Defining the Metabolic Syndrome Construct: Multi-Ethnic Study of Atherosclerosis (MESA) cross-sectional analysis

The design of MESA has been previously described (12). The study was approved by institutional review boards of all participating centers. At baseline (2000–2002), 6,814 white, African-American, Chinese, or Hispanic men and women who were free of clinical cardiovascular disease were recruited. The majority (3,199 men, 3,581 women) had complete data required for classification by the modified National Cholesterol Education Program Adult Treatment Panel III definition (10,11) for these analyses. Subjects completed standardized medical history questionnaires. Fasting blood measurements of glucose, insulin, and a lipid panel were performed. Umbilicus-level waist circumference was measured using a standard tape measure with 4 oz tension. Resting seated blood pressure was measured three times using a Dinamap model Pro 100 automated oscillometric sphygmomanometer (Critikon, Tampa, FL). The average of the last two measurements was used in analysis. The metabolic syndrome is the presence of three or more of the components (10,11): 1) waist ≥102 (men) or ≥88 cm (women), 2) triglycerides ≥150 mg/dl, 3) HDL cholesterol ≤40 (men) or ≤50 mg/dl (women), 4) blood pressure ≥130/85 mmHg or on antihypertensive treatment, and 5) fasting plasma glucose ≥100 mg/dl or on anti-hyperglycemic treatment.

HOMA was calculated as fasting plasma insulin (μU/ml) × glucose (mmol/l)/22.5 (9). The maximum intimal-medial thickness (IMT) was measured at 12 bilateral locations for the internal carotid arteries (IIMT) and four locations for the common carotid arteries (CIMT) using B-mode ultrasound and averaged.

Analyses were stratified by sex, because the criteria for the metabolic syndrome are sex specific.

Association of each component with HOMA was tested using nonparametric rank-sum tests. Sensitivity analyses excluded individuals taking antidiabetic medications. The association of HOMA with the number of co-prevalent components was tested using the Kusick nonparametric trend test. Receiver-operating characteristic analysis for the metabolic components count used the upper quartile of HOMA to define insulin resistance (13).

The prevalence of each metabolic component was calculated by the number of other co-prevalent components. Trends in prevalences were tested by the Kruskal and Goodman γ statistic. Expected co-prevalences of 0, 1, … 5 components were calculated based on the marginal prevalences of each metabolic component (see the online appendix at http://dx.doi.org/10.2337/dc07-0147) and compared against the observed distribution using χ² statistics.

For constructs 1 and 2, stratified analyses were done to confirm if the qualitative conclusions differed by race/ethnicity.

CIMT and IIMT were analyzed on the log scale. Coefficients for multiple linear regression were adjusted for age and race/ethnicity. In model series 1, the population was restricted to those who had no prevalent components, exactly one component, or three or more components. (Those who had exactly two components neither have the defined metabolic syndrome nor reflect the individual component contributions.) Regression models specified separate coefficients for the five isolated components and a coefficient for the metabolic syndrome cluster, compared with the referent group that has no prevalent component. The difference between the linear combination of the five component coefficients and the cluster coefficient was tested for statistical significance. In model series 2, individuals with all five metabolic syndrome components co-prevalent were included with those with exactly one, or none, of the components prevalent. In confirmatory model series 3, all components (irrespective of being isolated) were included, and the metabolic syndrome cluster variable was also included as a single interaction variable. However, this model series does not account for all possible two-component clusters, and hence the coefficients of the individual components cannot be interpreted as unambiguously as in model series 1 and 2. Racial/ethnic heterogeneity of supra-additivity was tested using omnibus F tests of the race-metabolic syndrome interaction variables in models including a race/ethnicity-specific coefficient for each isolated component and the metabolic syndrome cluster.

In separate models adjusted for age and racial/ethnic group, an interaction coefficient was estimated to test whether the presence of each of 26 possible co-prevalent component combinations was associated with IMT above and beyond the sum of the component coefficients.

The proportions of MESA participants with the individual components of the metabolic syndrome were as follows: large waist circumference (38% men, 69% women), high triglycerides (32% men, 29% women), low HDL cholesterol (40% men, 49% women), high blood pressure (59% men, 59% women), and high fasting plasma glucose (49% men, 36% women), with 40% of men and 46% of women having three or more components.

For each metabolic syndrome component, median HOMA was higher if the component was present (all differences P < 0.001, supplementary Table 1 in the online appendix). Table 1 shows that an increasing number of co-prevalent components was associated with higher HOMA levels in men and women (all trends, P < 0.001). This was true in race/ethnicity-stratified analysis (all groups, P < 0.001) (supplementary Table 2 in the online appendix).

For the number of co-prevalent components as the diagnostic for insulin resistance, the area under the receiver-operating characteristic curve was 0.80 (95% CI 0.78–0.81) for men and 0.80 (0.79–0.82) for women. The metabolic syndrome (three or more components) results in sensitivity and specificity levels of 74 and 29% in men and 82 and 34% in women, respectively. A moderate to good diagnostic efficiency was present in men and women of all four race/ethnicities (area under the receiver-operating characteristic curve range 0.74–0.86 for different groups) (detailed in supplementary Table 3 in the online appendix).

In both men and women, there is a graded relationship between the prevalence of any given component and the number of other components (for all components, P < 0.001) (supplementary Fig. 1 in the online appendix).

The observed distribution of co-prevalence of components was significantly different from the co-prevalence expectation given the prevalence of each of the individual components (Fig. 1, P < 0.001 for men and women). Specifically, nonprevalence of any component, and co-prevalence of four and five components were observed more frequently than expected (i.e., tails of the distribution were heavier), suggesting that clustering is not a chance event. Similarly, significantly heavier-tailed distributions were observed for men and women in analyses stratified by race/ethnicity (all groups, P < 0.001). In all men, the largest contributors to χ² were 1) no components; 2) all components; 3) obesity, high triglycerides, low HDL cholesterol, and high blood pressure, without high glucose (all more frequent than expected); and 4) isolated high blood pressure and low HDL cholesterol. Isolated high blood pressure and high triglycerides were less frequently observed than expected. The highest deviating clusters in all women include those listed above in 1) through 4), as in men, with the cluster of isolated obesity and high triglycerides less frequently observed than expected.

Figure 2 shows that the association of CIMT or IIMT with the co-prevalence of three or more components (model series 1), or all five components (model series 2), is no greater than the added isolated component associations (i.e., no supra-additive interaction) either in men or women. Note that the small nonsignificant (P = 0.66) negative association of CIMT with isolated glycemia among women is conditioned on the absence of all other components. There are no significant racial/ethnic differences in the lack of supra-additive interaction of the metabolic syndrome cluster above the five isolated components. The lack of supra-additive association of the metabolic syndrome with IMT variables was confirmed in models estimating the marginal rather than isolated component effects (model series 3).

Using the Bonferroni criterion level (α = 0.05/26 = 0.0019), no positive additive interaction was found for any of the combinations of two or more metabolic components in relation to either CIMT or IIMT.

We showed that the metabolic syndrome qualified as a “syndrome” under some, but not all constructs hypothesized in the present study.

Construct 1 postulated a common denominator for the components of the syndrome. We confirmed the graded association of HOMA with the number of components. If direct measurement of insulin resistance is not feasible, the metabolic syndrome as defined may be used as a clinical diagnosis of insulin resistance.

The assumption of the primacy of insulin resistance is controversial, since abdominal obesity has been suggested as an essential etiological basis for the metabolic syndrome (14). Nevertheless, the metabolic syndrome could be used as a marker of insulin resistance. If treatment of insulin resistance per se may not affect the syndrome components, the practical significance of this finding is not clear.

Construct 2 postulated that clustering was not by chance and thus was hypothesis generating. We presented a metric and directly tested for the hypothesis of clustering that allowed for additive component count, because components of syndromes are used additively by clinicians (15,16). We showed that the components were correlated, such that component counts of four or five, and correspondingly, zero components, occurred more frequently than chance expectation. Data reduction techniques such as factor analysis generate new linear combinations of components (reviewed by Lawlor et al. [6]) and do not test the syndrome construct hypothesis. Though factor analysis may be useful for generating latent phenotypes (6,17), this was outside the scope of our analyses.

Practically, the detection of any one component should alert the clinician to seek other abnormalities. Clustering generates the hypothesis of a common source for the abnormalities. This source may be biochemical, such as insulin resistance, but our analyses do not exclude the possibility that other sources are responsible for the clustering, such as psychosocial causes.

For construct 3, “syndrome” implied that co-prevalence of components was associated with a supra-additive association with pathology, more than expected by summing individual component contributions. This understanding of the “syndrome” construct has been mentioned in a recent debate (1–3). This construct was also implicit in the demonstration by Golden et al. (18) that 18 of the 57 six-component combinations were associated with supra-additive carotid IMT. Surprisingly, this was not found in our analyses, which show no supra-additive association. However, our results may not be directly comparable to those of Golden et al. (18), which used hyperinsulinemia (≥110 pmol/l) and BMI ≥25 kg/m² as metabolic components, and more stringent criteria for hypertension (≥140 mmHg systolic), high triglycerides (≥200 mg/dl), high glucose (≥110 mg/dl), and low HDL in women (≤40 mg/dl) in a younger population (45–64 years). Carotid bulb IMT, needed to exactly replicate the analyses of Golden et al. (18), was not measured in MESA. In analysis that used the metabolic criteria of Golden et al. (18) among individuals 45–64 years, our conclusion regarding construct 3 for common, internal, or mean IMT remained unchanged (results not shown). It is possible that there are differences between the cohorts studied and that construct 3 may be valid in some populations. Further, in our analysis, the association of each component with IMT conditioned on the absence of all other components was somewhat weak, with a nonsignificant negative association in one instance. The weak association of isolated components should result in an enhanced power to detect supra-additive association of clustered components, yet no supra-additive association was seen in our analysis.

Interestingly, supra-additive association with a harmful outcome is not expected in other clinically useful “syndromes.” For example, the only prognosticator for acute rheumatic fever sequelae is carditis (19) and not the other components of the Duckett-Jones criteria (15); thus, there is no proven supra-additive risk for component clustering. To examine supra-additive association of clustering with upstream etiology, we can consider the example of the Down's syndrome using an eight-component syndrome definition (16). In a series of 19,000 live births, there was no incremental association of trisomy 21 with the component counts beyond 5, suggesting a saturation of the association of cause at higher scores, which is statistically equivalent to a sub-additive effect of clustering.

However, should this construct come to be generally accepted for metabolic abnormalities, our results indicate that the metabolic syndrome is not “syndromic” in terms of carotid IMT as pathology. The practical implication is that every metabolic component should be vigorously treated on its own, with no expectation of a putative supra-additive risk reduction for subclinical carotid atherosclerosis.

Our analysis has certain limitations. HOMA is not the gold standard for the measurement of insulin resistance. However, the random error associated with the estimation of insulin resistance by HOMA would be more likely to dilute the findings resulting from the analysis of construct 1. We have used IMT, an intermediate cardiovascular outcome that is available in MESA. A valid outcome measure for construct 3 needs longitudinal follow-up for incident diabetes, myocardial infarction, or death. This limitation must be borne in mind while comparing the results of Golden et al. (18), which found construct 3 to be valid for a slightly different measure of IMT in a different population using different metabolic components.

In summary, while we could not observe a supra-additive interaction resulting from the simultaneous presence of metabolic syndrome components in terms of intimal-medial thickening of the carotid artery, co-prevalence of multiple components of the metabolic syndrome was more frequent than expected by chance alone. Detection of any of the components of the syndrome in a patient should raise the index of suspicion in the health care provider that the other components may also be present. This clustering is consistent with the presence of a common underlying cause. If the underlying cause is insulin resistance, our results suggest that the metabolic syndrome provides a good proxy diagnostic for insulin resistance, which is not assessed in routine clinical tests. However, it remains to be determined if identification of insulin resistance will result in a more effective therapeutic approach than that aimed at the individual components of the metabolic syndrome.

Calculation of expected component co-prevalences: Let p₁, p₂, … p₅ (q₁, q₂,… q₅) be the marginal prevalence (nonprevalence) of the five components of the metabolic syndrome.

The expected co-prevalences are given by the following:

• \[0\ \mathrm{components}:\ {\Pi}_{i}(q_{i})\]
• \[1\ \mathrm{component}:\ {\sum}_{i}\ {\{}{\Pi}p_{i}q_{j{\neq}i}{\}}\]
• \[2\ \mathrm{components}:\ {\sum}_{i}{\sum}_{j{\neq}i}{\{}{\Pi}p_{i}p_{j}q_{k{\neq}i,j}{\}}\]
• \[3\ \mathrm{components}:\ {\sum}_{i}{\sum}_{j{\neq}i}{\{}{\Pi}q_{i}q_{j}p_{k{\neq}i,j}{\}}\]
• \[4\ \mathrm{components}:\ {\sum}_{i}{\{}{\Pi}q_{i}p_{j{\neq}i}{\}}\]
• \[5\ \mathrm{components}:\ {\Pi}_{i}(p_{i})\]

This research was supported by contracts N01-HC-95159 through N01-HC-95165 and N01-HC-95169 from the National Heart, Lung, and Blood Institute and M01-RR-02719 from the National Center for Research Resources.

The authors thank the other investigators, the staff, and the participants of MESA for their valuable contributions.

A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org.