Comparison between the euglycemichyperinsulinemic clamp and surrogate measures
Abstract
OBJECTIVE—In this study we compared fasting insulin and measures of insulin sensitivity (M) based on fasting insulin and glucose (i.e., homeostasis model assessment [HOMA], quantitative insulin sensitivity check index [QUICKI], and fasting glucose–to–insulin ratio [FGIR]) or triglycerides to the insulin clamp in a cohort of children/adolescents.
RESEARCH DESIGN AND METHODS—The subjects were Minneapolis fifth to eighthgrade students. Euglycemichyperinsulinemic clamps were performed on 323 adolescents at mean age 13 and were repeated on 300 of these subjects at mean age 15. Insulin sensitivity was determined by glucose uptake (milligrams per kilogram per minute) adjusted for lean body mass (M_{LBM}) and steadystate insulin (M_{LBM}/ln SSI). Comparisons were made for the whole cohort and by body size (BMI <85th percentile vs. BMI ≥85th percentile). Receiver operating characteristic (ROC) curves were used to test whether specific fasting insulin cut points separated truepositive from falsepositive approximations of insulin resistance.
RESULTS—Fasting insulin was significantly correlated with HOMA (r = 0.99), QUICKI (r = 0.79), FGIR (r = −0.62), and (ln fasting insulin + ln triglycerides) (r = 0.88). Correlations of the surrogates with M_{LBM} were significantly lower than those with M for the total cohort and ≥85th percentile group. In general, correlations in the ≥85th percentile group were higher than those in the <85th percentile group. Correlations with M_{LBM} and M_{LBM}/ln SSI decreased in the total cohort and ≥85th percentile group from age 13 to 15. ROC curves showed only a modest capability to separate true from falsepositive values.
CONCLUSIONS—Surrogate measures are only modestly correlated with the clamp measures of insulin sensitivity and do not offer any advantage over fasting insulin. In general, lower correlations are seen with M_{LBM} than with M and with thinner than with heavier individuals.
 AUC, area under the receiver operating characteristic curve
 FGIR, fasting glucose–to–insulin ratio
 HOMA, homeostasis model assessment
 LBM, lean body mass
 QUICKI, quantitative insulin sensitivity check index
 ROC, receiver operating characteristic
 SSI, steadystate insulin
The increasing prevalence of childhood obesity, the strong association between insulin resistance and obesity, and the relation of insulin resistance to the metabolic syndrome have led to an intense clinical and research interest in the measurement of insulin resistance in children (1–3). The “gold standard” method for measuring insulin resistance is the euglycemichyperinsulinemic clamp (4), in which a constant intravenous infusion of insulin is balanced by a simultaneous infusion of glucose in a clinical research setting. Because of the invasive, timeconsuming nature of the clamp procedure, it is difficult to use in routine clinical practice or in large epidemiological studies, particularly in children.
In an attempt to simplify the measurement of insulin resistance, a number of surrogate measures based on fasting levels of insulin and glucose (i.e., the homeostasis model assessment [HOMA], the quantitative insulin sensitivity check index [QUICKI], and the fasting glucose–to–insulin ratio [FGIR]) have been developed and widely used in adult (5–7) and, subsequently, pediatric studies. Others have used fasting insulin and triglycerides as the surrogate measures (8). However, these surrogate measures have been directly compared to insulin clamp–derived measures of insulin resistance in children and adolescents in only two studies (9,10), and most studies have been in relatively small cohorts. Thus, there are ongoing questions about the validity of the surrogate measures and their applicability in the pediatric population.
The purpose of the present study was to compare fasting insulin, HOMA, QUICKI, FGIR, and (ln fasting insulin + ln triglycerides) to euglycemic insulin clamp measures of insulin resistance in a cohort of >300 randomly selected children studied at mean age 13 and again at mean age 15. The results show that the surrogate measures do not offer any advantage over fasting insulin, they have only a moderate correlation to the insulin clamp, their correlation to the insulin clamp is lower in thin than in heavy children, and the correlation may change with increasing age.
RESEARCH DESIGN AND METHODS—
This study was approved by the University of Minnesota Committee for the Use of Human Subjects in Research. Informed consent was obtained from the parents and informed assent from the children. The subjects were participants in a longitudinal study of the relation between insulin resistance and cardiovascular risk factors in children. The original cohort was randomly recruited after blood pressure screening of Minneapolis fifth to eighthgrade public school children, as described previously (11). From this cohort, 323 insulin clamps were completed in pubertal children (Tanner stage II or greater) at mean age 13.1 ± 1.2 years. A second clamp was performed 2 years later in 300 of these participants at mean age 15.0 ± 1.2 years.
Before each insulin clamp, participants underwent a clinic examination and Tanner staging, as determined by pubic hair development in boys and breast and pubic hair development in girls. The greater of the two values in the girls was used to avoid underestimation of pubertal maturation. Height was measured with a wallmounted stadiometer, and weight was determined using a balance scale. BMI was calculated as weight in kilograms divided by height in meters squared. BMI percentiles were determined using the 2000 Centers for Disease Control and Prevention BMI for age percentile growth charts (12). Triceps and subscapular skinfold thicknesses were measured twice to the nearest millimeter with Lange calipers, and the mean value was used to predict percent body fat and lean body mass (LBM) using the Slaughter regression equation developed specifically for this agegroup (13).
Euglycemichyperinsulinemic clamps have been described in detail previously (11). Participants were admitted to the University of Minnesota Clinical Research Center after a 10h overnight fast. An arm vein was cannulated for infusion of potassium phosphate, insulin, and dextrose, and a contralateral vein was cannulated for blood sampling with the hand placed in a heated box (65°C) to arterialize venous blood. Baseline insulin and glucose levels were determined from samples drawn at 15, 10, and 5 min before beginning the insulin and glucose infusions. Baseline triglyceride levels were determined in samples drawn 15 min before the infusions were begun. The insulin infusion was started at time 0 and continued at a rate of 1 mU · kg^{−1} · min^{−1} for 3 h. An infusion of 20% glucose was started at time 0 and adjusted to maintain euglycemia (serum glucose level at 5.6 mmol/l) with plasma glucose determined every 5 min.
Plasma glucose was measured immediately at the bedside with a Beckman Glucose Analyzer II (Beckman, Fullerton, CA). Insulin samples were collected on ice and centrifuged within 20 min. Serum insulin levels were determined in the University of Minnesota Hospital Laboratory by radioimmunoassay using a double antibody method. Triglyceride levels were determined in the same laboratory as described previously (11). Insulin sensitivity (M) was calculated as the average amount of glucose (milligrams per kilogram per minute) required to maintain euglycemia during the last 40 min of the clamp and was expressed as M_{LBM} (milligrams of glucose infused per kilogram LBM per minute). Insulin sensitivity was also expressed as M_{LBM}/ln steadystate insulin (SSI) with SSI calculated from samples drawn at 160 and 180 min after the insulin infusion was begun. Lower M, M_{LBM}, and M_{LBM}/ln SSI values represent a greater degree of insulin resistance.
The mean of the three fasting insulin and glucose levels and the triglyceride level obtained at baseline were used to calculate fasting insulin, HOMA [(fasting insulin in microunits per milliliter) × (fasting glucose in millimoles per liter)/22.5)], QUICKI [1/ln(fasting insulin in microunits per milliliter) + ln(fasting glucose in milligrams per deciliter)], FGIR, and (ln fasting insulin + ln triglycerides). Fasting insulin, HOMA, and (ln fasting insulin + ln triglycerides) are inversely related, whereas QUICKI and FGIR are directly related to the degree of insulin sensitivity.
Means ± SD values were computed for the total cohort and for each BMI category (<85th and ≥85th percentiles) at each clamp at each age. Correlation coefficients in three forms are presented. The first is for a pair of variables (e.g., the correlation for M_{LBM} versus fasting insulin) in independent subgroups of people. The difference in this form of correlation was tested in a twosample t test using Fisher’s Z = ½ ln[(1 + r)/(1 − r)], with variance 1/(n − 3), where n is the number of individuals in the analysis. The second form is for two different variables with a third common variable in a constant sample of people (e.g., correlation between M versus fasting insulin and M_{LBM} versus fasting insulin). The difference in this form of correlation was tested using the method of Meng et al. (14). To compare correlations r_{13} and r_{23} in a sample of n people, calculate the difference of the Fisher’s Z for r_{13} and r_{23}. The difference is tested as a t test with variance (n − 3)/(2 · (1 − r_{12}) · h), where h = (1 − f · r^{2} mean)/(1 − r^{2} mean), r^{2} mean is the average of r_{13}^{2} and r_{23}^{2}, and f is (1 − r_{12})/(2 · (1 − r_{2} mean)), with a maximum value of 1. The third form is for the correlation between two variables measured at one time compared with the correlation between the same two variables measured at another time in a sample that is partially or completely overlapping between the two measurement times (e.g., between M versus fasting insulin at age 13 and M versus fasting insulin at age 15). Here we used the P value of the regression coefficient for interaction of the first variable with time in a repeatedmeasures regression of the repeated second variable on an intercept, an indicator for time, the repeated first variable, and the interaction mentioned above.
In an attempt to find an alternative method for assessing insulin resistance, receiver operating characteristic (ROC) curves were developed as described previously in adults (15). The curves describe the ability of a surrogate measure to separate truepositive (i.e., sensitivity) from falsepositive (i.e., specificity) insulin resistance at a series of cut points. The test accuracy is determined by the area under the ROC curve (AUC) with a value of 0.5 indicating no significant relation and a value of 1 indicating a perfect relation. In this study ROC curves were used to describe the accuracy with which levels of fasting insulin, HOMA, or QUICKI can identify individuals with insulin resistance, defined as an M_{LBM} value below the 10th percentiles of the total M_{LBM} distribution.
RESULTS—
The clinical characteristics and measures of insulin resistance at mean ages 13 and 15 are provided for the total cohort and by BMI groups in Table 1. The mean fasting glucose levels for the whole cohort were 4.9 ± 0.4 mmol/l (range 3.5–6.3 mmol/l) at mean age 13 and 4.8 ± 0.4 mmol/l (range 3.5–6.5 mmol/l) at mean age 15. The levels were not significantly different between the BMI groups at mean age 13; at mean age 15 fasting glucose was slightly, but significantly, greater in the ≥85th percentile BMI group. Fasting insulin levels for the whole cohort were 70.2 ± 55.8 pmol/l (range 9–477.6 pmol/l) at mean age 13 and 72 ± 52.8 pmol/l (range 3.6–413.4 pmol/l) at mean age 15. Fasting insulin, HOMA, and triglyceride levels were significantly higher, whereas QUICKI and FGIR were significantly lower in the ≥85th percentile BMI group at both ages. Insulin resistance by all three clamp measurements (M, M_{LBM}, and M_{LBM}/ln SSI) was significantly higher in the ≥85th percentile group at age 13, but only M remained significantly higher at mean age 15. Figure 1 shows the similar relations between fasting insulin and M_{LBM} at mean ages 13 and 15. Of particular note is the considerable degree of scatter of fasting insulin levels for any given level of M_{LBM}.
Fasting insulin was significantly correlated with HOMA (r = 0.99), QUICKI (r = −0.79 and −0.78), FGIR (r = −0.62 and −0.52), and (ln fasting insulin + ln triglycerides) (r = 0.88 and 0.87) at ages 13 and 15, respectively. The longitudinal correlations between the surrogate values at age 13 and insulin sensitivity at age 15 were modest and varied by group. The correlations for the entire cohort ranged from 0.15 to 0.17 (P = 0.006–0.1). The correlations for the ≥85th and <85th percentile groups were within the same range (0.12–0.18), but they were not significant (P = 0.9–0.25). The same pattern was seen for the correlations between the surrogates at age 13 and change in insulin sensitivity from age 13 to 15. In contrast, the correlations between change in the surrogate values from age 13 to 15 and change in insulin sensitivity from age 13 to 15 for the entire cohort (r = 0.24–0.25), the <85th percentile group (r = 0.24–0.26), and the ≥85th percentile group (r = 0.31–0.35) all were significant (P = 0.005–0.0001).
The correlations between the surrogate and clamp measures of insulin resistance are shown in Table 2. With the exception of FGIR (0.11) at mean age 15, the correlations were significantly different from 0 (P < 0.0001) for M, M_{LBM}, and M_{LBM}/ln SSI at both ages. In general, the correlations were slightly higher for the girls (r = 0.45–0.66) than for the boys (r = 0.37–0.51), but the differences were statistically significant (P < 0.03 for all) only at age 13. There were differences in the pattern of the correlations. In the total cohort, correlations between the surrogate measures and M_{LBM} were significantly lower than for M and M_{LBM}/ln SSI at age 13, with the exception of FGIR; however, at age 15 the correlations for the surrogates with M_{LBM}/ln SSI were similar to the correlations with M_{LBM} and significantly lower than the correlations with M.
Among the BMI subgroups, there were significantly higher correlations with M than with M_{LBM} for fasting insulin and HOMA and with M_{LBM}/ln SSI than with M_{LBM} for all the surrogates in the <85th percentile group at age 13. In the ≥85th percentile group at age 15, the correlations between the surrogates and M were significantly greater than with M_{LBM} or M_{LBM}/ln SSI. The correlations between surrogate measures and both M_{LBM} and M_{LBM}/ln SSI also changed with age. The correlations for M_{LBM} and M_{LBM}/ln SSI with fasting insulin, HOMA, QUICKI, and FGIR in the total cohort and in the ≥85th percentile BMI group decreased from mean age 13 to 15. In particular, the change was statistically significant for all the surrogate measures versus M_{LBM}/ln SSI in the total group and ≥85th percentile group. In general, the correlations in the ≥85th percentile group were higher than those in the <85th percentile group.
The ROC analysis showed there was no significant difference in the AUC among fasting insulin, HOMA, and QUICKI in prediction of dichotomous insulin resistance. The AUC values (0.771, 0.770, and 0.771, respectively) indicated only a modest capability to separate truepositive from falsepositive insulin resistance. For instance, a fasting insulin cut point of 22 μU/ml (the upper 10% for fasting insulin) identified only 12 of 32 participants in the lower 10% of M_{LBM} (true positive), but 22 participants with M_{LBM} above the lower 10% (false positive) also were included. The AUC results were similar when the definition for insulin resistance was increased to the lower 20% for M_{LBM} and also were similar for both the <85th percentile and ≥85the percentile BMI groups.
CONCLUSIONS—
This large cohort of randomly selected adolescents participating in a longitudinal study of insulin resistance offered the unique opportunity to compare surrogate measures of insulin resistance based on fasting insulin and glucose levels (fasting insulin, HOMA, QUICKI, and FGIR) and (ln fasting insulin + ln triglycerides) with the gold standard insulin clamp measure of insulin resistance at two separate ages. The results showed that the surrogate measures were only modestly correlated with the euglycemichyperinsulinemic insulin clamp and do not appear to offer any advantage over fasting insulin alone. This study further showed that the correlations vary with BMI, with lower correlations for M_{LBM} than for M, and higher correlations in the ≥85th percentile group compared with the <85th percentile group. When SSI was added to M_{LBM}, the correlations were greater than with M at age 13 but significantly lower than with M at age 15.
In general, HOMA is the most widely used of the surrogate measures in children. The high correlation between HOMA and fasting insulin (r = 0.99) in this study is not surprising, considering the HOMA formula [(fasting insulin × fasting glucose)/22.5] and the finding in adults of a 24fold variation in fasting insulin compared with a 1.8fold variation in fasting glucose (16). The same was true in this study with a 1.8fold variation in fasting glucose and a 53fold variation in fasting insulin. Fasting glucose also is maintained within a narrow range among obese children (3) and children with abnormal glucose tolerance tests (1). Studies in adults have reported comparisons between fasting insulin and HOMA similar to those in the present study (16,17). The relative influences of fasting insulin and glucose would be expected to be the same in the QUICKI and FGIR equations, and both of these also were significantly correlated with fasting insulin in this study, although the r values were lower; (ln fasting insulin + ln triglycerides) also did not appear to offer any advantage over fasting insulin, with high correlations with insulin resistance at age 13 but lower correlations at age 15.
The correlations between the surrogate measures and clampderived insulin resistance unadjusted for fatness (M) were ∼0.5 at both ages. However, after removal of fatness from the estimate of glucose disposal (M_{LBM}), the correlations became lower, particularly at mean age 15. Similar differences were found in a young adult AfricanAmerican cohort (18). Among the BMI subgroups, the greatest differences between M and M_{LBM} correlations were in the ≥85th percentile group at age 15. Fatness is omitted from the calculation of clampderived insulin resistance (4) because most glucose uptake is localized to skeletal muscle (19). Thus, the use of M rather than M_{LBM} appears to artificially improve the relation between the clamp and surrogate measures, particularly in fatter adolescents. It has been suggested that M_{LBM} should be corrected for SSI levels (4). Although M_{LBM}/ln SSI in this study increased the correlations with the surrogate measures, the effect was small relative to the correlations with M_{LBM}.
Previous studies have compared surrogate measures to clampderived insulin resistance with various results. In a cohort of 31 children aged 6–11 years, primarily prepubertal and obese, similar correlations to the present study were found for HOMA (r = −0.51) and for FGIR (r = 0.37), but there was a higher correlation for QUICKI (r = 0.69) (9). In a cohort of 131 prepubertal and pubertal children, correlations of 0.91 and 0.86 were found for fasting insulin, QUICKI, HOMA, and FGIR in whites and African Americans, respectively (10). Others have compared surrogate measures to the frequently sampled intravenous glucose tolerance test, also with disparate results. In a small study of 18 obese children, the correlations for fasting insulin, HOMA, QUICKI, and FGIR all were ∼0.9 (20), and in a study of 30 obese and 36 lean children the correlations with HOMA and QUICKI were <0.2 (21). In adults, the European Group for the Study of Insulin Resistance reported a correlation of 0.37 between fasting insulin and the insulin clamp in 1,140 subjects (4), and correlations of 0.41–0.53 were found between insulin resistance defined by the frequently sampled intravenous glucose tolerance test and fasting insulin, HOMA, and QUICKI (22). It is not clear why the correlations differ among the studies and, in particular, with regard to the present study, but the differences could be related to differences in age, pubertal status, BMI, or cohort size. We believe the results in the present cohort are accurate for pubertal adolescents because of the large cohort, selection of participants across the entire distribution of BMI, and use of the gold standard insulin clamp.
A new finding in this study is the higher correlations in the heavier (BMI ≥85th percentile) children. This finding also has been reported in adults in whom correlations with fasting insulin, HOMA, and QUICKI were ∼0.35 for normal weight, 0.55 for overweight, and 0.60 for obese subjects, respectively (16). Although an explanation for this difference between heavy and thin individuals is not readily apparent, it may be related to the fact that an increase in BMI is most commonly related to an increase in fatness, and insulin resistance could increase at a faster than linear rate as adiposity increases. As discussed previously (16), obesity is associated with higher levels of fasting insulin than those in normal weight individuals, even when the weight groups are matched for glucose intolerance or insulin resistance. This difference in fasting insulin also was found between the ≥85th and <85th percentile groups in the present study. It seems reasonable to suggest that higher levels of fasting insulin, representing altered insulin secretion or clearance, could influence the correlations between the surrogate measures, which are dependent on levels of fasting insulin, and the clamp. The low correlations in the <85th percentile group suggest that the use of surrogate measures to assess insulin resistance in thin children is particularly unreliable.
The results from this study demonstrate the dilemma faced in assessing the role of insulin resistance during childhood and adolescence. Although it generally is not feasible to conduct invasive tests for insulin resistance in large cohort studies, the correlation data, graphs, and ROC analyses show that fasting insulin and other surrogate measures are imprecise substitutes. This is not surprising, as the clamp measures only cellular glucose uptake, whereas fasting insulin represents the integrated effect of glucose and insulin release and clearance. That is not to say that fasting insulin may not be a marker of cardiovascular risk, but only that it is a poor representation of insulin resistance.
Acknowledgments
This study was supported by grants HL52851 and M01RR00400 from the National Institutes of Health.
Footnotes

Published ahead of print at http://care.diabetesjournals.org on 3 January 2008. DOI: 10.2337/dc071376.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C Section 1734 solely to indicate this fact.
 Accepted December 20, 2007.
 Received July 17, 2007.
 DIABETES CARE