Waist to hip ratio
in middle-aged women. Associations with behavioral and psychosocial factors and with changes in cardiovascular risk factors. Arterioscler Thromb 1991;11:1250–7.
7. Donahue RP, Bean JA, Donahue RD, et al. Does insulin resistance unite the separate components of the insulin resistance
syndrome? Evidence from the Miami Community Health
Study. Arterioscler Thromb Vasc Biol 1997;17:2413–17.
8. Edwards KL, Austin MA, Newman B, et al. Multivariate analysis of the insulin resistance syndrome in women. Arterioscler
Thromb 1994;14:1940–5.
9. Lempiainen P, Mykkanen L, Pyoralak K, et al. Insulin resistance syndrome predicts coronary heart disease events in elderly nondiabetic men. Circulation 1999;100:123–8.
10. Leyva F, Godsland IF, Worthington M, et al. Factors of the
metabolic syndrome: baseline interrelationships in the first
follow-up cohort of the HDDRISC Study (HDDRISC-1).
Heart disease and diabetes risk indicators in a screened cohort.
Arterioscler Thromb Vasc Biol 1998;18:208–14.
11. Leyva F, Godsland IF, Ghatei M, et al. Hyperleptinemia as a
component of a metabolic syndrome of cardiovascular risk.
Arterioscler Thromb Vasc Biol 1998;18:928–33.
12. Lindblad U, Langer RD, Wingard DL, et al. Metabolic syndrome and ischemic heart disease in elderly men and women.
Am J Epidemiol 2001;153:481–9.
13. Pyorala M, Miettinen H, Halonen P, et al. Insulin resistance
syndrome predicts the risk of coronary heart disease and
stroke in healthy middle-aged men: the 22-year follow-up
results of the Helsinki Policemen Study. Arterioscler Thromb
Vasc Biol 2000;20:538–44.
14. Meigs JB, D’Agostino RB Sr, Wilson PW, et al. Risk variable
clustering in the insulin resistance syndrome. The Framingham Offspring Study. Diabetes 1997;46:1594–600.
15. Sakkinen PA, Wahl P, Cushman M, et al. Clustering of procoagulation, inflammation, and fibrinolysis variables with metabolic factors in insulin resistance syndrome. Am J Epidemiol
2000;152:897–907.
16. Farbrigar LR, Wegener DT, MacCallum RC, et al. Evaluating
the use of exploratory factor analysis in psychological
research. Psychol Methods 1999;4:272–99.
17. Hu FB. Re: “Clustering of procoagulation, inflammation, and
fibrinolysis variables with metabolic factors in insulin resistance syndrome.” (Letter). Am J Epidemiol 2001;153:717–18.
18. Meigs JB. Invited commentary: insulin resistance syndrome?
Syndrome X? Multiple metabolic syndrome? A syndrome at
all? Factor analysis reveals patterns in the fabric of correlated
metabolic risk factors. Am J Epidemiol 2000;152:908–11; discussion 912.
19. Comrey AL, Lee HB. A first course in factor analysis. Hillsdale, NJ: LEA, 1992.
20. Bentler PM. EQS structural equations program manual.
Encino, CA: Multivariate Software, 1995.
21. Long JS. Confirmatory factor analysis: a preface to LISREL.
Beverly Hills, CA: Sage, 1983.
22. Bossé R, Ekerdt DJ, Silbert JE. The Veteran Administration
Normative Aging Study. In: Mednick SA, Harway M, Finello
KM, eds. Handbook of longitudinal research. Vol 2. Teenage
and adult cohorts. New York, NY: Praeger, 1984:273–95.
23. Leon LP, Chu DK, Stasiw RP. New, more specific methods
for the SMA 12/60 multichannel biochemical analyzer:
advances in automated analysis. Tarrytown, NY: Mediad,
1977.
24. Hoyle RH. Structural equation modeling: concepts, issues,
and applications. Thousand Oaks, CA: Sage, 1995.
25. Bentler PM, Weeks DG. Linear structural equations with
latent variables. Psychometrika 1980;45:289–308.
26. Reaven GM. Syndrome X: 6 years later. J Intern Med Suppl
1994;736:13–22.
27. Lemieux S, Despres JP. Metabolic complications of visceral
obesity: contribution to the aetiology of type 2 diabetes and
implications for prevention and treatment. Diabetes Metab
1994;20:375–93.
28. Peiris AN, Struve MF, Mueller RA, et al. Glucose metabolism
in obesity: influence of body fat distribution. J Clin Endocrinol Metab 1988;67:760–7.
29. Reaven GM. Banting Lecture 1988. Role of insulin resistance
in human disease. Diabetes 1988;37:1595–607.
30. Reynisdottir S, Angelin B, Langin D, et al. Adipose tissue
lipoprotein lipase and hormone-sensitive lipase. Contrasting
findings in familial combined hyperlipidemia and insulin
resistance syndrome. Arterioscler Thromb Vasc Biol 1997;17:
2287–92.
31. Ferrannini E, Buzzigoli G, Bonadonna R, et al. Insulin resistance in essential hypertension. N Engl J Med 1987;317:350–
7.
32. Pollare T, Lithell H, Berne C. Insulin resistance is a characteristic feature of primary hypertension independent of obesity.
Metabolism 1990;39:167–74.
33. Rowe JW, Young JB, Minaker KL, et al. Effect of insulin and
glucose infusions on sympathetic nervous system activity in
normal man. Diabetes 1981;30:219–25.
34. Zavaroni I, Mazza S, Dall’Aglio E, et al. Prevalence of hyperinsulinaemia in patients with high blood pressure. J Intern
Med 1992;231:235–40.
35. Saad MF, Lillioja S, Nyomba BL, et al. Racial differences in
the relation between blood pressure and insulin resistance. N
Engl J Med 1991;324:733–9.
36. Expert Panel on Detection, Evaluation, and Treatment of High
Blood Cholesterol in Adults. Executive summary of the third
report of the National Cholesterol Education Program (NCEP)
Expert Panel on Detection, Evaluation, and Treatment of High
Blood Cholesterol in Adults (adult treatment panel III).
JAMA 2001;285:2486–97.
Am J Epidemiol 2003;157:701–711
Structure of Metabolic Syndrome 711
37. American Heart Association. 2001 heart and stroke statistical
update. Dallas, TX: American Heart Association, 2001.
38. Niaura R, Banks SM, Ward KD, et al. Hostility and the metabolic syndrome in older males: the Normative Aging Study.
Am J Epidemiol 2003;157:701–711
Psychosom Med 2000;62:7–16.
39. Vitaliano PP, Scanlan JM, Zhang J, et al. A path model of
chronic stress, the metabolic syndrome, and coronary heart
disease. Psychosom Med 2002;64:418–35.