| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Other Original Articles |
Indiana University School of Medicine (K.J.M., A.E.H., H.O.S., G.P., G.H., A.D.B.), Division of Endocrinology and Metabolism, Indianapolis, Indiana 46250; and Hypertension-Endocrine Branch (A.K., M.J.Q.), National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland 20892
Address all correspondence and requests for reprints to: Kieren J. Mather, M.D., FRCPC, Division of Endocrinology, Indiana University School of Medicine, CL 459, 541 North Clinical Drive, Indianapolis, Indiana 46202. E-mail: kmather{at}iupui.edu
Abstract
The objectives of this study were to evaluate test characteristics, such as normality of distribution, variation, and repeatability, of simple fasting measures of insulin sensitivity and to use the results to choose among these measures. Duplicate fasting samples of insulin and glucose were collected before 4 h of euglycemic hyperinsulinemic clamping using insulin infusion rates ranging from 40–600 mU/m2·min. Currently recommended estimates of insulin sensitivity, including the fasting insulin, 40/insulin, the homeostasis model assessment, the logarithmic transformation of the homeostasis model assessment, and the Quantitative Insulin Sensitivity Check Index, were evaluated. The normality of distribution and the variability of the tests (coefficient of variation and discriminant ratio) were compared between the measures and against the "gold standard" hyperinsulinemic clamp. Data from 253 clamp studies in 152 subjects were examined, including 79 repeated studies for repeatability analysis. In subjects ranging from lean to diabetic, the log transformed fasting measures combining insulin and glucose had normal distributions and test characteristics superior to the other simple indices (logarithmic transformation of the homeostasis model assessment coefficient of variation, 0.55; discriminant ratio, 13; Quantitative Insulin Sensitivity Check Index coefficient of variation, 0.05; discriminant ratio, 10) and statistically comparable to euglycemic hyperinsulinemic clamps (coefficient of variation, 0.10; discriminant ratio, 6.4). These favorable characteristics helped explain the superior correlations of these measures with the hyperinsulinemic clamps among insulin-resistant subjects. Furthermore, therapeutic changes in insulin sensitivity were as readily demonstrated with these simple measures as with the hyperinsulinemic clamp. The test characteristics of the logarithmic transformation of the homeostasis model assessment and the Quantitative Insulin Sensitivity Check Index are superior to other simple indices of insulin sensitivity. This helps explain their excellent correlations with formal measures both at baseline and with changes in insulin sensitivity and supports their broader application in clinical research.
THE PAST DECADE has seen a surge of interest in insulin resistance, both as an etiological factor in the pathogenesis of type 2 diabetes mellitus (1) and as a key component of the dysmetabolic cardiovascular syndrome (also known as the insulin resistance syndrome) (2). There is a need for simple, accessible tools for the measurement of insulin sensitivity. Classic steady state hyperinsulinemic clamps (3) arguably represent the current "gold standard." Both steady state and dynamic tests are time intensive for both the subject and the investigator, somewhat invasive, and incur significant costs because they require multiple insulin level measurements. These disadvantages have prevented the general application of these tests in the both clinical and research arenas.
Most large scale epidemiological studies have simply correlated fasting insulin levels with the outcomes of interest. More than 15 yr ago, mathematical modeling of the normal physiological balance of insulin and glucose produced the homeostasis model assessment (HOMA), which provided equations for estimating insulin resistance (HOMA-IR) and ß-cell function from simultaneous fasting measures of insulin and glucose levels (4). Although this estimate of insulin resistance has been used in some epidemiological studies (5, 6, 7, 8, 9, 10, 11), few clinical studies have relied on this measure alone because of reportedly poor accuracy compared with hyperinsulinemic clamps and a reliance on notoriously variable insulin measurements.
The direct relationships between insulin levels or HOMA-IR and clamp measures of insulin sensitivity are hyperbolic rather than linear (12, 13, 14, 15). Recently, a number of groups have suggested that improved correlations of fasting measures with glucose clamps might be obtained with linearizing transformations. Both logarithmic (6, 15, 16, 17, 18) and reciprocal transformations (15, 19) have been proposed. We hypothesized that the improved correlations reflected improvements in test characteristics and that examining these characteristics would support a rational choice between the various mathematical transformations. To this end, we evaluated the repeatability, variability, and discriminant power (20) of various calculated indices of insulin sensitivity and compared them to the hyperinsulinemic euglycemic clamp. These findings form the basis for the correlations between the clamps and the fasting indices, so these were compared as well.
Materials and Methods
Subjects were selected from a database of hyperinsulinemic
euglycemic clamp data collected at our institution. Three groups of
subjects existed in the database: lean, healthy controls (body mass
index 
27); obese nondiabetic (insulin-resistant) subjects (body
mass index >27); and subjects with type 2 diabetes mellitus (defined
according to American Diabetes Association criteria). Extremes of blood
pressure and lipid measures are routinely considered exclusion criteria
for our studies, so these subjects differed principally in their degree
of insulin resistance. We excluded clamp studies performed before
September 1993, at which time a commercial supersensitive insulin assay
was adopted for routine use in our core laboratory. The data collected
for these studies included duplicate fasting insulin and glucose
measurements (separated by 10 min) before the initiation of insulin
infusions. Standard measures of weight and height were performed, and
percentage body fat was measured by dual energy x-ray absorptiometry or
water displacement (for subjects >122 kg, the upper limit for the dual
energy x-ray absorptiometry machine). Diabetic subjects receiving oral
agents were withdrawn from therapy 2–3 wk before clamp studies.
Diabetic patients treated with insulin discontinued long-acting insulin
5 d before and short-acting insulin 24 h before studies.
We chose to divide subjects by clinical characteristics rather than with a post hoc definition of insulin sensitivity based on clamp outcome. Although this results in some overlap of degrees of insulin sensitivity across groups, it provides a better match with the more general prospective approach of classifying subjects on the basis of clinical characteristics, and any statistical effects of this overlap can be accounted for by examining continuous relationships within and across groups.
The database included euglycemic hyperinsulinemic (EH) clamp studies
performed at a range of insulin doses, and four levels were included in
the present analysis: 40, 120, 300, and 600
mU/m2·min (I40, I120, I300, and I600). Except
for insulin infusion rate, the design of clamp studies was uniform.
Subjects were studied after an overnight (10–14 h) fast. At least 30
min after placement of vascular accesses, an 
20-min baseline period
took place during which fasting insulin and glucose samples were taken
10 min apart. An unprimed 4-h insulin infusion was then applied, with a
variable infusion of 20% glucose in water to maintain euglycemia. The
glucose disposal rate (GDR) was calculated as the mean of the final two
20-min GDRs during the last 40 min of the 4-h clamp. Lean and obese
subjects were studied at all four clamp levels, whereas clamp studies
in diabetic subjects were available only at the higher two infusion
rates. Under these hyperinsulinemic conditions, hepatic glucose output
is assumed to be completely suppressed (21), so tracer
methods for the measurement of hepatic glucose output were not
used.
Insulin determinations were made using a dual-site RIA specific for human insulin with cross-reactivity with proinsulin less than 0.2%. The lower detection limit is 0.56 pmol/liter, and in our laboratory the interassay and intraassay coefficients of variation (CV) are 4.1% and 2.6%, respectively. Because the sensitive insulin assay is known to become increasingly unreliable as measured values approach the lower detection limit, we retested all samples with reported values less than 50 pmol/liter using an ultrasensitive human double antibody insulin assay (detection limit, 0.056 pmol/liter ; intratest and intertest CV, 12.4% and 9.0%), for a repeat correlation analysis.
Our data set included 13 young women with the polycystic ovarian syndrome who had undergone 3 months treatment with oral troglitazone (600 mg daily) and who had undergone I120 clamp studies as described above before and after this treatment. Using these data, the ability of the simple fasting measures to demonstrate the change in insulin sensitivity was compared with the EH clamp.
Test characteristics
Repeat fasting insulin and glucose measures were available for all subjects who had participated in more than one clamp study. The minimum interval for such repeat studies was 4 wk. Where subjects had undergone treatment with identical clamp conditions on more than one occasion, only those studies that had been repeated within 6 months were included. Furthermore, if any significant change in body mass index, drug therapy, or metabolic status (e.g. the new development of type 2 diabetes mellitus) had occurred, such repeat studies were excluded from the repeatability analysis.
Four assessments of test characteristics were undertaken. In addition
to assessing for a uniform distribution of the data itself, uniform
distribution of the error on repeat testing across the measurement
range (homoscedasticity) was investigated using Altman-Bland plots.
Variability and repeatability were assessed using the CV and a newly
proposed measure, the discriminant ratio (DR) (20). For
the CV, the standard formula was used:
 |
Unlike the CV, the interpretation of the DR is not dependent on the
absolute value of the population mean. Also, the DR includes both of
the principal sources of systematic error, i.e.
between-subject and within-subject error. It is easily calculated from
a repeated measures ANOVA using the error terms
(MSB, between-subject error term;
MSW, within-subject error term across repeat
studies) derived from a standard repeated measures ANOVA table for the
repeated tests:
 |
2 calculations required
for these comparisons were carried out using the UCLA online
statistical calculators (http://www.ucla.stat.edu). This measure also
provides a means for correcting an observed correlation between tests
for the known measurement error of each test (giving an improved
estimate of the true underlying correlation between the two tests)
using a correction factor incorporating the observed DRs of the two
measures being compared (20). Estimates of insulin sensitivity
Test characteristics were determined for 1) the fasting insulin level itself; 2) 40/insulin (µU/ml); 3) the logarithm of fasting insulin (logInsulin); 4) the HOMA-IR; 5) the logarithm of the HOMA-IR (logHOMA-IR); and 6) the Quantitative Insulin-Sensitivity Check Index (QUICKI) as well as for the clamp-derived GDR.
The HOMA-IR was first put forward in 1985 by Matthews et al.
(4). The authors recommended using triplicate fasting
measures of insulin (reported in µU/ml, at the time measured with
single-site antibody assays) and glucose (in mmol/liter ). A constant
was applied to correct the value to unity in normal subjects, providing
the following formula:
 |
 |
Correlations with GDR
Comparisons of the estimates of insulin sensitivity provided by each of the above measures were performed relative to the gold standard measure provided by the GDR. Although it is possible that the GDR is underestimated in the I40 clamps as a result of an unmeasured contribution from hepatic glucose production, this concern is minimized by the 3- to 4-h duration of the clamp procedure (21). This permits a reliance on the GDR measures at all insulin infusion rates as the standard for comparison. The GDR and fasting measurements were derived from values measured in each individual on the same day. Pearson’s correlation coefficients, and the r-to-z estimate of statistical significance, were calculated using StatView 5.0 for Macintosh (SAS Institute, Inc., Chicago, IL). The adjusted correlations, taking into account the measured variability of the tests themselves, were calculated according to the method of Levy et al. (20).
Results
Subjects
We evaluated 253 clamp studies in 152 subjects. The
characteristics of the subjects are detailed in Table 1
. Three of the 11 diabetic subjects were
treated with insulin, 1 was treated with a sulfonylurea medication, and
the others were newly diagnosed. The obese subjects and type 2 diabetic
subjects had lipid levels and blood pressure within the normal range,
but they had slightly higher low density cholesterol, lower high
density cholesterol, higher triglycerides, and higher blood pressure
than the lean subjects. This is consistent with the known
characteristics of the insulin resistance syndrome. The racial
distribution was 91 (59%) Caucasian, 61 (40%) African American, and 3
(1%) Mexican American. The 48 women studied included 13 with a known
diagnosis of the polycystic ovarian syndrome.
|
Test characteristics
Logarithmic transformations but not inversions normalized the
distribution of the data based on fasting insulin levels. The
logarithmic transformations also served to normalize the distribution
of error across the range of measurements (Fig. 1). The Altman-Bland plot for insulin
alone (Fig. 1, top right) is clearly heteroscedastic (the
variation in the measurement increases across the range of measured
values), and although the two-test correlation of insulin appears
satisfactory, it is heavily dependent on a minority of data points at
the upper end of the range. By contrast, the logHOMA-IR and QUICKI have
much more uniform variability across the range of values and good
two-test correlations, with more uniform distributions of measured
values. GDR (Fig. 1, top left) has these desirable
attributes without mathematical transformations.
|
). The difficulties of depending on CV
alone were apparent in comparing logHOMA-IR and QUICKI, which had
markedly different CV despite their simple mathematical relationship.
This counterintuitive difference results from the effects of
logarithmic and inversion transformations on numbers near one and zero.
This makes the CV difficult to interpret, but intuitively the inherent
repeatability of these measures must mirror each other. The DRs for
these two tests were superior to all of the other simple fasting
measures and statistically equivalent to each other and to the
clamp-derived GDR (Table 2).
|
).
The alternative approaches of using the insulin concentration alone,
the inversion of insulin, or the untransformed HOMA-IR suffered from
the heteroscedasticity inherent in the insulin measurement itself. The
CV of these values simply reflected the variability of the insulin
measurement (Table 2). The DRs of these values were no better than that
of insulin alone. The ultrasensitive insulin assay made no significant
improvements to any of the above characteristics (data not shown).
In summary, the best combinations of these characteristics (a normal distribution, a favorable distribution of measurement error across the range of values, a low CV, and a high DR) was seen with the GDR derived from EH clamps. The heteroscedasticity of insulin-based indices was improved by both inversion and logarithmic transformations, although only the latter produced a normal distribution of data. The apparent effects of these transformations on the CV are in part an artifact of the mathematics, and these divergent results in fact do not (and logically cannot) represent a significant alteration in this measure of variability. The transformed HOMA and QUICKI values, however, both showed improved DRs equivalent to that of the clamp-derived GDR. This marks these two measures as superior to the other fasting measures of insulin sensitivity.
Correlations to EH clamps
This theoretical grounding informs the assessment of the
relationships between these fasting indices and the EH clamp. Simple
correlations depend on a normal distribution of the related variables
and of their measurement error. These characteristics were achieved by
logarithmic transformation, validating correlation analysis with this
subset of the fasting measures. The correlations of these measures with
GDR, and the corrected correlations after adjustment for the
variability of the tests, are presented in Table 3. Untransformed fasting insulin,
40/insulin, and HOMA-IR had only moderately inferior correlation
coefficients (data not shown), but they exhibited nonlinear
relationships with GDR. Although the observed correlations differed
markedly by degree of obesity (Table 3), they were unchanged overall
when examined with the groups divided by age (younger or older than 45
yr: I40, P = 0.19; I120, P = 0.39;
I300, P = 0.86; I600, 0.60; age <45 yr, r
= -0.466, age >45 yr, r = -0.878, P = 0.04) or
gender (I40, P = 0.13; I120, P = 0.20;
I300, P = 0.39; I600, P = 0.32).
Similarly, there was no apparent effect of race on the correlation
(I40, P = 0.54; I120, P = 0.69). The
best correlations overall were seen in the diabetic subjects when
compared with 600 mU/m2·min insulin infusions,
although this includes a small number of subjects. Adjustment of these
correlations for the measured variability in the test, to better
estimate the true underlying correlation, produced small improvements
in the estimated correlation (Table 3).
|
). The correlations with logHOMA-IR and
QUICKI were in the 0.35–0.40 range, achieving statistical significance
only under I40 conditions (n = 69). Obese subjects, in contrast,
exhibited very good correlations with fasting measures of insulin
sensitivity (Table 3). The correlations observed at different clamp
levels ranged from 0.46–0.73 (adjusted, 0.47–0.88; Table 3), although
where fewer subjects were studied these correlations did not achieve
statistical significance. Despite the relatively small number of
subjects with type 2 diabetes mellitus included in the data set,
excellent correlations of logHOMA-IR and QUICKI with GDR were observed
(Table 3), with r values as high as 0.94. These subjects were all obese
and had increases in both fasting insulin and glucose levels (Table 1). The use of the ultrasensitive insulin assay in subjects with low insulin values produced a small leftward shift in the reported insulin values, particularly in those samples in which the originally reported value was less than 30 pmol/liter. The logHOMA-IR and QUICKI derived using these results gave only small improvements in the observed correlations at less than 40 mU (lean subjects: logHOMA-IR, -0.385 to -0.411; QUICKI, 0.359 to 0.389). Under I120 conditions, the correlations were not improved.
Tracking changes in insulin sensitivity
Pharmacological intervention to improve insulin sensitivity using
troglitazone (n = 13) produced the changes in GDR
presented in Fig. 2. The concurrent
changes in fasting insulin and glucose resulted in changes in QUICKI
and logHOMA-IR. The GDR (I120 clamps) increased from 5.56 ± 0.64
to 7.29 ± 0.88 mg/kg·min (31% change; P =
0.005), whereas the QUICKI increased from 0.308 ± 0.007 to
0.325 ± 0.007 mg/kg·min (23% change; P <
0.001) and logHOMA-IR decreased from 0.65 ± 0.07 to 0.50 ±
0.07 mg/kg·min (6% change; P < 0.001). As an
expected extension of the comparable discriminant powers of these
tests, in this small group of patients these changes could be
distinguished equally well with the simple measures as with the
hyperinsulinemic clamps.
|
We have evaluated the test characteristics of various estimates of insulin sensitivity based on fasting plasma insulin and glucose and compared them with those of the EH clamp. The distribution of values and of measurement error across the range of measured values was assessed, and measures of repeatability, including the CV and the DR, were calculated and compared. The logarithmic transformations achieved the intended normalization of the insulin-dependent data. The measures of variability suggested that the logarithmic transformations (logHOMA-IR and QUICKI) were preferable to the other fasting measures, particularly with regard to the DR.
These characteristics help explain and support the recent favorable comparisons of logHOMA-IR with GDR (13), and we were able to confirm this in our own data set by examining the correlations of the various fasting measures of insulin sensitivity with GDR across a wide range of clamp conditions. The untransformed estimates had the disadvantage of nonlinear relationships in the correlations, in addition to their worse repeatability characteristics. Remarkably good correlations were observed for the logInsulin, logHOMA-IR, and QUICKI with GDR in obese and type 2 diabetic subjects, despite the small number of the latter. In contrast to previous reports, we found the correlations observed among lean subjects to be inferior compared with those observed among obese and diabetic subjects. This is in many ways an expected statistical consequence of the greater variability of current insulin assays in the lower end of the range, separate from any physiological explanations. Unfortunately, the newer ultrasensitive assay did little to improve this situation, as might be expected if the variability was attributable to biological rather than test variation.
Finally, the discriminant power and correlations of logHOMA-IR and QUICKI with GDR among insulin-resistant subjects allowed the tracking of changes in insulin after treatment with troglitazone comparably to the EH clamp.
Comparing the simple measures of insulin sensitivity
The logarithmic transformations produced the desired normalization of the insulin-dependent data, including the measurement error, and compressed the range of data to reduce the effect of extreme values. This validates the application of statistical comparisons that assume underlying normal distributions. These effects in turn allowed improved linear modeling of relationships to GDR with the log-transformed measures.
Historically, the repeatability of a test has been expressed as the CV.
Although conceptually simple, this measure carries disadvantages; most
notably, when the mean value is close to zero, even a test with good
precision may have a high CV (20). This was observed in
the present report. For lean subjects, a comparatively large CV for
logHOMA-IR results from a SD of 0.13, with the numerically
small mean of 0.08. In fact, this measure is equally variable across
the groups, as is evident in the comparable SDs (Table 2).
Furthermore, the QUICKI, which is related to the logHOMA-IR by
inversion and a constant term, appeared to have a superior CV, despite
the inherently linked variability of these two measures. The DR, a new
measure of variability that takes into account both the between-subject
and within-subject variations (20), was comparable for the
QUICKI and logHOMA-IR. Importantly, the DRs of these two measures and
the GDR were statistically comparable. Therefore, the logHOMA-IR and
QUICKI are as powerful at discriminating differences in their estimates
of insulin sensitivity across the population as the GDR is at
discriminating the formally measured insulin sensitivity.
Of course, the utility of these measures in estimating insulin sensitivity per se depends on the underlying correlation of the estimate and the formal measure. These correlations were excellent among the obese and diabetic subjects using the logarithmically transformed variables, reflecting the overall improvements in test characteristics. The comparatively poor correlations with GDR of all of these estimates in lean nondiabetic subjects suggests that they are imperfect surrogate measures in insulin-sensitive populations. Better correlations (r = 0.47) have been reported in nonobese subjects using QUICKI and the SIclamp, although with this technique the correlations among obese and diabetic subjects were again superior (r = 0.89 and 0.7, respectively) (15). This finding is also in contrast to a recent report (13) of equal correlations between logHOMA-IR and GDR in lean and obese as well as diabetic and nondiabetic subsets of subjects. Importantly, the latter data set apparently includes lean diabetic subjects, who would bias toward a correlation among lean subjects. Also, those subjects were studied at insulin concentrations that did not completely suppress hepatic glucose output in all cases; therefore, they provided a measure of hepatic as well as peripheral insulin sensitivity. An analogous balance may have contributed to our finding of a significant relationship in lean subjects only at the lowest insulin infusion rate. Overall, the correlations among lean subjects appear significant but less robust than those among obese and diabetic insulin-resistant subjects.
It is tempting to argue that the logInsulin is a simpler and more accessible correlate of insulin sensitivity than either the QUICKI or the logHOMA-IR, in view of the comparable correlations between these measures among lean and obese subjects. Certainly, under any circumstances in which the glucose was completely normal, this value in effect becomes a constant term added to the equation and therefore contributes little. However, the inclusion of glucose makes the formulas more generalizable to all circumstances with variable glucose levels, including the range of subdiabetic glucose readings seen in some obese subjects. Also, in accounting for this variation, the repeatability characteristics of the test are improved, as is evident in the superior DRs of QUICKI and logHOMA-IR compared with logInsulin. These considerations support the selection of the QUICKI or logHOMA-IR over measures based on insulin alone when concurrent glucose values are accessible.
The data set included a comparatively small number of diabetic subjects, so conclusions regarding the comparisons of test characteristics and correlations in this group must be weighted accordingly. However, these subjects represent a natural extension of the range of insulin resistance, and the results are consistent with the trend suggested by the larger samples of lean and obese subjects. Also, despite the small numbers, the correlations themselves were the strongest of all the groups, perhaps because the QUICKI and logHOMA-IR account for both the insulin and glucose levels. The calculation of the DR of necessity includes all subjects; therefore, it is strengthened by the inclusion of even this small number of subjects at one extreme of the range.
In summary, the logHOMA-IR and QUICKI were the superior simple measures of insulin sensitivity with regard to test characteristics, and this was reflected in very good correlations across a range of EH clamps among insulin-resistant obese and diabetic subjects.
Limitations of insulin assays
All of these simple fasting estimates of insulin sensitivity are highly dependent on the fasting insulin level. The original description of the HOMA included significant caveats in this regard (4). Although assays have improved, these remain imperfect tests with regard to test-to-test variability (23, 24). Furthermore, the underlying biological variability in insulin levels, arising from the combination of its short serum half-life, the known cyclicity of insulin secretion (25), and the rapid responsiveness to changes in hormonal and metabolic milieu, will remain a source of variation regardless of improvements in insulin assays. The use of a newer, ultrasensitive insulin assay did not improve any of the characteristics of the tests, presumably because it did not improve any of these sources of variability. To account for some of this variability, Matthews et al. (4) recommended the mean of three insulin samples, taken over a 15-min period, be used. Our data include two rather than three values, taken 10 min apart, and this produced acceptable results. Cost savings can be obtained by pooling multiple samples before measurement, but the use of at least two samples over a 10- to 15-min period remains prudent. These efforts to reduce the variability of the measurement are necessary to achieve the test characteristics described herein.
Applications
Three main categories of studies are candidates for the use of
measures of insulin sensitivity: metabolic studies, intervention
studies, and epidemiological studies. With metabolic studies, insulin
sensitivity will most likely be a primary end point, and the use of
simplified estimates is probably not worthwhile given the availability
of expertise in formal EH clamp measurements, which remain the gold
standard. Intervention studies targeting insulin resistance or the
insulin resistance syndrome are becoming increasingly common, with the
advent of PPAR-
agonists and renewed interest in other interventions
such as metformin and changes in diet and exercise. Many such studies
will be primarily interested in the changes in insulin sensitivity
per se, and again the EH clamp should be the preferred
method for these studies.
The time and cost of formal clamp testing is prohibitive in large scale epidemiological studies, and to date these studies have largely chosen serum insulin as a surrogate measure of insulin sensitivity. For these studies, the logHOMA-IR or QUICKI would be appropriate surrogate measures of insulin sensitivity, as has been suggested (26). Importantly, large prospective databases already exist with sufficient data to examine associations between these simple indices of insulin resistance and cardiovascular outcomes. However, one important caveat is suggested by our findings: the superior estimate of insulin sensitivity provided among an insulin-resistant subset could systematically affect any apparent correlations with other measures in the population. This will potentially need to be accounted for in statistical analyses of such data sets.
In most interventional studies, changes in insulin sensitivity will be a secondary interest, with the effects on (for example) cardiovascular parameters being the primary end points. In these situations, the logHOMA-IR or QUICKI (using sensitive insulin-specific assays, with at least duplicate samples) would provide a useful, minimally invasive, relatively inexpensive surrogate for the EH clamp. This is further supported by the demonstrated ability of these measures to reveal clinically relevant changes in insulin sensitivity in comparatively small numbers of subjects.
Conclusion
Simple mathematical combinations of fasting insulin and glucose measures (logHOMA-IR or QUICKI) provide estimates of insulin sensitivity with variability and discriminant power comparable to those of EH clamps and superior to measures based on insulin alone. In accounting for both the glucose and insulin levels, these measures are more generalizable to the full range of metabolic conditions associated with insulin resistance. This underlies the excellent correlations of these measures with clamps seen in obese and diabetic insulin-resistant subjects. Furthermore, changes in insulin sensitivity can be demonstrated using these tools in comparatively small groups of subjects. Our results suggest that caution needs to be exercised in applying these estimates to groups including insulin-sensitive subjects. With this caveat, the current report provides the statistical underpinnings for the broader application of these inexpensive, accessible estimates of insulin sensitivity in clinical investigations and large scale epidemiological studies.
Footnotes
This work was supported by NIH Grants DK42469, DK20452, and MO1-RR750-19; and a Veterans Affairs Merit Review Award. K.J.M. was supported by an Alberta Heritage Fund for Medical Research clinical research fellowship.
Abbreviations: CV, Coefficient(s) of variation; DR, discriminant ratio; EH, euglycemic hyperinsulinemic; GDR, glucose disposal rate; HOMA-IR, homeostasis model assessment of insulin resistance; I40, I120, I300, and I600, insulin doses of 40, 120, 300, and 600 mU/m2·min; logHOMA-IR, logarithmic transformation of HOMA-IR; logInsulin, logarithm of fasting insulin; QUICKI, Quantitative Insulin-Sensitivity Check Index.
Received January 16, 2001.
Accepted June 13, 2001.
References
This article has been cited by other articles:
 |
|
 |
|
  S. K. Hutchison, C. Harrison, N. Stepto, C. Meyer, and H. J. Teede Retinol-Binding Protein 4 and Insulin Resistance in Polycystic Ovary Syndrome Diabetes Care, July 1, 2008; 31(7): 1427 - 1432. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J Polak, Z Kovacova, C Holst, C Verdich, A Astrup, E Blaak, K Patel, J M Oppert, D Langin, J A Martinez, et al. Total adiponectin and adiponectin multimeric complexes in relation to weight loss-induced improvements in insulin sensitivity in obese women: the NUGENOB study. Eur. J. Endocrinol., April 1, 2008; 158(4): 533 - 541. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. B. Sverrisdottir, T. Mogren, J. Kataoka, P. O. Janson, and E. Stener-Victorin Is polycystic ovary syndrome associated with high sympathetic nerve activity and size at birth? Am J Physiol Endocrinol Metab, March 1, 2008; 294(3): E576 - E581. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Lee, R. Muniyappa, X. Yan, H. Chen, L. Q. Yue, E.-G. Hong, J. K. Kim, and M. J. Quon Comparison between surrogate indexes of insulin sensitivity and resistance and hyperinsulinemic euglycemic clamp estimates in mice Am J Physiol Endocrinol Metab, February 1, 2008; 294(2): E261 - E270. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Muniyappa, S. Lee, H. Chen, and M. J. Quon Current approaches for assessing insulin sensitivity and resistance in vivo: advantages, limitations, and appropriate usage Am J Physiol Endocrinol Metab, January 1, 2008; 294(1): E15 - E26. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. Ugur-Altun, A. Altun, S. Guldiken, E. Tatli, M. Kara, and A. Tugrul Silent Myocardial Ischemia in Middle-Aged Asymptomatic Patients With Type 2 Diabetes in Turkish Population Angiology, November 1, 2007; 58(5): 535 - 542. [Abstract] [PDF]
|
|
|
|
 |
|
 P. Velasquez-Mieyer, C. P. Neira, R. Nieto, and P. A. Cowan Review: Obesity and cardiometabolic syndrome in children Therapeutic Advances in Cardiovascular Disease, October 1, 2007; 1(1): 61 - 81. [Abstract] [PDF]
|
|
|
|
|
|
A. Zanchi, A. G. Dulloo, C. Perregaux, J.-P. Montani, and M. Burnier Telmisartan prevents the glitazone-induced weight gain without interfering with its insulin-sensitizing properties Am J Physiol Endocrinol Metab, July 1, 2007; 293(1): E91 - E95. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Zanchi, C. Perregaux, M. Maillard, D. Cefai, J. Nussberger, and M. Burnier The PPAR{gamma} agonist pioglitazone modifies the vascular sodium-angiotensin II relationship in insulin-resistant rats Am J Physiol Endocrinol Metab, December 1, 2006; 291(6): E1228 - E1234. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Barinas-Mitchell, L. H. Kuller, K. Sutton-Tyrrell, R. Hegazi, P. Harper, J. Mancino, and D. E. Kelley Effect of Weight Loss and Nutritional Intervention on Arterial Stiffness in Type 2 Diabetes Diabetes Care, October 1, 2006; 29(10): 2218 - 2222. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Gambineri, V. Vicennati, S. Genghini, F. Tomassoni, U. Pagotto, R. Pasquali, and B. R. Walker Genetic Variation in 11{beta}-Hydroxysteroid Dehydrogenase Type 1 Predicts Adrenal Hyperandrogenism among Lean Women with Polycystic Ovary Syndrome J. Clin. Endocrinol. Metab., June 1, 2006; 91(6): 2295 - 2302. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. Chalasani, R. Vuppalanchi, N. S. Raikwar, and M. A. Deeg Glycosylphosphatidylinositol-Specific Phospholipase D in Nonalcoholic Fatty Liver Disease: A Preliminary Study J. Clin. Endocrinol. Metab., June 1, 2006; 91(6): 2279 - 2285. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. Tang, J. Glanville, C. J. Hayden, D. White, J. H. Barth, and A. H. Balen Combined lifestyle modification and metformin in obese patients with polycystic ovary syndrome. A randomized, placebo-controlled, double-blind multicentre study Hum. Reprod., January 1, 2006; 21(1): 80 - 89. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. S. Fox, M. G. Larson, E. P. Leip, J. B. Meigs, P. W.F. Wilson, and D. Levy Glycemic Status and Development of Kidney Disease: The Framingham Heart Study Diabetes Care, October 1, 2005; 28(10): 2436 - 2440. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. A.R. Doi, M. Al-Zaid, P. A. Towers, C. J. Scott, and K. A.S. Al-Shoumer Irregular cycles and steroid hormones in polycystic ovary syndrome Hum. Reprod., September 1, 2005; 20(9): 2402 - 2408. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A.A. Lteif, K. Han, and K.J. Mather Obesity, Insulin Resistance, and the Metabolic Syndrome: Determinants of Endothelial Dysfunction in Whites and Blacks Circulation, July 5, 2005; 112(1): 32 - 38. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. J. Karne, H. Chen, G. Sullivan, and M. J. Quon Letter re: Limited Accuracy of Surrogates of Insulin Resistance during Puberty in Obese and Lean Children at Risk for Altered Glucoregulation J. Clin. Endocrinol. Metab., July 1, 2005; 90(7): 4418 - 4419. [Full Text] [PDF]
|
|
|
|
|
|
A. Gambineri, L. Patton, R. De Iasio, B. Cantelli, G. E. Cognini, M. Filicori, A. Barreca, E. Diamanti-Kandarakis, U. Pagotto, and R. Pasquali Efficacy of Octreotide-LAR in Dieting Women with Abdominal Obesity and Polycystic Ovary Syndrome J. Clin. Endocrinol. Metab., July 1, 2005; 90(7): 3854 - 3862. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Chen, G. Sullivan, and M. J. Quon Assessing the Predictive Accuracy of QUICKI as a Surrogate Index for Insulin Sensitivity Using a Calibration Model Diabetes, July 1, 2005; 54(7): 1914 - 1925. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. K. Koh, S. H. Han, M. J. Quon, J. Yeal Ahn, and E. K. Shin Beneficial Effects of Fenofibrate to Improve Endothelial Dysfunction and Raise Adiponectin Levels in Patients With Primary Hypertriglyceridemia Diabetes Care, June 1, 2005; 28(6): 1419 - 1424. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. K. Koh, M. J. Quon, S. H. Han, J. Y. Ahn, D. K. Jin, H. S. Kim, D. S. Kim, and E. K. Shin Vascular and Metabolic Effects of Combined Therapy With Ramipril and Simvastatin in Patients With Type 2 Diabetes Hypertension, June 1, 2005; 45(6): 1088 - 1093. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
V. Jayagopal, E. S. Kilpatrick, S. Holding, P. E. Jennings, and S. L. Atkin Orlistat Is as Beneficial as Metformin in the Treatment of Polycystic Ovarian Syndrome J. Clin. Endocrinol. Metab., February 1, 2005; 90(2): 729 - 733. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. J. Huggett, J. Burns, A. F. Mackintosh, and D. A.S.G. Mary Sympathetic Neural Activation in Nondiabetic Metabolic Syndrome and Its Further Augmentation by Hypertension Hypertension, December 1, 2004; 44(6): 847 - 852. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Engeli, J. Janke, K. Gorzelniak, J. Bohnke, N. Ghose, C. Lindschau, F. C. Luft, and A. M. Sharma Regulation of the nitric oxide system in human adipose tissue J. Lipid Res., September 1, 2004; 45(9): 1640 - 1648. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Gambineri, C. Pelusi, E. Manicardi, V. Vicennati, M. Cacciari, A. M. Morselli-Labate, U. Pagotto, and R. Pasquali Glucose Intolerance in a Large Cohort of Mediterranean Women With Polycystic Ovary Syndrome: Phenotype and Associated Factors Diabetes, September 1, 2004; 53(9): 2353 - 2358. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. H. Kim, F. Abbasi, and G. M. Reaven Impact of Degree of Obesity on Surrogate Estimates of Insulin Resistance Diabetes Care, August 1, 2004; 27(8): 1998 - 2002. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. J. Karne, H. Chen, and M. J. Quon Diagnosing Insulin Resistance by Simple Quantitative Methods in Subjects With Normal Glucose Metabolism: Response to Ascaso et al. Diabetes Care, May 1, 2004; 27(5): 1247 - 1248. [Full Text] [PDF]
|
|
