Correspondence Between the Adult Treatment Panel III Criteria for Metabolic Syndrome and Insulin Resistance

The sample consisted of 256 asymptomatic adults (133 men and 123 women) who had previously participated in the Rochester Family Heart Study (RFHS) or were spouses, siblings, or offspring of previous RFHS participants. The RFHS cohort included 3,974 members of three-generation pedigrees who were ascertained through households with children enrolled in the schools of Rochester, Minnesota, in 1984 (7), without regard to the health status of family members. Recruitment for the present study required that individuals were between the ages of 20 and 60 years and were not taking any medications that could affect lipid or carbohydrate metabolism. No pregnant or lactating women were recruited. The study protocol was approved by the Institutional Review Board of the Mayo Clinic and was carried out in accordance with institutional guidelines after each participant signed a consent form.

Each participant underwent an initial evaluation that included blood sampling, a review of prior medical history (including use of medications), physical examination that included measurements of blood pressure by random zero sphygmomanometer, height by wall stadiometer, weight by electronic scale, and waist and hip circumferences by tape measure (all recorded by trained personnel). The examining physician administered a standardized medical questionnaire and recorded the subjects’ responses as part of this evaluation. The blood pressure measurements were taken after subjects sat quietly for at least 5 min, and averages of the three readings were used in the analyses. Blood was drawn after an overnight fast of at least 8 h, and samples were placed on the ice immediately and centrifuged within 1 h, after which serum and plasma were separated, frozen, and stored at −80°C until assays were performed. Waist circumference was measured with the subject standing upright; the examiner, positioned at the right of the subject, made an imaginary horizontal mark just above the uppermost lateral border of the right iliac crest and crossed it with a vertical mark on the midaxillary line. The measuring tape was placed in a horizontal plane around the abdomen at the level of this marked point; the plane of the tape was parallel to the floor and the tape was snug but did not compress the skin. The measurement was made at normal, minimal inspiration. Hip circumference was defined as being the widest circumference over the buttocks and below the iliac crest. To obtain an accurate measurement, measurements were made at several positions and the widest circumference was recorded. Subjects underwent dual-energy X-ray absorptiometry scanning (DPX-IQ; Lunar Radiation, Madison, WI) for body composition assessment (body fat percentage) at the General Clinical Research Center (GCRC). Blood was drawn after an overnight fast (at least 8 h) for measurement of plasma glucose and lipid (total cholesterol, triglycerides, and HDL cholesterol) concentrations. Subjects were required to discontinue any prescription medications known to influence blood pressure, plasma lipids, or glucose. Two of the subjects had a previous history of drug-treated hypertension, but these medications were withdrawn at least 4 weeks before study participation. However, none of the subjects had a previous history of diabetes and none was previously treated with antidiabetic medications.

The metabolic syndrome definition was determined by the categorical ATP-III criteria. The following cutoff points were used: waist circumference >102 cm (>40 inches) in men and >88 cm (>35 inches) in women; HDL cholesterol <40 mg/dl in men and <50 mg/dl in women; triglycerides ≥150 mg/dl in both sexes; systolic blood pressure ≥130 mmHg or diastolic blood pressure ≥85 mmHg in both sexes; and fasting glucose ≥110 mg/dl in both sexes (1).

Subjects consumed a controlled, high-carbohydrate diet to increase insulin sensitivity for 6 days before the FSIVGTT (8). The diet consisted of ∼55% carbohydrate, 30% fat, 15% protein; daily, the diet contained 1 g of calcium, 150 mmol of sodium, and 90 mmol of potassium per 2000 kcal with sufficient kilocalories to maintain weight. The meals were prepared in the metabolic kitchen of the GCRC, where participants ate two of their three meals each day.

On the morning of day 7 between 7 and 9 a.m. after an overnight fast, an intravenous catheter was placed in each of the subjects’ forearms—one for bolus injections of glucose and tolbutamide and the other for rapid, repeated blood sampling for measurement of glucose and insulin concentrations. After baseline samples, a bolus of glucose (0.3 g/kg body weight) was injected over a 60-s period. Blood was then sampled 36 times over the next 180 min, with an initial frequency of 1 sample per min. Twenty minutes after the glucose bolus, a single dose of tolbutamide was injected (100–300 mg) to elicit a secondary insulin response, which facilitates estimation of insulin resistance in subjects with impaired endogenous insulin release.

The reference standard for insulin resistance was determined by Bergman’s minimal model. Consistent with previous literature, insulin resistance was defined as an insulin sensitivity index (S_i) below the lowest quartile for nondiabetic subjects in this study, i.e., S_i <2.0 × 10 min⁻¹ · μU⁻¹ · ml⁻¹ (9,10).

Data were summarized by calculating means ± SD for quantitative variables and percentages for categorical variables. Each individual was assessed for the presence of the metabolic syndrome using data collected on waist circumference, blood pressure, and fasting plasma concentrations of HDL cholesterol, glucose, and triglycerides. Using the ATP-III definition of the metabolic syndrome described above, we calculated the sensitivity, specificity, and positive predictive value of detecting insulin resistance defined as stated above. Receiver operating characteristic (ROC) curves were constructed to provide a graphical representation of the relationship between false-positive (i.e., 1 − specificity) and true-positive (sensitivity) detection rates for the counting of categorical metabolic syndrome components. In addition, ROC curves were constructed for each quantitative trait underlying the component measures. A standard way to evaluate diagnostic accuracy is by calculating the area under the ROC curve, and the optimum value for each quantitative measure can be defined by the value on the ROC curve closest to the upper left corner of the graph, using the formula: square root of (1 − sensitivity)² + (1 − specificity)². To determine whether the areas under the curve (AUC) obtained using different predictors were significantly different, we used the method of Delong et al. (11).

In each sex, multiple linear regression was used to identify metabolic syndrome component measures that made additional contributions to the prediction of S_i after adjustment for age. Predictor variables were selected in a forward stepwise fashion with standardized variable entry (0.05) and elimination (0.10) criteria. The utility of the additional predictive measures was assessed by comparing R² values of a full model that included the additional predictor with a reduced model that did not. All analyses were conducted using the statistical software JMP, 5.0 version (12).

The sample was composed of 123 men and 133 women between the ages of 20 and 60 years (mean ± SD 39.1 ± 12 years). On average, men were significantly taller and heavier and had larger waist circumference, glucose values, and triglyceride concentrations and lower HDL cholesterol concentrations than women (Table 1). Overall, 69 of the 256 (26%) subjects were insulin resistant based on having S_i values <2 × 10⁻⁴ min⁻¹ · μU⁻¹ · ml⁻¹. Mean S_i did not differ significantly between sexes nor did the prevalence of insulin resistance (26% in men, 27% in women; P = 0.348). Based on the ATP-III criteria, the prevalence of the metabolic syndrome was 15.6% and did not differ significantly between sexes (16.2% in men and 15.1% in women; P = 0.465).

Sensitivity of the metabolic syndrome to identify insulin resistance was 42% (45% in men and 39% in women; P = 0.137) and specificity was 94% (93% in men and 95% in women; P = 0.345) (Table 2). Sensitivity and 1 − specificity are plotted in Fig. 1 as functions of the number of components of the metabolic syndrome. Sensitivity could be improved to 70% (from 42%) by requiring only two metabolic syndrome components at the expense of decreasing specificity to 78% (from 94%). Diagnostic accuracy of counting metabolic syndrome components, judged by the AUC, was 0.768 (0.797 in men and 0.747 in women).

We constructed ROC curves of each metabolic syndrome component to identify optimum values to identify insulin resistance (Table 3). By using these optimum values as cut points for the predefined metabolic syndrome components, overall diagnostic accuracy of counting components, judged by the AUC, was improved to 0.847 in men (from 0.797; P = 0.18) and to 0.787 in women (from 0.747; P = 0.43). However, the AUC for waist circumference alone considered as a quantitative trait was greater in both men (0.906; P = 0.001) and women (0.822; P = 0.10) and equivalent to counting metabolic syndrome components when the sexes were pooled (0.813). Triglyceride concentration was the only other single component that provided greater overall diagnostic accuracy than counting the metabolic syndrome components. However, the AUC for triglyceride concentrations alone (0.806 in men and 0.806 in women) was less than that for waist circumference or for counting optimum value components.

To further evaluate the predictive utility of quantitative levels of component measures of the metabolic syndrome, we conducted multiple variable linear regression analyses in which the outcome variable was quantitative level of insulin sensitivity (logS_i). In men, waist circumference was the only metabolic syndrome component to enter the model as an additional predictor of logS_i after controlling for age. Waist circumference accounted for an additional 39% of the interindividual variation in logS_i (model R² = 0.47, P = 0.001). In women, after we controlled for age, not only waist circumference but also triglycerides and fasting glucose entered the model as additional predictors of logS_i, together accounting for 43% of the interindividual variation of logS_i (model R² = 0.44, P = 0.001). However, waist circumference accounted for most of the additional variation in logS_i explained by the latter model (i.e., 34%).

In this study of asymptomatic subjects from the white non-Hispanic population of Rochester, Minnesota, insulin resistance was determined by the minimal model technique under standardized conditions of diet and activity. Our findings confirm that the ATP-III criteria for the metabolic syndrome are an insensitive but fairly specific method to identify subjects with insulin resistance (13). The individual component measures of the metabolic syndrome, as defined by ATP-III, vary both in terms of their prevalences and in their relationships to insulin resistance. Moreover, cut point values for each quantitative metabolic syndrome component measure that optimized diagnostic accuracy for insulin resistance in our sample from the general Caucasian population differed from those recommended by ATP-III. A single measure of central obesity, waist circumference, alone appeared to provide greater overall diagnostic accuracy than counting metabolic syndrome components as advocated by ATP-III guidelines.

The notion that the metabolic syndrome is a cluster of related abnormalities that commonly occur among pre-diabetic, insulin-resistant individuals was first brought forward by Reaven in 1988 (14). According to this conceptualization, insulin resistance is considered to be an underlying pathophysiologic disturbance that not only contributes to glucose intolerance, dyslipidemia, and hypertension but also may make an additional, independent contribution to CHD risk. However, the ATP-III panel did not recommend routine measurement of insulin resistance or include assessment of insulin resistance per se as identifying criterion for the metabolic syndrome, primarily because of the laborious, time-consuming methods required to measure insulin sensitivity.

To our knowledge, only two previous reports have assessed the relationship between the ATP-III criteria and a measure of insulin resistance. In a small nondiabetic sample, Liao et al. (13) compared component metabolic syndrome measures to insulin resistance measured by hyperinsulinemic-euglycemic clamp in 74 unselected Caucasian individuals. They reported that ATP-III criteria have low sensitivity to identify insulin resistance among nondiabetic subjects. Also in a nondiabetic sample, Cheal et al. (15) compared each individual component of the metabolic syndrome to insulin resistance determined by a modified insulin suppression test (steady-state plasma glucose concentration) in 443 subjects who were selected from a database of individuals who had participated in metabolic studies over the previous 10 years. They reported that the ATP-III criteria for metabolic syndrome do not provide a sensitive approach to identify insulin-resistant individuals and that measures of obesity and dyslipidemia are more useful predictors. However, this study did not measure waist circumference but instead used BMI as a measure of obesity in the definition of metabolic syndrome. Moreover, the sample selection criteria effectively leave the population of inference unknown. The estimated overall prevalence of the metabolic syndrome in their sample was higher than that in ours by almost 5% (20.5 versus 15.6%), and this increase in prevalence, seen especially in women (22.6% compared with 15.1% in our sample) might be explained by the use of BMI instead of waist circumference as a measure of obesity.

The relationship between the presence of the metabolic syndrome by ATP-III criteria and insulin resistance is not perfect and suffers mostly from lack of sensitivity. Depending upon the screening objective, more sensitive criteria may be desirable. For example, recognizing that insulin resistance may be the underlying metabolic disorder contributing to increased cardiovascular risk (6), reducing the number of required ATP-III components from three to two would provide greater sensitivity to identify individuals at increased risk. It is also apparent that the relationship of each metabolic syndrome component to insulin resistance differs greatly among the recommended ATP-III components. Waist circumference and triglyceride concentrations appeared to have the greatest diagnostic accuracy for insulin resistance, whereas blood pressure, glucose value, and HDL cholesterol concentration had lesser predictive utility. Our results suggest that simple measurement of just waist circumference may provide greater overall diagnostic accuracy than counting categorical components of the metabolic syndrome. That is, use of additional measures beyond waist circumference added essentially nothing to either sensitivity or specificity in the identification of insulin-resistant individuals.

A limitation of this study is that the results may only be strictly applicable to healthy Caucasians without type 2 diabetes. The FSIVGTT analyzed by the minimal model technique in the present study is not perfectly correlated with estimates of logS_i from the euglycemic clamp, which is the gold standard for measuring insulin sensitivity. However, application of the minimal model to measurements of glucose and insulin obtained during a FSIVGTT provides well-correlated estimates of logS_i and is more feasible for epidemiologic studies (9).

In summary, application of the ATP-III metabolic syndrome criteria provides good specificity but low sensitivity for screening asymptomatic white adults for insulin resistance. Measuring just waist circumference is simpler and may provide greater accuracy for identification of insulin resistance.