A simple intravenous glucose tolerance test for assessment of insulin sensitivity
 Robert G Hahn^{1, 2}Email author,
 Stefan Ljunggren^{2, 3},
 Filip Larsen^{4} and
 Thomas Nyström^{3}
DOI: 10.1186/17424682812
© Hahn et al; licensee BioMed Central Ltd. 2011
Received: 14 April 2011
Accepted: 2 May 2011
Published: 2 May 2011
Abstract
Background
The aim of the study was to find a simple intravenous glucose tolerance test (IVGTT) that can be used to estimate insulin sensitivity.
Methods
In 20 healthy volunteers aged between 18 and 51 years (mean, 28) comparisons were made between kinetic parameters derived from a 12sample, 75min IVGTT and the M_{bw} (glucose uptake) obtained during a hyperinsulinemic euglycemic glucose clamp. Plasma glucose was used to calculate the volume of distribution (V_{d}) and the clearance (CL) of the injected glucose bolus. The plasma insulin response was quantified by the area under the curve (AUC_{ins}). Uptake of glucose during the clamp was corrected for body weight (M_{bw}).
Results
There was a 7fold variation in M_{bw}. Algorithms based on the slope of the glucoseelimination curve (CL/V_{d}) in combination with AUC_{ins} obtained during the IVGTT showed statistically significant correlations with M_{bw}, the linearity being r^{2} = 0.630.83. The best algorithms were associated with a 2575^{th} prediction error ranging from 10% to +10%. Sampling could be shortened to 3040 min without loss of linearity or precision.
Conclusion
Simple measures of glucose and insulin kinetics during an IVGTT can predict between 2/3 and 4/5 of the insulin sensitivity.
Introduction
The best established methods of measuring insulin resistance are the hyperinsulinemic euglycemic glucose clamp and the intravenous glucose tolerance test (IVGTT), of which former is the "gold standard" [1–3]. These methods have a long history as investigative tools in diabetes research but are too cumbersome to be used during surgery, although insulin resistance develops in this setting [4, 5].
The aim of this project is to evaluate a simplified IVGTT test that lasts for 30, 40 or 75 min. This test is less labourintensive than both the glucose clamp and the conventional IVGTT. Analysis of the data is based on a comparison between the "strength" of the insulin response and the elimination kinetics of glucose. A commonly used expression for the "strength" of a physiological factor is the area under the curve (AUC), which was applied here on insulin, while the slope of the elimination curve for glucose served to quantify the "effect".
The hypothesis was that the test could predict insulin resistance with the same or higher precision than the "minimal model" (MINMOD) which is typically based on a longer IVGTT and quite demanding mathematically [6, 7]. We assessed this objective by comparing the simplified IVGTT with the result of the glucose clamp in 20 healthy volunteers.
Materials and methods
Baseline data and key results for the IVGTT and the glucose clamp.
Parameter  Mean (SD), or median (25^{th}75^{th} percentiles)  Unit 

Health status  
Body mass index  23.4 (2.3)  kg/m^{2} 
HbA1c  44 (0.5)  mmol/mol 
Blood Hb concentration  126 (14)  mmol/L; 
Serum creatinine concentration  83 (3)  μmol/L 
Serum sodium and potassium concentrations  141 (2); 3.9 (0.3)  mmol/L 
IVGTT  
Plasma glucose, baseline  4.8 (0.5)  mmol L^{1} 
Plasma insulin, baseline  21 (1224)  pmol L^{1} 
Volume of distribution (V_{d})  14.0 (6.5)  L 
per kg body weight  0.20 (0.09)  L kg^{1} 
Clearance (CL)  0.63 (0.26)  L min^{1} 
per kilo body weight  9.3 (3.8)  ml min^{1} kg^{1} 
Insulin sensitivity (S_{I}) of MINMOD  16 (732)  10^{5} L pmol^{1} min^{1} 
Glucose effectiveness (S_{G}) in MINMOD  13 (526)  10^{3} min^{1} 
Glucose clamp  
Plasma glucose, baseline  5.0 (1.0)  mmol L^{1} 
Plasma insulin, baseline  16 (730)  pmol L^{1} 
Plasma glucose, mean 90120 min  5.7 (0.3)  mmol L^{1} 
Plasma insulin, mean 90120 min  167 (34)  pmol L^{1} 
Glucose metabolism, M, 90120 min  3.1 (1.2)  mmol min^{1} 
M_{bw} = per kg body weight  45 (15)  μmol min^{1} kg^{1} 
Euglycemic hyperinsulinemic clamp
The subjects reported at the laboratory between 7.308.00 AM. A superficial dorsal hand vein was cannulated in retrograde direction with a small threeway needle and kept patent by repeated flushing with saline solution. The hand and lower arm were warmed by a heating pad for intermittent sampling of arterialized venous blood for glucose determination (Hemocue, Ängelholm, Sweden). In the opposite arm an intravenous catheter was inserted into the left antecubital vein for insulin and glucose infusion.
During the 120min test, insulin 20 mU · BSA m^{2} · min^{1} (Human Actrapid, NovoNordisk A/S, Bagsverd, Denmark) was infused along with 20% dextrose (Fresenius Kabi, Uppsala, Sweden). Baseline blood samples were drawn and the euglycemic hyperinsulinemic clamp was initiated by infusion of a bolus dose of insulin for 4 minutes followed by a stepwise increase in glucose for 10 min. The glucose infusion rate was adjusted to keep the subjects' blood glucose level constant at 5 mmol/L on the basis of arterialized samples withdrawn every 5 min from the dorsal hand vein catheter [8]. The infusion rate during the last 30 min, after correction for body weight, was taken to represent the metabolism of glucose (M_{bw}) [1–3].
Intravenous glucose tolerance test
On the second occasion, 12 days apart from the clamp study and after 12 h of fasting, a regular intravenous glucose tolerance test (IVGTT) was performed to determine the early insulin response phase (010 min), as well as the areaunderthecurve for insulin (AUC _{ ins } being total insulin and ΔAUC _{ ins } above baseline) and Cpeptide for up to 75 minutes. A bolus of glucose (300 mg/kg in a 30% solution) was given within 60 sec into the antecubital vein. Blood was sampled from the contralateral antecubital vein at 0, 2, 4, 6, 8, 10, 20, 30, 40, 50, 60 and 75 min for assessment of the plasma glucose, insulin, and Cpeptide concentrations. Plasma glucose was measured by the glucose oxidase method used by the hospital's routine laboratory. Plasma insulin and Cpeptide were measured using ELISA kits (Mercodia AB, Uppsala, Sweden).
Calculations
where G_{ b } is the baseline glucose, V_{d} is the volume of distribution, CL the clearance and CL/V_{d} the slope of the glucose elimination curve. The halflife (T_{1/2}) of the exogenous glucose load was obtained as (ln 2 V_{d} /CL). The AUC for plasma insulin was calculated by using the linear trapezoid method.
where S_{I} = glucose sensitivity, S_{G} = glucose effectiveness, X(t) is insulin action in the interstitial fluid space, and F(t) a function for the elevation of plasma insulin above the basal level. p_{2} is the removal rate of insulin from the interstitial fluid space while p_{3} describes the movement of circulating insulin to the interstitial space.
The best estimates for the unknown parameters in these models were estimated for each of the 20 experiments individually by nonlinear leastsquares regression. No weights were used. The mathematical software was Matlab R2010a (MathWorks, Natick, MA, USA).
where CS_{1} a surrogate measure for insulin sensitivity, K_{G} is the slope of the glucose elimination curve (same as CL/V_{d}) and T is the time after 10 min.
Statistics
The results were presented as mean and standard deviation (SD) and, when there was a skewed distribution, as the median (25^{th}75^{th} percentile range). Simple or multiple linear regression analysis, in which r^{2} is the coefficient of determination, was used to express "linearity" when studying the relationship between the M_{bw} of the glucose clamp (control) and various algorithms for insulin sensitivity derived from data collected during the IVGTT. The error in the prediction of M_{bw} associated with each regression analysis was obtained as [100% (fittedmeasured)/measured]. The change in prediction error obtained by restricting the analysis period from 75 to 40 and 30 min was tested by Friedman's test. All reported correlations were statistically significant by P < 0.05.
Results
Clamp
M_{bw} of the glucose clamp varied 7fold (Table 1, middle). Between 2/3 and 4/5 of this variability could be predicted by linear regression based on indices of glucose and insulin turnover obtained from the data collected during the IVGTT.
IVGTT
First key algorithm
Linear correlations between the IVGTT and the glucose clamp.
Y  X  Equation  Time period  r^{2}  25^{th}75^{th} percentiles of prediction error  

M_{bw} 
 Y = 172 + 1040 X  75 min  0.63  10%  +16% 
Y = 201 + 1179 X  40 min  0.63  8%  +20%  
Y = 219 + 1256 X  30 min  0.62  12%  +26%  
Same equation, but using total insulin AUC  Y = 220 + 1310 X  75 min  0.68  11%  +9%  
Y = 218 + 1287 X  40 min  0.63  8%  +12%  
Y = 248 + 1419 X  30 min  0.66  8%  +20%  
M_{bw} 
 Y = 19 +124 X  Baseline "Quicki"  0.41  14%  +11% 
M_{bw}  S_{I} of MINMOD 10^{5}  Y = 36 + 0.38 X  75 min  0.34  16%  +24% 
Consistently weaker correlations were obtained on correcting M_{bw} for the steady state plasma glucose and insulin concentrations (data not shown, r^{2}≈0.400.50).
This key algorithm has the same construction as "Quicki" which uses only the baseline values of plasma glucose and insulin. The original "Quicki" equation correlated with M_{bw} with a linearity of only r^{2} = 0.41 (Figure 2B) which was still slightly stronger than for other similar expressions, such as HOMAIR (r^{2} = 0.35) and the G/I ratio (r^{2} = 0.39) [2].
MINMOD and Tura's equation
Weaker correlations were also obtained when comparing M_{bw} with the insulin sensitivity as obtained by "minimal model analysis" (MINMOD) of the IVGTT data (r^{2} = 0.34, Figure 2C). Plots of X(t) obtained by MINMOD indicated that the insulin concentration at the effect site was highest at 18 min (1333) min.
The recently published equation by Tura et al. [11] correlated with M_{bw} with a linearity of r^{2} = 0.54 for the period 040 min. Logarithmtransformation of Tura's surrogate measure for insulin sensitivity increased r^{2} to 0.65.
Second key algorithm
Further linear correlations between the IVGTT and the glucose clamp.
Y  X  Equation  Time period  r^{2}  25^{th}75^{th} percentiles of prediction error  

M_{bw} 
 Y = 2.5 + 45.4 X  75 min  0.64  10%  +16% 
Y = 8.6 + 51.5 X  40 min  0.64  8%  +21%  
Y = 13.8 + 54.9 X  30 min  0.64  12%  +25%  
Same equation, but using total insulin AUC  Y = 2.8 + 53.4 X  75 min  0.68  10%  +9%  
Y = 6.1 + 54.0 X  40 min  0.64  8%  +13%  
Y = 14.5 + 60.0 X  30 min  0.67  8%  +20%  
M_{bw} 
 Y = 206  49.0 X + 340 CL/V_{ d }  75 min  0.70  11%  +16% 
Y = 224  56.4 X + 480 CL/V_{d}  40 min  0.74  10%  +20%  
Y = 223  57.9 X + 580 CL/V_{ d }  30 min  0.70  10%  +23%  
Same equation, but using total insulin AUC  Y = 265  63.6 X + 383 CL/V_{ d }  75 min  0.83  9%  +11%  
Y = 262  65.4 X + 488 CL/V_{ d }  40 min  0.82  10%  +11%  
Y = 260  67.1 X + 602 CL/V_{ d }  30 min  0.79  8%  +14%  
M_{bw} 
 Y = 99 + 54.0 X  75 min  0.63  10%  +16% 
Y = 9 + 51.5 X  1040 min  0.64  8%  +21%  
Y = 14 + 54.9 X  1030 min  0.64  12%  +26% 
A promising modification of this second key algorithm inserted the parameters of the glucose kinetics and the AUC for plasma insulin in a multiple regression equation, which yielded a maximum linearity of r^{2} = 0.83 for the relationship between the IVGTT and M_{bw} (Table 3, Figure 3B).
Slight strengthening of the linearity was always obtained by using AUC _{ ins } without correction for the baseline plasma insulin level (Tables 2 and 3).
Exploratory analyses
Replacing AUC _{ ins } by the sum of the plasma insulin concentrations for various periods of time did not greatly impair linearity or the prediction error (Table 3, Figure 3C).
The overall linear correlation between the AUC for Cpeptide and insulin was r^{2} = 0.66. However, replacing AUC_{ins} by AUC for Cpeptide in the equations proposed above greatly reduce their linearity with M_{bw} (r^{2} ≈ 0.20).
Discussion
IVGTT versus the glucose clamp
The present study searched for an approach to estimate insulin sensitivity that requires only minimum of resources. The results are presented as a number of regression equations that compare M_{bw} of the glucose clamp (control) with minor mathematical variations of two key algorithms based on data derived from a short IVGTT. Any of them may be used as substitutes for a glucose clamp in healthy volunteers, although some offer stronger linearity and a smaller prediction error than others.
The first of the key algorithms, shown on top of Table 2, is constructed in a way similar to the "Quicki" [10]. However, the linearity was much stronger when based on the IVGTT as compared to the baseline data used in the "Quicki" (Figure 2A, B).
Various modifications of the second key algorithm, presented in Table 3, were also tested. A promising change was to consider the sum of the slope of the glucose elimination curve, CL/V_{d}, and the insulin "pressure", AUC _{ins} , in a multiple regression equation. This approach could explain up to 83% of the interindividual variability in M_{bw} (Figure 3).
Reducing the sampling time from 75 min to 40 min, or even 30 min, had only small undue effects on our quality measures, i.e. the linearity and the prediction error.
Corrections for baseline concentrations
The relationship between plasma insulin and glucose is not a simple one. The doseresponse curve is hyperbolic (saturation kinetics) [2, 3] and the CL of glucose is related to the ^{10}log of the insulin level [3, 12].
The saturation kinetics makes it questionable to correct M_{bw} for the steady state insulin level in plasma to yield the M_{bw}/I ratio, although this is often done. The high concentration of insulin at the effect site at the end of a glucose clamp probably changes CL very little for a large increment in plasma insulin. Correcting M_{bw} for steady state plasma insulin also resulted in poorer correlations visàvis the IVGTT.
Likewise, one may question whether baseline insulin should be subtracted from AUC_{ins} when estimating M_{bw} from an IVGTT test. Although being a logical and commonly used correction, disregarding the baseline strengthened the correlations in the present study. Inhibition of the endogenous glucose production taking place early during the IVGTT is likely to make the insulin concentration below baseline govern the disposition of both the exogenous and the endogenous glucose later during the test. Differences in the mathematical correlations between the glucose clamp and the IVGTT were fairly small, however, and we therefore conclude that correcting for baseline insulin can be done, but is not essential.
Comparison with other methods
The precision by which our 12sample IVGTT could predict insulin sensitivity stands out favourably in comparison with other and more complex approaches, as presented in a review by Borai et al. [1].
A previous study of MINMOD based on a series of 25 blood samples showed a linearity to the glucose clamp that was quite similar to the r^{2} = 0.34 found here [13]. The new algorithms thus offered far better linearity than MINMOD in the present setting. MINMOD contains four unknown parameters that become gradually more difficult to estimate with good precision the fewer samples there are available. Moreover, MINMOD is not well suited for short sampling times. In contrast, the new algorithms included leastsquare regression estimation of only two parameters, CL and V_{d}, which makes them less sensitive for a reduction of sampling time and/or sampling intensity. With 12 samples, CL and V_{d} were estimated with the standard errors that averaged less than 10% (data not shown).
Tura et al. [11] recently compared the ratio of the glucose disappearance rate and AUC_{ins} with S_{I} and M_{bw} in a retrospective analysis of studies comprising both volunteers and diabetic and postoperative patients who had undergone a frequently sampled 50min IVGTT and a conventional 2hour glucose clamp. Good correlations between these indices of insulin sensitivity were claimed for all subgroups. The basic equation used is quite similar to the one we propose on the top of Table 3. However, they did not use the ^{10}log of AUC_{ins} and corrected this area for the group average S_{I} value. They also divided the expression by the sampling time, which we find questionable since plasma insulin but not K_{G} decreases with time. This fact must be handled by using a unique equation for each sampling time, as in Tables 2 and 3.
Limitations during surgery
The present study suggests two key algorithms, together with various modifications thereof, that may be used to estimate insulin sensitivity based on data derived from a short IVGTT performed in healthy volunteers. In a subsequent study, these algorithms will be validated in the pre and postoperative settings. Our interest in this topic stems from a wish to study insulin resistance during surgery. Virtually all nondiabetic patients develop transient type 2 diabetes as a part of the stress response to surgery [4, 5]. Too little research has been performed to investigate the reasons and consequences of this insulin resistance, which is probably due to the demanding and complex nature of both the glucose clamp and the IVGTT. In this setting, it is important that the blood sampling and the time and resources required for the test are kept low. Moreover, the test should impose only a slight burden on the body's physiology.
Conclusion
The ratio of the slope of the glucose elimination curve and the AUC for plasma insulin during a short IVGTT showed a strong linear correlation (r^{2} = 0.630.83) with the insulin sensitivity as obtained by the glucose clamp technique in healthy volunteers.
Abbreviations
 AUC:

area under the curve
 CL :

clearance
 IVGTT:

intravenous glucose tolerance test
 MINMOD:

minimal model analysis
 V _{d} :

volume of distribution
 T_{1/2}:

halflife.
Declarations
Acknowledgements and Funding
Tobias Gebäck, Chalmers School of Technology, Gothenburg, Sweden, programmed the MINMOD in the Matlab environment. Financial support was received from the Stockholm County Council (Grant number 20090433), Olle Engkvist Byggmästare Foundation, Karolinska institute, Swedish Society for Medical Research, and the Swedish Society of Medicine. The work was performed at The Metabolic Laboratory of the Endocrinology Department at Södersjukhuset, Stockholm, Sweden.
Authors’ Affiliations
References
 Borai A, Livingstone C, Ferns GAA: The biochemical assessment of insulin resistance. Ann Clin Biochem. 2007, 44: 324342. 10.1258/000456307780945778.View ArticlePubMedGoogle Scholar
 Muniyappa R, Lee S, Chen H, Quon MJ: Current approaches for assessing insulin sensitivity and resistance in vivo: advantages, limitations, and appropriate usage. Am J Physiol Endocrinol Metab. 2008, 294: E1526.View ArticlePubMedGoogle Scholar
 Ferrannini E, Mari A: How to measure insulin sensitivity. J Hypertens. 1998, 16: 895906. 10.1097/0000487219981607000001.View ArticlePubMedGoogle Scholar
 Brandi LS, Frediani M, Oleggini M, Mosca F, Cerri M, Boni C, Pecori N, Buzzigoli G, Ferrannini E: Insulin resistance after surgery: normalization by insulin treatment. Clin Sci. 1990, 79: 443450.View ArticlePubMedGoogle Scholar
 Ljungqvist O, Thorell A, Gutniak M, Häggmark T, Efendic S: Glucose infusion instead of preoperative fasting reduces postoperative insulin resistance. J Am Coll Surg. 1994, 178: 329336.PubMedGoogle Scholar
 Bergman RN, Ider YZ, Bowden CR, Cobelli C: Quantitative estimation of insulin sensitivity. Am J Physiol. 1979, 236: 667677.Google Scholar
 Nittala A, Ghosh S, Stefanovski D, Bergman R, Wang X: Dimensional analysis if MINMOD leads to definition of the disposition index of glucose regulation and improved simulation algorithm. Biomed Engineering OnLine. 2006, 5: 4457. 10.1186/1475925X544.View ArticleGoogle Scholar
 deFronzo RA, Tobin JD, Andres R: Glucose clamp technique: a method for quantifying insulin secretion and resistance. Am J Physiol. 1979, 273: E21423.Google Scholar
 Sjöstrand F, Hahn RG: Validation of volume kinetic analysis of glucose 2.5% solution given by intravenous infusion. Br J Anaesth. 2003, 90: 600607. 10.1093/bja/aeg102.View ArticlePubMedGoogle Scholar
 Katz A, Nambi SS, Mather K, Baron AD, Sullivan G, Quon MJ: Quantitative insulin sensitivity check index: a simple, accurate method for assessing insulin sensitivity in humans. J Endocrinol Metab. 2000, 85: 24022410. 10.1210/jc.85.7.2402.View ArticleGoogle Scholar
 Tura A, Sbrignadello S, Succurro E, Groop L, Sesti G, Pacini G: An empirical index of insulin sensitivity from short IVGTT: validation against the minimal model and glucose clamp indices in patients with different clinical characteristics. Diabetologia. 2010, 53: 144152. 10.1007/s0012500915479.View ArticlePubMedGoogle Scholar
 Berndtson D, Olsson J, Hahn RG: Hypovolaemia after glucoseinsulin infusions in volunteers. Clin Sci. 2008, 115: 371378. 10.1042/CS20080032.View ArticlePubMedGoogle Scholar
 Beard JC, Bergman RN, Ward WK, Porte D: The insulin sensitivity index in nondiabetic man: correlation between clampderived and IVGTTderived values. Diabetes. 1986, 35: 362369. 10.2337/diabetes.35.3.362.View ArticlePubMedGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.