Is There a Single Underlying Factor for the Metabolic Syndrome in Adolescents?: A confirmatory factor analysis

The 1999–2002 National Health and Nutrition Examination Survey used a multistage, stratified sampling design to represent the noninstitutionalized civilian U.S. population (25,26). Participants were interviewed at home and were invited to attend a mobile examination center where they provided a blood sample and were examined. We limited the analyses to boys and girls aged 12–17 years who attended the morning medical examination, had fasted ≥8 h, and had complete data on all variables (n = 1,262). Only non-Hispanic Caucasians, non-Hispanic African Americans, and Mexican Americans were included in the final analyses.

Serum specimens were frozen at <−70°C, shipped on dry ice, and stored at <−70°C until analysis. Plasma insulin concentration was measured via a Pharmacia insulin radioimmunoassay kit (Pharmacia Diagnostics, Uppsala, Sweden). Plasma glucose concentration was measured by using an enzymatic reaction (Cobas Mira Chemistry System; Roche Diagnostic Systems, Montclair, NJ). Details on insulin assays can be found elsewhere (25–27). The HOMA-IR was calculated as [glucose (mmol/l) × insulin (μU/ml)]/22.5 (28). Triglyceride concentration level was measured enzymatically in serum after being hydrolyzed to glycerol using a series of coupled reactions. HDL cholesterol level was directly measured on a Hitachi 704 Analyzer after the precipitation of other lipoproteins with a heparin-manganese chloride mixture. The triglycerides–to–HDL cholesterol level was calculated using triglyceride concentration divided by HDL cholesterol concentration.

Waist circumference was measured by two trained health technicians using a steel measuring tape to the nearest 0.1 cm at the high point of the iliac crest at minimal respiration when the participant was in a standing position (29). Up to four SBP and DBP readings were obtained from participants. The average of the last two measurements of SBP and DBP for the participants who had three or four measurements, the last measurement for the participants with only two measurements, and the only measurement for the participants who had one measurement were used to establish high blood pressure status. MAP was calculated as DBP + 1/3(SBP − DBP).

Fasting insulin, HOMA-IR, triglycerides, and triglycerides–to–HDL cholesterol were log transformed to approximate a normal distribution. All variables were standardized (mean = 0 and SD = 1) for age and sex. The four-factor model and the one-factor CFA model in adults were specified according to Shen et al. (16) and Pladevall et al. (19). The one-factor model proposed in adolescents was specified as follows: waist circumference, triglycerides, fating insulin, and SBP may be influenced by an underlying factor, the “metabolic syndrome,” and a measurement error. No correlated measurement errors between any two measures were assumed. The factor loading (λ) of each measured variable indicates the strength of its association with the underlying factor. We used a cutoff value of ±0.3 as the minimal level of a practically significant factor loading (30) and α = 0.05 as the significance level for two-tailed statistical tests.

The estimates of the parameters were obtained using the maximum likelihood method of the Mplus software (31). The χ² test, comparative fit index (CFI), and standardized root mean square residual were used to assess the goodness of fit of the hypothesized model to the data (30,31). A cutoff value between 0.90 and 0.95 for CFI is recommended as an acceptable fit (32), and a cutoff value of ≥0.95 for CFI or ≤0.08 for standardized root mean square residual is recommended for a good fit (32).

A multiple-group analysis was conducted to test the invariance of factor loadings in the CFA model across sex and race/ethnicity. The χ² difference test was used to determine whether the factor loadings between the two groups were statistically significant. The Bonferroni adjustments of the P values were applied for the comparisons of overall factor loadings among the three racial or ethnic subgroups (0.05/3 = 0.017), the four individual factor loadings (0.05/4 = 0.013) by sex, and the six pairs of factor loadings (0.05/6 = 0.008) of the final model for the total sample.

In the final analytic sample, 51.2% were boys, 29.4% Caucasians, 30.7% African Americans, and 39.9% Mexican Americans. The means ± SD and correlation coefficients of waist circumference, triglycerides, fasting insulin, and SBP are shown in Table 1.

The four-factor model proposed by Shen et al. (16) was carefully specified, yet resulted in no convergence. The correlation coefficients of DBP (ranged from 0.02 to 0.10) or fasting glucose (0.04 to 0.22) with other components were low. The estimates of goodness of fit for the one-factor CFA model proposed in adults by Pladevall et al. (19) were adequate (Fig. 1A). However, the factor loading for MAP was low (λ = 0.12), indicating a poor validity of this measure.

The estimates of goodness of fit for the one-factor CFA model proposed in the present study were adequate (Fig. 1B). The overall estimates of factor loadings for the total sample were 0.76, 0.46, 0.81, and 0.42 for waist circumference, triglycerides, fasting insulin, and SBP, respectively. All estimates of factor loadings were >0.3, indicating an acceptable validity of the four directly measured variables.

The forthcoming analyses were based on the one-factor model in adolescents proposed in the present study. The estimates of goodness-of-fit indexes were excellent for Caucasian (Fig. 2A), African-American (Fig. 2C), and Mexican-American (Fig. 2E) boys and Mexican-American girls (Fig. 2F); good for Caucasian girls (Fig. 2B); and acceptable for African-American girls (Fig. 2D). Among the four measures, fasting insulin had the largest factor loading for the metabolic syndrome in all sex- and race/ethnicity-specific subgroups, except in African-American girls.

The estimates of factor loadings of the four measures for the metabolic syndrome were similar between boys and girls (χ² = 4.88 [3 d.f.], P = 0.18). There was no statistical significance in the estimates of factor loadings between boys and girls among Caucasians (χ² = 6.26 [3 d.f.], P = 0.10), African Americans (χ² = 0.73 [3 d.f.], P = 0.87), and Mexican Americans (χ² = 3.21 [3 d.f.], P = 0.36), suggesting similarity in the construct validity of the measured variables for the metabolic syndrome across sex. No statistically significant differences in the overall factor loadings among the three racial/ethic subgroups were detected using Bonferroni adjustments for P values at the 0.017 level (Table 2).

There were no statistically significant differences in the estimates of factor loadings between waist circumference and fasting insulin (χ² = −0.06 [1 d.f.], P = 0.81) and between triglycerides and SBP (χ² = 0.77 [1 d.f.], P = 0.38). The factor loading of waist circumference was greater than that of triglycerides (χ² = 59.78 [1 d.f.], P < 0.01) and SBP (χ² = 77.43 [1 d.f.], P < 0.01). The factor loading of fasting insulin was greater than that of triglycerides (χ² = 79.14 [1 d.f.], P < 0.01) and SBP (χ² = 89.37 [1 d.f.], P < 0.01).

HDL cholesterol was inversely correlated with triglycerides (r = −0.42). The correlation between triglycerides and triglycerides–to–HDL cholesterol ratio (r = 0.93) was stronger than that between HDL cholesterol and triglycerides–to–HDL cholesterol (r = −0.69). The factor loading of triglycerides (λ = 0.46) was similar to triglycerides–to–HDL cholesterol ratio (λ = 0.50), but slightly larger than that of HDL cholesterol (λ = −0.38). Multiple group analyses indicated that the factor structure of the model using HDL cholesterol differed between African Americans and Mexican Americans (χ² = 10.82 [3 d.f.], P = 0.0127). In addition, the factor loading of fasting insulin (λ = 0.81) was similar to that of HOMA-IR (λ = 0.78) but larger than that of fasting glucose (λ = 0.17).

Based on the adequate fit and valid factor structures of the one-factor model proposed in the present study, waist circumference, triglycerides, fasting insulin, and SBP may be potentially useful as the four phenotypic traits of an underlying factor that defines the metabolic syndrome in adolescents. Of the four simple and clinically available measures, waist circumference and fasting insulin appeared to be the major components in the syndrome. In particular, the one-factor model seemed to be generalizable in various subpopulations because no significant differences in the factor structures of the model across sex and race/ethnicity were detected in our study.

Despite an overall adequate fit for the model proposed by Pladevall et al. (19) in adults, MAP as a potential component for the metabolic syndrome in adolescents may be questionable because of its poor construct validity (λ = 0.12). In fact, MAP, as a measure of average pressure throughout the cardiac cycle, has been studied less in relation to insulin resistance or obesity among both children and adults. In contrast, SBP was positively associated with insulin resistance, whereas DBP was not or was weakly associated with insulin resistance (24). Thus, the use of SBP as a potential component for the metabolic syndrome seemed to be more tenable than either MAP or a combination of SBP and DBP in adolescents.

There are several advantages of using fasting insulin as a potential component for the metabolic syndrome. At the simplest level, it is as good a surrogate estimate of insulin resistance (20–22) as various combinations of fasting insulin and glucose concentration such as HOMA-IR (28). Of greater clinical relevance may be the pathophysiological role that hyperinsulinemia plays in the development of the clinical abnormalities that occur more frequently in individuals who are insulin resistant. Finally, the construct validity of fasting insulin was similar to that of HOMA-IR, yet much greater than that of fasting glucose in the definition of the metabolic syndrome.

The use of triglycerides in lieu of a triglycerides–to–HDL ratio or HDL cholesterol as a possible component of the metabolic syndrome in adolescents may have the following advantages: 1) triglycerides correlated more closely to triglycerides–to–HDL cholesterol ratio than HDL cholesterol; 2) the model using triglycerides was less variant in factor structures than that using HDL cholesterol across sex and race/ethnicity; 3) elevated triglyceride concentrations have been considered a key marker for atherogenic dyslipidemia or the lipid triad, i.e., raised triglyceride levels, small LDL particles, and low HDL cholesterol (23); and 4) low HDL cholesterol was a component of the metabolic syndrome only in the presence of hypertriglyceridemia in patients with type 2 diabetes (33). Therefore, triglycerides appeared to be a preferable measure of dyslipidemia in the definition of the metabolic syndrome in adolescents.

Insulin resistance and abdominal obesity have been proposed as major underlying causes for the metabolic syndrome (34,35). Direct comparison of the relative importance of insulin resistance and abdominal obesity in the metabolic syndrome is difficult and scarce in literature. A previous study (15) showed that obesity might be a stronger component of the metabolic syndrome in adolescents than hyperinsulinemia. Our results, however, that fasting insulin and waist circumference were approximately equally associated with the metabolic syndrome, suggest that both insulin resistance and abdominal obesity may be the key features of the syndrome.

Our results provide a conceptual framework of the metabolic syndrome in adolescents. To be useful in clinical practice, research, and surveillance, findings from factor analyses have to be translated into a practical definition of the metabolic syndrome. One approach would be to emulate the definitions of the metabolic syndrome among adults, such as the ones developed by the National Cholesterol Education Program and the World Health Organization, and use threshold values for the components specific to children and adolescents, as recommended in guidelines for waist circumference (4,36,37) and SBP (38). Adult thresholds for triglycerides would need to be adapted to children and adolescents (2). The threshold value of >20 mU/l for fasting plasma insulin concentration proposed by the American Heart Association may be potentially useful for the clinical assessment of insulin resistance in pediatric population (39). Another approach would be to develop a risk score for the metabolic syndrome based on multivariate modeling or on summing z-scores for the components. Because such a risk score is a continuous measure, one or more cut points could be established, leading to a classification such as having or not having the metabolic syndrome or such as low, medium, or high risk for the metabolic syndrome.

In conclusion, our findings have implications in clinical practice, epidemiologic research, and public health surveillance. The one-factor model uses valid, simple, and easily available measures to define the metabolic syndrome; thus, it may facilitate the diagnosis of the syndrome in clinical settings and the development of a case definition for use in surveillance. In addition, the model appeared to be consistent across sex and racial/ethnic subgroups and therefore could be generalized to diverse populations. It might be of particular interest to use valid and universally applicable measures to define the metabolic syndrome in the pediatric population, given the lack of a standard pediatric definition of the syndrome to date.