jurnal ir

Dening Insulin Resistance FromHyperinsulinemic-Euglycemic ClampsCHARMAINE S. TAM, PHDWENTING XIE, MSWILLIAM D. JOHNSON, PHD

WILLIAM T. CEFALU, MDLEANNE M. REDMAN, PHDERIC RAVUSSIN, PHD

OBJECTIVEdThis study was designed to determine a cutoff point for identifying insulinresistance from hyperinsulinemic-euglycemic clamp studies performed at 120 mU/m2 z minin a white population and to generate equations from routinely measured clinic and bloodvariables for predicting clamp-derived glucose disposal rate (GDR), i.e., insulin sensitivity.

RESEARCH DESIGN AND METHODSdWe assembled data from hyperinsulinemic-euglycemic clamps (120 mU/m2 z min insulin dose) performed at the Pennington BiomedicalResearch Center between 2001 and 2011. Subjects were divided into subjects with diabetes(n = 51) and subjects without diabetes (n = 116) by self-report and/or fasting glucose $126mg/dL.

RESULTSdWe found that 75% of individuals with a GDR ,5.6 mg/kg fat-free mass (FFM) +17.7 zmin were truly insulin resistant. Cutoff values for GDRs normalized for body weight, bodysurface area, or FFM were 4.9 mg/kg z min, 212.2 mg/m2 z min, and 7.3 mg/kgFFM z min,respectively. Next, we used classication tree models to predict GDR from routinely measuredclinical and biochemical variables. We found that individual insulin resistance could be esti-mated with good sensitivity (89%) and specicity (67%) from the homeostasis model assessmentof insulin resistance (HOMA-IR) .5.9 or 2.8, HOMA-IR ,5.9 with HDL ,51 mg/dL.

CONCLUSIONSdWe developed a cutoff for dening insulin resistance from hyperinsulinemic-euglycemic clamps. Moreover, we now provide classication trees for predicting insulin re-sistance from routinely measured clinical and biochemical markers. These ndings extend theclamp from a research tool to providing a clinically meaningful message for participants inresearch studies, potentially providing greater opportunity for earlier recognition of insulinresistance.

Diabetes Care 35:16051610, 2012

There is substantial evidence that in-sulin resistance, typically dened asdecreased sensitivity or responsive-ness to the metabolic actions of insulin,is a precursor of the metabolic syndromeand type 2 diabetes. The gold standard forassessing insulin resistance in humans isthe hyperinsulinemic-euglycemic clamp.Developed by DeFronzo et al. in 1979 (1),this procedure assumes that at high dosesof insulin infusion (.80mU/m2 zmin), thehyperinsulinemic state is sufcient to com-pletely suppress hepatic glucose produc-tion and that there is no net change inblood glucose concentrations under steady-state conditions. Under such conditions,

the rate of glucose infused is equal to therate of whole-body glucose disposal (GDR)or metabolizable glucose (M) and reectsthe amount of exogenous glucose neces-sary to fully compensate for the hyperin-sulinemia. GDR is expressed as a functionof metabolic body size, such as bodyweight (kg), body surface area (m2; BSA),fat-free mass (kg; FFM), or metabolic size(kgFFM+17.7) (2).

Hyperinsulinemic-euglycemic clampsare used in cross-sectional studies and inprospective studies designed to test theeffect of interventions (weight loss, weightgain, or pharmacological treatment) on in-sulin sensitivity. The question of what is a

normalMvalue is largely unknown but isdependent on the dose of insulin infused.In 1985, Bergman et al. (3) examined M val-ues across 18 independent clamp studieswith an insulin infusion rate of 40 mU/m2 zmin. For nonobese normal glucose-tolerantsubjects, themeanM value was between 4.7and 8.7 mg glucose per kilogram of bodymass perminute. From these data, Bergmanet al. (3) proposed a conservative deni-tion for insulin resistance as an M value,4.7mg/kg zmin. To our knowledge, onlyone study has used a statistical approach todetermine a cutoff point for identifying in-sulin resistance from the clamp (4). In thisanalysis, results from 2,321 (2,138 sub-jects without diabetes) euglycemic clampprocedures (40 mU/m2 zmin) in an ethni-cally diverse population were assembled.It was found that insulin resistance wasbest predicted when subjects had a GDR(M value) ,28 mmol/kgFFM z min (4).Together, these previous studies give areasonable estimate of the distribution ofGDR for clamps performedwith an insulininfusion rate of 40 mU/m2 z min.

The question of what is a normal andclinically relevant M value derived fromclamps using other insulin infusion doses,such as 120 mU/m2 z min (57), is un-known. As such, it is difcult for clini-cians to explain the importance of theresults from these clamps to research par-ticipants. The 120 mU/m2 z min insulinwas chosen for this analysis for two majorreasons: 1) endogenous glucose produc-tion is likely to be fully suppressed, and 2) ashorter time is needed to reach steady-stateconditions with a higher infusion rate, thusmaking this insulin dose more cost-effectivewhile also reducing participant burden.Moreover, although the clamp technique isthe gold standard for directly assessinginsulin resistance in humans, it is time-consuming, labor-intensive, and overall ex-pensive. Therefore, the ability to predictresults of the clamp from other clinicaldata, which are both easier to obtain andless expensive to measure, is important.

In the current study, we assembleddata from clamp studies performed understandard operating procedures at thePennington Biomedical Research Center(PBRC; Baton Rouge, LA), between 2001and 2011, with an insulin infusion rate

c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c

From the Pennington Biomedical Research Center, Louisiana State University System, Baton Rouge, Louisiana.Corresponding author: Eric Ravussin, [email protected] 2 December 2011 and accepted 23 February 2012.DOI: 10.2337/dc11-2339This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10

.2337/dc11-2339/-/DC1. 2012 by the American Diabetes Association. Readers may use this article as long as the work is properly

cited, the use is educational and not for prot, and thework is not altered. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.

care.diabetesjournals.org DIABETES CARE, VOLUME 35, JULY 2012 1605

C a r d i o v a s c u l a r a n d M e t a b o l i c R i s kO R I G I N A L A R T I C L E

of 120 mU/m2 z min. Our aims were to1) determine a cutoff point for identifyinginsulin resistance for hyperinsulinemic-euglycemic clamp studies performed at120 mU/m2 z min in a white populationand 2) generate equations from commonlymeasured clinic and blood variables forpredicting insulin sensitivity (M values de-rived from the clamp).

RESEARCH DESIGN ANDMETHODSdThe Pennington CenterLongitudinal Study is an ongoing inves-tigation of the effects of obesity andlifestyle factors on the development ofchronic diseases such as type 2 diabetesand cardiovascular disease. The sample iscomprised of volunteers who have par-ticipated in nutrition, weight loss, andother metabolic intervention and observa-tional studies at the PBRC since 1992. Thecurrent cross-sectional analysis is limitedto adult participants who had a single-step120 mU/m2 z min hyperinsulinemic-euglycemic clamp before any interventionand a dual-energy X-ray absorptiometry(DXA) scan between 2001 and 2011 (n =263). Subjects with fasting plasma glucose.180 mg/dL or a diagnosis of diabetes.5years were excluded. Our cohort consistedof 167 whites, 94 African Americans, andtwo others. When euglycemic-hyperinsuli-nemic clamp studies were rst establishedat PBRC by one of the authors (E.R.), thecapability for measuring isotope tracers,and therefore hepatic glucose produc-tion, was unavailable. Therefore, clampswere performed at 120 mU/m2 z min asthis insulin dose was considered highenough to completely suppress hepaticglucose production.

Diabetes status was dened by self-report or fasting plasma glucose concentra-tions$126 mg/dL. Subjects were classiedas being diabetic if they 1) self-reportedyes to having diabetes (n = 62) or 2)self-reported no to diabetes but had afasting glucose $126 mg/dL (n = 4). Sub-jects were classied as being nondiabetic ifthey self-reported no and had a fastingglucose ,126 mg/dL (n = 197). Further-more, to verify diabetes status, we also ob-tained HbA1c data, which were available in104 of 263 subjects. HbA1c levels werehigher in subjects classied as having dia-betes (6.16 0.6%) comparedwith subjectsclassied as not having diabetes (5.4 60.4%). All procedures were approved bythe PBRC Institutional Review Board, andall participants provided written, informedconsent.

Anthropometry and bodycompositionHeight was measured with a wall-mountedstadiometer and metabolic weight witha digital scale. BMI was calculated asweight in kilograms divided by heightin meters squared. Whole-body percentbody fat was measured by DXA (HologicsQDR 4500A; Hologics, Bedford, MA),and fat mass (FM) and FFM were calcu-lated from the percent body fat and bodyweight. BSA was calculated using the DuBois equation (8).

Insulin sensitivityIn vivo insulin sensitivity was assessedwith a hyperinsulinemic-euglycemicclamp with a primed continuous infusionof insulin at 120 mU/m2 z min to achieveendogenous steady-state insulin concen-trations and to draw steady-state bloodsamples. An intravenous catheter wasplaced in an antecubital vein for infusionof insulin and glucose. A second catheterwas placed retrograde in a dorsal vein ofthe contralateral hand for blood draws.The hand was placed in a heating box at418C for arterialization of venous blood.A 20% glucose solution was infused at avariable rate necessary to maintain plasmaglucose concentrations between 90 and100 mg/dL. For clamps performed at120 mU/m2 z min, the duration was 2 h,and for clamps at 80 mU/m2 z min, theclamp duration was at least 3 h. Themean rate of exogenous glucose infusionduring the last 30 min was dened as theGDR (1). First, GDR was adjusted for glu-cose concentrations during this steady-state interval (GDR 3 [average groupsteady-state glucose/individual steady-state glucose]) (3). Next, to adjust for met-abolic size, GDR was normalized forweight, BSA, FFM, or FFM+17.7 (2).

Blood analysisPlasma glucose was analyzed with a YellowSprings Instruments 2300 STAT GlucoseAnalyzer (Yellow Springs Instruments Inc.,Yellow Springs, OH). Plasma insulin wasmeasured by chemiluminescent immuno-assays on an Immulite 2000 Analyzer (Di-agnostic Products), and lipid (FFA, totalcholesterol, and HDL) concentrations weremeasured with a Beckman Coulter Syn-chron DXC 600 Pro. LDL was calculatedwith the Friedewald calculation.

Statistical methodsDescriptive data are presented as mean6SD. Statistical analysis was performedwith SAS version 9.2 (SAS Institute,

Cary, NC). The x2 test was used to testsignicance of group differences in sex.Independent sample t tests and Mann-Whitney U tests were used to test groupdifferences for continuous variables. GDRdetermined from the hyperinsulinemic-euglycemic clamp and adjusted forweight, BSA, FFM, or FFM+17.7 wasused as a biomarker for separating subjectswho were insulin resistant (had type 2 di-abetes mellitus) from those who were in-sulin sensitive (did not have type 2diabetes). We dened the optimal cutoffpoint for discriminating insulin-resistantsubjects from insulin-sensitive subjectsas the GDR that provided both high sen-sitivity (ability to detect insulin-resistantsubjects) and simultaneously providedhigh specicity (ability to detect insulin-sensitive subjects). Thus, sensitivity is theproportion of true positives (individualsknown to be insulin resistant who wereclassied as insulin resistant based on theirGDR), whereas 1 specicity is the pro-portion of false positives (individualsknown to be insulin sensitive who wereclassied as insulin resistant based on theirGDR). We used a receiver-operating char-acteristic (ROC) curve, a plot of the sensi-tivity versus 1 specicity, to display thetrade-off between true positives and falsepositives across incremental choices ofdiscriminating cutoff points for glucosedisposal rate (9,10). The cutoff point cor-responding to the optimal balance oftrue versus false positives indicates thebest threshold for discriminating insulin-resistant individuals from insulin-sensitiveindividuals.Principles of classication trees. Clas-sication trees were used to determine aspecic set of rules (algorithm) for classi-fying subjects as insulin resistant or insulinsensitive from GDR (11). This methodmakes no assumptions about the underly-ing distribution of data and takes into ac-count interactions among covariates andoutcome variables. An advantage of theclassication tree method is that it allowsobservations with missing values in a data-set, as opposed to traditional logistic regres-sion analysis, which removes observationswith missing predictor values. Tree-basedmethods recursively partition the predictorvariables into a set of decision points orclassication nodes. A binary decisiontree is fully grown when classication rulesare completely diagrammed. Withoutpruning, a large and complex tree canbe built at the cost of overtting the data(12). Thus, pruning the tree is imple-mented so that we can select a tree size

1606 DIABETES CARE, VOLUME 35, JULY 2012 care.diabetesjournals.org

Dening insulin resistance from clamps

that minimizes the cross-validated errorand avoids overtting.

Decision rules for classifying subjectscan be obtained by assigning each termi-nal node a class label as either 1 (pres-ence of insulin resistance) or 0 (absenceof insulin resistance). The choice of a classlabel is based on a dened cutoff pointand also the proportion of true event oc-currences at that terminal node. For ex-ample, if 0.25 or above is chosen as acutoff point, then only terminal nodeswith true event occurrence proportionequal to or larger than 0.25 are classiedas insulin resistant and the remaining areclassied as 0 or insulin sensitive. Theproportion displayed at each terminalnode can be considered as a risk scorefor insulin resistance.Single Classication Tree Method. Inorder to select the best-performing mod-els for exploring the relationship betweeninsulin sensitivity and clinical variables,subjects were randomly split into a train-ing subset for building the model (n =125) and a testing subset for evaluatingthe performance of the model (n = 42) at aratio of 3:1.

The Single Classication Tree Methodwas chosen as the statistical exploratorymethod because it outperformed otherapproaches such as logistic regression andboosting regression trees in terms of sensi-tivity, specicity, and area under the ROCcurve, denoted aROC (10), on the testingdataset. Thus, aROC is used as ameasure ofaverage prediction performance of differentclassication algorithms. Because aROCrepresents area of a portion under a unitsquare, its value ranges between 0 and 1.Random guessing results in an aROC ap-proximately equal to 0.5. A realistic classi-cation method should have an aROC.0.5, and a classication rule with aROC.0.70 is generally considered adequate;aROC very close to 1 indicates exception-ally good performance. Our goal was toexamine the selected decision trees for pre-dicting insulin resistance to accommodatedifferent clinical/research settings wherevalues for all variables would not necessar-ily be available.

The tree analysis was performed withdata from 167 whites. The main outcomevariable used in this analysis was GDRadjusted for FFM+17.7 (GDRadj) becausewe believe that this adjustment is the bestavailable method accounting for meta-bolic size (2). Covariates included sex,age, weight, height, BMI, BSA, FFM,self-reported family history of diabetes,fasting glucose, fasting insulin, cholesterol,

HDL, LDL, triglycerides, FFA, and homeo-stasismodel assessment of insulin resistance(HOMA-IR) [fasting insulin (mU/mL) 3fasting glucose (mg/dL)/405]. Three classi-cation tree models were developed to an-swer different questions. Model 1 wasconstructed with all available input vari-ables, mimicking a clinical situation wherebody composition and blood markers wereall included. Model 2 uses only fasting glu-cose and insulin concentrations, BMI, age,and sex as input variables. Model 3 is amin-imalist model constructed using age, sex,and BMI.

RESULTS

Subject characteristicsOur white cohort comprised 51 subjectswith diabetes and 116 subjects withoutdiabetes (Table 1). Subjects with diabe-tes were signicantly older (age 57.2 68.6 vs. 44.2 6 13.2 years, P , 0.001)and had higher LDL (P = 0.05) and glu-cose levels (by study design; P , 0.001)than subjects without diabetes. Therewere no signicant sex differences be-tween subjects with and without diabe-tes (P = 0.10).

Distribution and optimal cutoff fordetermination of insulin resistanceThe distribution of GDR, whether it wasadjusted for weight, BSA, FFM, or FFM+17.7was bimodal (summarized in Table 2).As expected, GDR was lower in subjectswith diabetes compared with subjectswithout diabetes, regardless of how

GDR was adjusted for metabolic size. ForGDRadj, the mean was 7.4 6 2.7 mg/kgFFM+17.7 z min in subjects without di-abetes and 4.6 6 1.7 mg/kgFFM+17.7 zmin in subjects with diabetes. A cutoff of5.6 mg/kgFFM+17.7 z min provided themaximum theoretical sensitivity (75%)and specicity (71%) with an aROC of80% for dening insulin resistance. Inother words, the optimal cutoff for iden-tifying insulin-resistant individuals fromthe 120 mU/m2 z min clamp was 5.6mg/kgFFM+17.7 zmin. Using this criterionin the whole group, 44.3% of individualsfell below this cut point and could be con-sidered insulin resistant. The same cutoff(5.6 mg/kgFFM+17.7 z min) was observedwhen this analysis was performed in thewhole cohort (n = 267) (data not shown).When GDR was adjusted for body weight,BSA, or FFM, the cutoff value for insulinresistance was 4.9 mg/kg weight zmin, 212mg/m2 z min, and 7.3 mg/kgFFM z min,respectively (Table 2).

As a subanalysis, we also compileddata from86hyperinsulinemic-euglycemicclamps performed at 80 mU/m2 z min.Similar to the 120-mU insulin dose, thedistribution of GDR was bimodal withlower GDR in subjects with diabetes (n =11) compared with subjects without dia-betes (n = 75). The average data and cutoffpoints for GDR expressed by weight, BSA,FFM, or FFM+17.7, and the subject char-acteristics for this cohort are provided inSupplementary Tables 1 and 2. In sub-jects without diabetes, the mean GDRadjwas 8.2 6 3.0 mg/kgFFM+17.7 z min

Table 1dSubject characteristics for whites undergoing hyperinsulinemic-euglycemicclamp studies at 120 mU/m2 min

Subjects without diabetes(n = 116)

Subjects with diabetes(n = 51) P

Sex 66 females, 50 males 22 females, 29 males 0.10a

Age, years 44.2 6 13.2 (1867) 57.2 6 8.6 (3570) ,0.001b

Weight, kg 93.3 6 20.6 (52.3140.1) 92.1 6 12.7 (65.212.1.0) 0.66c

BMI, kg/m2 32.7 6 6.9 (19.746.7) 31.9 6 4.4 (24.040.7) 0.41b

% Fat 35.6 6 9.6 (10.453.4) 34.0 6 8.0 (18.651.2) 0.16b

HDL, mg/dL 50.5 6 12.7 (26.994.8) 48.6 6 12.2 (24.482.6) 0.54b

LDL, mg/dL 125.2 6 38.8 (54.0324.6) 111.9 6 37.9 (43.0248.8) 0.05c

Triglycerides,mg/dL 155.5 6 93.5 (42.0613.0) 180.6 6 131.2 (59.0698.0) 0.38b

Cholesterol,mg/dL 206.3 6 41.6 (142.0411.0) 196.0 6 46.9 (103356) 0.16c

FFA, mmol/L 0.6 6 0.2 (0.061.13) 0.6 6 0.2 (0.31.0) 0.43c

Fasting glucose,mg/dL 96.2 6 7.9 (82.0123.0) 119.1 6 19.3 (75.0187.0) ,0.001b

Data are presented as mean6 SD (range). Diabetes status was dened by self-report or fasting glucose values$126 mg/dL. ax2 test. bMann-Whitney U test. cIndependent samples t test.


Tam and Associates

compared with 3.3 6 0.7 mg/kgFFM+17.7 z min in subjects with diabetes,and the cutoff for dening insulin resis-tance was 4.1 mg/kgFFM+17.7 z minwith a maximal sensitivity and specicityof 83 and 96%, respectively (aROC=0.98).When GDR was adjusted for body weight,BSA, or FFM, the cutoff value for insulinresistance was 4 mg/kgweight z min, 192mg/m2 z min and 5 mg/kgFFM z min, re-spectively (Supplementary Table 2). Treemodel analyses were not performed at the80-mU insulin dose because of the rela-tively small sample size and unbalancedproportion of subjects with and withoutdiabetes.

Tree models for predicting insulinresistance using all predictorsAs indicated in RESEARCHDESIGNANDMETHODS,tree models and decision rules were de-veloped to predict GDRadj based on a ran-domly selected training set (n = 125)followed by evaluating the performanceof the models in the remaining testing set(n = 42).Model 1: all predictors. In the treemodel based on all predictors (Model 1),the following predictors were statisticallysignicant: HOMA-IR, HDL, and fastingglucose resulting in an aROC of 87%.Figure 1 depicts the classication treemodel using these predictors. The numberof insulin-resistant individuals and thenumber of insulin-sensitive individualsare shown in Fig. 1. Using an arbitrarycutoff of 0.25 as the risk score of havinginsulin resistance, nodes $25% (propor-tion of insulin-resistant individuals) pre-dict insulin resistance and nodes ,25%predict insulin sensitivity. Therefore, inFig. 1, the associated decision rules forpredicting an individual as being insulinresistant is either having the following: 1) aHOMA-IR .5.9 or 2) HOMA-IR = 2.85.9 and HDL ,51 mg/dL. Use of otherchoices for the predictive cutoff value (e.g.,50%) yielded similar sensitivity (89%) andspecicity (67%) results (data not shown).

Model 2: anthropometry, fasting glu-cose, and insulin measurements. Thenext tree model was performed using onlybody composition, fasting glucose, insu-lin, and age and sex as predictor variables.Only fasting insulin and glucose weresignicant predictors in building the treemodel with an aROC of 86% in the test-ing subset (Fig. 2). Similar to the abovetreemodel using all predictors, we assignedan arbitrary cutoff of 0.25 as the risk scoreof being classied as insulin resistant.Therefore, in Fig. 2, the associated decision

rule for predicting an individual to be in-sulin resistant is a fasting insulin concen-tration .10.6 mU/mL. This decision rulehas an estimated sensitivity of 100% andspecicity of 54%.Model 3: age, sex, and BMI. The lasttree model we tested for predicting in-sulin sensitivity from the clamp only usedage, sex, and BMI as predictor variables.Again based on an arbitrary cutoff of0.25% for detecting insulin resistance,the only signicant predictor variablewas BMI with an aROC of 61% and a

Table 2dGDR values at 120 mU/m2 min for subjects with and without diabetes adjusted for metabolic size, including bodyweight, BSA, FFM, and FFM+17.7 kg

Subjects withoutdiabetes (n = 116)

Subjects withdiabetes (n = 51)

Cutoff value for insulinresistance Sensitivity (%) Specicity (%) aROC

GDR/kg body weight 6.4 6 2.8 4.0 6 1.8 4.9 67 64 0.76GDR/m2 BSA 275.6 6 98.4 177.3 6 69.6 212.2 70 69 0.79GDR/kg FFM 9.8 6 3.8 6.0 6 2.3 7.3 71 72 0.80GDR/kg FFM+17.7(GDRadj) 7.4 6 2.7 4.6 6 1.7 5.6 75 71 0.80

Data are presented as mean 6 SD, unless otherwise indicated.

Figure 1dTree model for insulin resistance determined using all available body composition andblood measures. HOMA-IR and HDL were the only signicant determinants in this model. Themodel is built on a randomly selected training cohort of 125 subjects and tested in 42 subjects.An arbitrary risk score of 0.25 is calculated. Therefore, if a terminal node has a .25% pro-portion, those subjects are more likely to be insulin resistant (dashed lines). The decision nodesfor being insulin resistant are as follows: 1) HOMA-IR.5.9 and 2) 2.8, HOMA-IR,5.9 andHDL ,51 mg/dL.



BMI .24.7 kg/m2 predicting insulin re-sistance. As expected, as the number ofpredictor variables decreases, the aROCand sensitivity and specicity resultsworsen.

The same tree-based model analysiswas also performed in a larger, moreethnically diverse cohort of 263 individuals(167 whites, 94 African Americans, oneHispanic, and one other). As expected,with a greater number of subjects, higheraROCs, and sensitivity and specicity re-sults were obtained in the larger cohort(data not shown).

CONCLUSIONSdThehyperinsulinemic-euglycemic clamp technique is the goldstandard for assessing insulin sensitivityin humans. This method is widely used inresearch studies to examine the effects ofan intervention, such as low-calorie dietor pharmacological therapy. Unfortunately,few data exist on what is considered anormal glucose infusion rate, i.e., normalinsulin sensitivity. Moreover, the abilityto compare results from clamp studies isclouded by the fact that GDR results areexpressed as a function of body weight,BSA, or FFM. The goal of the current studywas to provide cutoff values for deninginsulin resistance and insulin sensitivityin clamps performed at an insulin dose of

120 mU/m2 z min and to provide decisiontrees for predicting insulin resistance fromroutinely measured clinical and biochemi-cal parameters. Similar to a previous study(4), we found that the distribution ofwhole-body glucose disposal rate is bi-modal. This bimodality allowed us to use astatistical approach to determine a cutoff fordening insulin resistance or insulin sensi-tivity. We found that a GDR or M value of5.6 mg/kgFFM+17.7 z min in whites givesan almost 80% probability of predictinginsulin resistance. In other words, indi-viduals with a GDR ,5.6 mg/kgFFM+17.7 z min have an 80% chance of beinginsulin resistant. The same cutoff was foundin a larger, more ethnically diverse cohort(data not shown).

In a smaller sample of clamps per-formed at 80 mU/m2 zmin, a cutoff of 5.3mg/kgFFM+17.7 z min (98% predictionprobability) was determined for deninginsulin resistance. That is, subjects with aGDR ,5.3 mg/kgFFM+17.7 z min have a98% probability of having insulin resis-tance. To expand the applicability of thesendings to the investigators who do notmeasure body composition, we deter-mined cut points for insulin resistancefor GDR expressed by weight, BSA, orFFM. These data allow researchers per-forming hyperinsulinemic-euglycemic

clamps at the 80 and 120 mU/m2 z min insu-lin infusion rates to gain valuable clinicalinformation on a subjects insulin sensitiv-ity and translate ndings from a primarilyresearch setting to a clinically relevant mes-sage. A summary of our results at 80 and120 mU/m2 z min and results from otherstudies at 40mU/m2 zmin (3,4) is providedin Supplementary Table 3.

Given the demanding nature, expense,and time involved in performing hyper-insulinemic-euglycemic clamps, we aimedto develop models for predicting insulinresistance by the clamp from routinelymeasured anthropometric, body composi-tion, and biochemical markers. Our rstclassication model (Model 1) included allavailable variables (listed in RESEARCHDESIGNAND METHODS) and found that subjectswith a 1) HOMA-IR .5.9 or 2) HOMA-IRbetween 2.8 and 5.9 and HDL,51 mg/dLpredicted insulin resistance with goodspecicity and sensitivity. The ndingthat HOMA-IRwas the strongest predictorof GDRadj is not surprising given the strongcorrelation between HOMA-IR and insulinsensitivity derived from the clamps demon-strated previously (13). In addition, ourndings are similar to those from the studyby Stern et al. (4) in which they generatedclassication trees from clamps performedat 40 mU/m2 z min. Stern et al. (4) foundHOMA-IR, BMI, waist circumference, andLDL as signicant predictors of insulin re-sistance, with similar sensitivity and speci-city results (our study: aROC = 0.87; Sternet al.: aROC = 0.90). Model 2 used onlyanthropometry, fasting glucose, insulin,and age and sex. Similar to Model 1, wefound that only fasting insulin was a sig-nicant predictor of GDRadj, with a fastinginsulin .10.6 mU/mL dening insulin re-sistance. Model 3 used only age, sex, andBMI to predict GDRadj. This model gavepoor specicity and sensitivity results thathighlighted the well-known fact that insu-lin resistance is a heterogeneous disorderthat is not only dependent on weight, sex,and age (14). Coupled with previous stud-ies dening insulin resistance from clampsperformed at 40 mU/m2 z min (3,4), ourresults at 80 and 120 mU/m2 z min insulindoses contribute substantially to the litera-ture determining insulin resistance fromclamps (Supplementary Table 3).

To account for the metabolic massthat uses insulin during a clamp proce-dure, GDR derived from the clamp mustbe adjusted for metabolic size. The ma-jority of studies in the literature tend touse total body weight or kilograms of FFMwith a smaller number of studies using

Figure 2dTree model for insulin resistance using only fasting glucose, insulin, age, sex, andBMI. Only fasting insulin was a signicant determinant in the model. The model is built ona randomly selected training cohort of 125 subjects and tested on 42 subjects. An arbitrary riskscore of 0.25 is calculated. Therefore, if a terminal node has a .25% proportion, those subjectsare more likely to be insulin resistant (dashed lines). The decision node for being insulin resistantis having a fasting insulin .10.6 mU/mL.


Tam and Associates

BSA. However, total body weight is nota completely appropriate method forcomparing individuals, because obesesubjects have a greater proportion of low-metabolizing mass (adipose tissue) andfemales have a greater percentage of fatthan men. BSA also poses a sex differenceproblem (2). Fromstudies ofmetabolic ratein the Pima Indians, Lillioja and Bogardus(2) showed that metabolic rate is directlyproportional to FFM+17.7 kg, suggestingthat this measure can be equated to meta-bolic size. On the basis of this nding, wehave chosen this normalization methodin our classication analysis. However,we have also considered that not all re-searchers may have access to measuringbody composition (DXA) and have also re-ported on these insulin resistance cutoffsnormalized by weight and BSA. Finally, itneeds to be noted that dividing GDRsduring a clamp by resting metabolic ratewould really represent the best way tocompare values from people with differ-ent body sizes.

Limitations of this study may be thesmall sample size of clamp studies at 120mU/m2 zmin, and particularly at 80mU/m2 zmin. However, we were able to achievegood statistical sensitivity and specic-ity results (.80% prediction probability)in this cohort. We also demonstrated thesame cutoff points in a larger, more ethni-cally diverse population. Another poten-tial limitation may be that we did not useradiolabeled glucose tracers to quantifyendogenous glucose production. How-ever, the high insulin infusion doses re-ported in this study generally suppressmost, if not all, the baseline splanchnicglucose output (15).We also acknowledgethat there is considerable overlap in insu-lin sensitivity (M) between subjects withand without diabetes (2). An alternativeapproach would have been to divide ourcohort into subjects with normal glucosetolerance, impaired glucose tolerance, ortype 2 diabetes; however, this would havefurther diminished our sample size.

In conclusion, our study providesnovel data for dening insulin resistance

from hyperinsulinemic-euglycemic clampsperformed at insulin doses of 120 and 80mU/m2 zmin. Moreover, we have providedclassication trees for predicting insulinresistance from routinely measured bio-chemical markers. Together, our ndingsextend the hyperinsulinemic-euglycemicclamp from what is largely considered aresearch tool to providing clinically mean-ingfulmessages for patients, thus providinggreater likelihood of earlier detection of in-sulin resistance.

AcknowledgmentsdThis study was sup-ported by funding from the National In-stitutes of Health (grants P30DK072476 andR01DK060412 and R00HD060762 andU01DK094418) to L.M.R.No potential conicts of interest relevant to

this article were reported.C.S.T. wrote the manuscript, contributed to

discussions about the results, and reviewed andedited the nalmanuscript.W.X. performed thestatistical analysis, contributed to discussionsabout the results, and reviewed and edited thenal manuscript. W.D.J. performed the statis-tical analysis, contributed to discussions aboutthe results, and reviewed and edited the nalmanuscript. W.T.C. contributed to discussionsabout the results and reviewed and edited thenal manuscript. L.M.R. contributed to dis-cussions about the results and reviewed andedited the nal manuscript. E.R. contributed todiscussions about the results and reviewed andedited thenalmanuscript. E.R. is the guarantorof this work and, as such, had full access to allthe data in the study and takes responsibility forthe integrity of the data and the accuracy of thedata analysis.We thank Dr. Frank Greenway for re-

peatedly asking us, what is a good (or bad)insulin sensitivity?

References1. DeFronzo RA, Tobin JD, Andres R. Glucose

clamp technique: a method for quantify-ing insulin secretion and resistance. AmJ Physiol 1979;237:E214E223

2. Lillioja S, Bogardus C. Obesity and insulinresistance: lessons learned from the PimaIndians. Diabetes Metab Rev 1988;4:517540

3. Bergman RN, Finegood DT, Ader M. As-sessment of insulin sensitivity in vivo.Endocr Rev 1985;6:4586

4. Stern SE, Williams K, Ferrannini E,DeFronzo RA, Bogardus C, Stern MP.Identication of individuals with insulinresistance using routine clinical measure-ments. Diabetes 2005;54:333339

5. Stull AJ, Cash KC, JohnsonWD, ChampagneCM, Cefalu WT. Bioactives in blueberriesimprove insulin sensitivity in obese, insulin-resistant men and women. J Nutr 2010;140:17641768

6. Plaisance EP, Greenway FL, BoudreauA, et al. Dietary methionine restriction in-creases fat oxidation in obese adults withmetabolic syndrome. J Clin EndocrinolMetab 2011;96:E836E840

7. Cefalu WT, Rood J, Pinsonat P, et al. Char-acterization of the metabolic and physiologicresponse to chromium supplementationin subjects with type 2 diabetes mellitus.Metabolism 2010;59:755762

8. Du Bois D, Du Bois EF. A formula to esti-mate the approximate surface area if heightandweight be known. 1916.Nutrition 1989;5:303311; discussion 312313

9. Bewick V, Cheek L, Ball J. Statistics review13: receiver operating characteristic curves.Crit Care 2004;8:508512

10. Bradley AP. The use of the area under theROCcurve in the evalution ofmachine learn-ing algorithms. Pattern Recognit 1997;30:11451159

11. Hastie T, Tibshirani R, Friedman J. Tree-basedmodels. In The Elements of StatisticalLearning, Second Edition. Springer, NewYork, 2008, p. 305316

12. Breiman L, Friedman JH, Olshen RA,Stone CJ. Classication and Regression Trees.Belmont, CA, Wadsworth InternationalGroup, 1984

13. Matthews DR, Hosker JP, Rudenski AS,Naylor BA, Treacher DF, Turner RC. Ho-meostasis model assessment: insulin re-sistance and beta-cell function fromfasting plasma glucose and insulin con-centrations in man. Diabetologia 1985;28:412419

14. Olefsky JM. Lilly lecture 1980. Insulinresistance and insulin action. An in vitroand in vivo perspective. Diabetes 1981;30:148162

15. DeFronzo RA: Pathogenesis of type 2 di-abetes mellitus. Med Clin North Am 88:787835, ix, 2004



jurnal ir

Documents