Insulin Sensitivity, Insulinemia, and Coronary Artery Disease the Insulin Resistance Atherosclerosis Study

Metabolic Syndrome/Insulin Resistance/Pre-Diabetes ORIGINAL ARTICLE

Insulin Sensitivity, Insulinemia, and Coronary Artery Disease The Insulin Resistance Atherosclerosis Study

1 5 MARIAN REWERS, MD, PHD JOE V. SELBY, MD, MPH tery disease (CAD). On the other hand, 2 6 DANIEL ZACCARO, MS RICHARD BERGMAN, PHD hyperinsulinemia (a marker of low in- 2 7 RALPH D’AGOSTINO,JR., PHD PETER SAVAGE, MD sulin sensitivity) has been related to CAD 3 STEVEN HAFFNER, MD, MPH FOR THE INSULIN RESISTANCE in numerous prospective (6–12) and 4 MOHAMMED F. SAAD, MD ATHEROSCLEROSIS STUDY INVESTIGATORS cross-sectional studies (13). Insulinemia is generally inversely related to insulin sensitivity, but the relationship is not lin- ear (14), and it is usually absent in dia- OBJECTIVE — The aim of this study was to evaluate whether low insulin sensitivity (Si) measured using a modified frequently sampled intravenous glucose tolerance test with minimal betic individuals (15,16) who account for model analysis is associated with coronary artery disease (CAD) independent of other cardio- a significant proportion of people with vascular risk factors. low insulin sensitivity. The Insulin Resis- tance Atherosclerosis Study (IRAS) (17) RESEARCH DESIGN AND METHODS — We studied 1,482 women and men, age and others (18,19) have previously 40–69 years old, African American (28%), Hispanic (34%), or non-Hispanic white (38%), with shown that low insulin sensitivity is asso- normal (45%), impaired (23%), or diabetic (32%) glucose tolerance. CAD defined as confirmed ciated with atherosclerosis, defined by the past myocardial infarction, coronary artery bypass graft, coronary angioplasty, or presence of a major Q-wave was found in 91 participants. intima-media thickening of the carotid arteries. In this study, we test the hypothesis RESULTS — The odds ratio (OR) for CAD was greatest among individuals in the two lowest that low insulin sensitivity is also cross-

quintiles of Si (2.4, 95% CI 1.0–5.6 and 4.7, 2.1–10.7) compared with the highest Si quintile. sectionally associated with clinical CAD, After adjusting for demographic and cardiovascular risk factors, a decrement from the 75th to independent of insulin levels and other ϭ 25th percentile in Si was associated with a 56% increase in CAD (P 0.028). Similar increments cardiovascular risk factors. in fasting or 2-h insulin levels were associated with, respectively, only 15 (NS) and 3% (NS)

increases in CAD. The association between Si and CAD was partially mediated by insulin, HDL cholesterol and triglyceride levels, hypertension, diabetes, and obesity, but not LDL cholesterol RESEARCH DESIGN AND or cigarette smoking. METHODS — The design of IRAS, a four-center epidemiological study explor- CONCLUSIONS — Low Si is associated with CAD independently of and stronger than ing relationships among insulin sensitiv- plasma insulin levels. Part of the association is accounted for by dyslipidemia, hypertension, ity, insulin levels, cardiovascular risk diabetes, and obesity. factors, and cardiovascular disease across Diabetes Care 27:781–787, 2004 a broad range of glucose tolerance, has been previously published (20). Briefly, IRAS evaluated 1,624 women and men ow insulin sensitivity underlies the measurement of insulin sensitivity is tech- aged 40–69 years, representing normal metabolic syndrome that includes nically difficult, and only a few relatively (44%), impaired (23%), and diabetic glu- L central obesity, dyslipidemia, hyper- small studies (3–5) have demonstrated a cose tolerance (33%). Of the 479 diabetic glycemia, hypertension, impaired fibrino- strong association between insulin sensi- patients included in this report, 294 were lysis, and atherosclerosis (1,2). However, tivity measured directly and coronary ar- previously diagnosed (average duration ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● of diabetes 6.9 Ϯ 6.4 years). Of those, From the 1Barbara Davis Center, University of Colorado HSC, Denver, Colorado; the 2Department of Public 73% were taking oral hypoglycemic 3 Health Sciences, Wake Forest School of Medicine, Winston-Salem, North Carolina; the Department of agents, whereas the remaining were Medicine, University of Texas HS, San Antonio, Texas; the 4Department of Medicine, University of California Los Angeles, Los Angeles, California; the 5Department of Physiology and Biophysics, University of Southern treated with diet alone. Individuals with California, Los Angeles, California; the 6Kaiser Research Center, Northern California, Division of Research, impaired glucose tolerance (IGT) and Oakland, California; and the 7Division of Epidemiology and Clinical Applications, National Heart, Lung, and type 2 diabetes were over-sampled to Blood Institute, Bethesda, Maryland. achieve sufficient statistical power in Address correspondence and reprint requests to Marian Rewers, MD, PhD, Barbara Davis Center, Uni- these subgroups. Nondiabetic IRAS par- versity of Colorado Health Sciences Center, B-140, 4200 E 9th Ave., Denver, CO 80262. E-mail: marian. [email protected]. ticipants had, however, fasting blood glu- Received for publication 1 September 2003 and accepted in revised form 1 December 2003. cose levels similar to those in nondiabetic Abbreviations: CAD, coronary artery disease; ECG, electrocardiogram; FSIGT, frequently sampled in- individuals of the same ethnic group in travenous glucose tolerance test; HOMA, homeostasis model assessment; IGT, impaired glucose tolerance; the general population (20). IRAS clinics IRAS, Insulin Resistance Atherosclerosis Study. A table elsewhere in this issue shows conventional and Syste`me International (SI) units and conversion in Oakland and Los Angeles, California, factors for many substances. studied non-Hispanic whites and African © 2004 by the American Diabetes Association. Americans recruited from Kaiser Perma-

DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 781 The Insulin Resistance Atherosclerosis Study nente health maintenance organizations. firmed by review of medical records or a viously described (17), but it was The centers in San Antonio, Texas and major Q-wave on IRAS examination elec- statistically not significant (P Ͼ 0.05). San Luis Valley, Colorado recruited non- trocardiogram (ECG). The IRAS Events Hispanic whites and Hispanics from on- Committee (M.R., S.H., and J.S.) re- RESULTS — This report includes 91% going population-based studies (21,22). viewed using standard criteria (28) all (1,482 of 1,624) of the study participants Race and ethnicity were assessed by self- events reported to occur before the IRAS who completed the FSIGT. Univariate report using the U.S. census definitions; examination. Myocardial infarction was comparison of the characteristics of the African-Americans comprised 29%, His- confirmed in 39 (72%) of 54, coronary 91 case subjects and 1,391 control sub- panics 34%, and non-Hispanic whites artery bypass graft in 19 of 21, and per- jects studied (Table 1) confirmed known 37% of the study participants. Exclusion cutaneous transluminal coronary angio- associations between CAD and type 2 di- criteria included insulin treatment in the plasty in 7 of 9 of case subjects. Standard, abetes, male sex, older age, central obesity past 5 years, fasting glucose Ն16.7 resting 12-lead ECG was performed using (higher waist-to-hip ratio), dyslipidemia mmol/l [300 mg/dl], unstable angina, de- the MAC/PC electrocardiograph (Mar- (low HDL cholesterol and high triglycer- compensated congestive heart failure, or quette Electronics, Milwaukee, WI). ECG ides), hypertension, and cigarette smok- serious illness within the past month. All tracings were read centrally using NOVA- ing. Case subjects had significantly lower study protocols were approved by institu- CODE ECG software and the Minnesota Si levels than control subjects. Fasting in- tional review boards, and informed con- Code (29) and revealed a major Q-wave sulin levels were only on the borderline of sent was obtained from all participants. (Minnesota code 1.1–1.2, except for being higher among case subjects than 1.28) in 59 of the participants. Of the control subjects. There was no difference Measurement of glucose tolerance, 1,482 IRAS participants who completed in the levels of 2-h insulin between case insulin, and insulin sensitivity FSIGT, 91 (47 nondiabetic and 44 dia- subjects and control subjects. An oral glucose tolerance test with glu- betic participants) had at least one of To explore the linearity of the rela- cose tolerance classification according to these events and were classified as case tionship between Si and the CAD, the ORs the WHO criteria (23) and a frequently subjects. of CAD were estimated by quintiles of the sampled intravenous glucose tolerance Si distribution, adjusting for age, sex, eth- test (FSIGT) were performed on two sep- Other measurements nicity, and clinic (Fig. 1). Adjusted CAD arate days 2–28 days apart. Participants Resting systolic and diastolic blood pres- ORs for quintiles of fasting and 2-h insu- were asked to refrain from heavy exercise sure were measured three times, and the lin levels were included for comparison. and alcohol consumption for 24 h and second and third measurements were av- The quintile of highest Si or lowest fasting fast for 12 h before each visit and abstain eraged. Hypertension was defined as sys- insulin or 2-h insulin levels served as the from smoking the morning of examina- tolic blood pressure Ն140 or diastolic reference. The ORs for CAD were greatest Ն tion. Plasma glucose was measured with blood pressure 90 mmHg or if they among individuals in the second lowest Si the glucose oxidase method on an auto- were currently taking antihypertensive quintile (OR ϭ 4.7, 95% CI 2.1–10.7), mated autoanalyzer (Yellow Springs medication. BMI was used as an estimate followed by those with the lowest Si (2.4, Equipment). Plasma insulin levels were of overall adiposity. The waist-to-hip ratio 1.0–5.6). The ORs were nearly identical measured using the dextran-charcoal ra- was used as an estimate of body fat distri- when the analysis was stratified by dia- dioimmunoassay method (24). bution. Cigarette smoking was catego- betic status. For instance, the ORs for Insulin sensitivity was assessed by the rized into “none,”“past,” or “current” CAD in nondiabetic participants were FSIGT with minimal model analysis (25). using a standard questionnaire. Plasma greatest among individuals in second ϭ Glucose (0.3 g/kg in 50% solution) was HDL and LDL cholesterol were measured lowest Si quintile (OR 4.7), followed by ␤ ϭ injected through an intravenous catheter in fresh fasting plasma using the -quan- those with the lowest Si (OR 2.5). at 0 min, and regular human insulin (0.03 tification according to the Lipid Research After adjustment for demographic U/kg) was injected at 20 min. Blood was Clinics. Triglycerides were measured by factors CAD (Table 2, model 1a), an in- collected at Ϫ5, 2, 4, 8, 19, 22, 30, 40, 50, enzymatic method in a glycerol blanked terquartile decrement from the 75th to ϫ Ϫ4 70, 100, and 180 min for insulin and glu- assay (Hitachi Autoanalyzer). 25th percentile in Si (2.21 to 0.41 10 ␮ Ϫ1 Ϫ1 cose determination. Si was calculated by min U ml ) was associated with a mathematical modeling (MINMOD, ver- Statistical analysis 91% increase in CAD (P ϭ 0.001). Similar sion 3.0, 1994). The injection of insulin All analyses were performed in SAS ver- interquartile differences in fasting insulin was necessary to ensure adequate plasma sion 6.08 statistical package (SAS Insti- (from 60 to 132 pmol/l) (model 1b) or 2-h insulin levels for accurate computation of tute, Cary, NC) using Student’s t test and insulin levels (from 216 to 816 pmol/l) insulin sensitivity in a diabetic person ␹2 test for univariate comparisons and (model 1c) were associated with, respec- (26). This version of the FSIGT was vali- logistic regression to estimate the relatively, only 34% (P ϭ 0.034) and 16% dated by comparison with the hyperinsu- tionship between Si and CAD, control- (NS) increases in CAD. A simultaneous linemic-euglycemic clamp (27). ling for potential confounders and effect estimation of the effects of Si, fasting, and modifiers. 2-h insulin (model 1d) indicated that only Definition of CAD The Si was estimated to be 0 for 231 of Si was significantly and independently as- ϭ Ͻ CAD was defined conservatively as past the 298 participants in the lowest Si quin- sociated with CAD (OR 1.84, P myocardial infarction, coronary artery tile. In all logistic regression models 0.006). bypass graft, or percutaneous translumi- shown, an indicator variable was in- Because nearly a one-half of the case ϭ nal coronary angioplasty only if con- cluded for individuals with Si 0, as pre- subjects had type 2 diabetes, which is

782 DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 Rewers and Associates

Table 1—Univariate comparison of the levels of selected cardiovascular risk factors among lated triglyceride levels and hypertension, IRAS participants with and without confirmed CAD but not LDL cholesterol and cigarette smoking, might mediate the effect of Si. CAD cases Control subjects P value Further adjustment for diabetes status (model 3) and obesity (model 4) re- n 91 1,391 moved most of the remaining association Ethnicity 0.712 between S and CAD. One may, however, African Americans 22 (24.2) 391 (28.1) i argue that this could be a case of over Hispanics 32 (35.2) 473 (34.0) adjustment, because the vast majority of Non-Hispanic whites 37 (40.6) 527 (37.9) participants with low S were diabetic Glucose tolerance 0.002 i and/or obese. After adjustment for all co- Normal 27 (29.7) 644 (46.3) variates (model 5), insulin levels did not Impaired 20 (22.0) 312 (22.4) have any effect on CAD, but the inter- Type 2 diabetes 44 (48.3) 435 (31.3) quartile decrement in S was still associ- Sex (women) 37 (40.7) 775 (55.7) 0.005 i ated with a 29% increase in the odds of Age (years) 59.5 Ϯ 6.7 55.3 Ϯ 8.5 Ͻ0.001 CAD. Although not statistically signifi- BMI (kg/m2) 30.2 Ϯ 5.7 29.3 Ϯ 5.8 0.123 cant, this finding may suggest that addi- Waist-to-hip ratio 0.98 Ϯ 0.07 0.94 Ϯ 0.08 Ͻ0.001 Ϫ Ϫ Ϫ tional factors not included in these S (10 4 min ␮U 1 ml 1) 1.07 Ϯ 1.22 1.68 Ϯ 1.92 Ͻ0.001 i analyses may also play a role in the asso- Fasting insulin (pmol/l) 144 Ϯ 93 129 Ϯ 115 0.083 ciation between low S and CAD. Mean fasting insulin (pmol/l)* 129 Ϯ 72 115 Ϯ 86 0.201 i 2-h insulin (pmol/l) 753 Ϯ 517 710 Ϯ 667 0.385 CONCLUSIONS — This is the largest HDL cholesterol (mmol/l) 1.03 Ϯ 0.31 1.16 Ϯ 0.39 Ͻ0.001 epidemiological study to date that has as- Triglycerides (mmol/l) 1.92 Ϯ 1.20 1.63 Ϯ 1.21 0.003 sessed directly insulin sensitivity and re- LDL cholesterol (mmol/l) 3.72 Ϯ 0.96 3.65 Ϯ 0.91 0.572 lated it to fasting and postload insulin Systolic blood pressure (mmHg) 132 Ϯ 18 124 Ϯ 17 Ͻ0.001 levels, traditional cardiovascular risk fac- Diastolic blood pressure (mmHg) 79 Ϯ 10 78 Ϯ 9 0.346 tors, and CAD. Our findings of an associ- Hypertension 52 (58.4) 527 (37.9) Ͻ0.001 ation between low insulin sensitivity and Cigarette smoking 0.049 CAD, largely independent of the effects of None 30 (33.0) 614 (44.2) major cardiovascular risk factors, are con- Past 47 (51.6) 542 (39.0) sistent with previous studies that used Current 14 (15.4) 234 (16.8) fasting insulin levels as a marker of insulin Ϯ Data are n (%) or mean SD. *Average fasting insulin on the oral glucose tolerance test day and on the FSIGT sensitivity (6–13). In contrast to some of day of IRAS examination 2–28 days apart. these previous studies (8,9,30), the association between Si and CAD was highly known to increase the risk of CAD, we A stepwise addition of these risk factors to significant and independent of the effects carried the analyses also stratified by dia- model 1a (data not shown) indicated that of lipids, hypertension, and cigarette betic status. The results were virtually HDL cholesterol levels or closely corre- smoking. These results are also consistent identical for diabetic and nondiabetic participants, and the interaction between the effects of Si and diabetes was nonsignifi- cant (P Ͼ 0.9) in all of the models. In further analyses, we combined diabetic and nondiabetic participants. Adjustment for HDL and LDL cholesterol levels, triglycerides, smoking, and hypertension attenuated the independent association between CAD and Si (model 2a) and removed any association between CAD and fasting (model 2b) or 2-h insulin levels (model 2c). Models 2d, 2e, and 2f further suggested that Si, rather than fasting or 2-h insulin levels, was the independent determinant of CAD. The de- crease in the CAD ORs with adjustment for cardiovascular disease risk factors, Figure 1—OR for CAD by Si quintiles (0–0.24, 0.25–0.83, 0.84–1.44, 1.45–2.67, and 2.68– from 1.91 for (model 1a) to 1.56 (model 19.4 ϫ 10Ϫ4 min ␮UϪ1 mlϪ1), quintiles of fasting insulin (179–1,830, 129–178, 93–128, 65–92, 2a), was consistent with the likely sce- and 7–64 pmol/l), and quintiles of 2-h insulin levels (1,048–6,458, 674–1,047, 453–573, 237–451, nario that some of these factors could me- and 214–236 pmol/l), adjusted for age, sex, ethnicity, and clinic. Individuals in the fifth quintile diate the association between Si and CAD. (highest Si, lowest fasting, or 2-h insulin levels) served as the reference group, IRAS 1992–1995.

DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 783 The Insulin Resistance Atherosclerosis Study

Table 2—OR for CAD associated with a difference between the 75th and 25th percentile in Si adjusted for demographic variables and conventional cardiovascular risk factors, IRAS 1992–1995

Model Independent variables in the model OR (95% CI) for: P value ϩ ϭ 1a Demographic* Si Si 1.91 (1.29–2.83) 0.001 1b Demographic ϩ log(fasting insulin) log(fasting insulin) ϭ 1.34 (1.02–1.76) 0.034 1c Demographic ϩ log(2-h insulin) log(2-h insulin) ϭ 1.16 (0.83–1.63) 0.380 ϩ ϭ 1d Demographic Si Si 1.84 (1.19–2.84) 0.006 ϩ log(fasting insulin) log(fasting insulin) ϭ 1.14 (0.80–1.64) 0.459 ϩ log(2-h insulin) log(2-h insulin) ϭ 0.88 (0.60–1.30) 0.530 ϩ ϩ ϭ 2a Demographic* CVD risk factors† Si Si 1.56 (1.03–2.34) 0.034 2b Demographic ϩ CVD risk factors ϩ log(fasting insulin) log(fasting insulin) ϭ 1.14 (0.83–1.55) 0.418 2c Demographic ϩ CVD risk factors ϩ log(2-h insulin) log(2-h insulin) ϭ 1.03 (0.72–1.47) 0.887 ϩ ϩ ϩ ϭ 2d Demographic CVD risk factors Si log(fasting insulin) Si 1.53 (0.98–2.40) 0.060 log(fasting insulin) ϭ 1.02 (0.72–1.46) 0.898 ϩ ϩ ϩ ϭ 2e Demographic CVD risk factors Si log(2-h insulin) Si 1.59 (1.04–2.44) 0.031 log(2-h insulin) ϭ 0.92 (0.63–1.35) 0.677 ϩ ϩ ϩ ϩ ϭ 2f Demographic CVD risk factors Si log(fasting insulin) log(2-h insulin) Si 1.55 (0.99–2.42) 0.053 log(fasting insulin) ϭ 1.07 (0.73–1.57) 0.740 log(2-h insulin) ϭ 0.90 (0.59–1.36) 0.608 ϩ ϩ ϩ ϭ 3 Demographic CVD risk factors Si diabetes status Si 1.41 (0.92–2.16) 0.116 ϩ ϩ ϩ ϩ ϩ ϭ 4 Demographic CVD risk factors Si diabetes status BMI WHR Si 1.26 (0.82–1.95) 0.286 ϩ ϩ ϩ ϩ ϩ ϭ 5 Demographic CVD risk factors Si diabetes status BMI WHR Si 1.29 (0.81–2.04) 0.283 ϩ log(fasting insulin) ϩ log(2-h insulin) log(fasting insulin) ϭ 0.98 (0.65–1.46) 0.903 log(2-h insulin) ϭ 0.97 (0.62–1.51) 0.880 *Age, sex, clinic, ethnicity; †HDL and LDL cholesterol, triglycerides, cigarette smoking, hypertension. with the previously reported (17–19) as- nificant part of the association between lin resistance but to a variable extent lim- sociation between low Si and carotid ar- low Si and CAD. ited by the pancreas’ ability to secrete tery wall thickness, which is an index of This study is the first to measure in- insulin and modified by ambient glyce- atherosclerosis. A comparison of the inti- sulin sensitivity directly in a large popu- mia and insulin clearance. Therefore, fast- ma-media thickness of the internal ca- lation of people with normal, impaired, or ing insulinemia is a less useful marker of rotid arteries in the IRAS CAD case diabetic glucose tolerance. Whereas it is insulin sensitivity in individuals with di- subjects and control subjects (Fig. 2), more difficult to measure Si than fasting abetes, impaired insulin secretion (e.g., a confirmed that the most insulin-resistant insulinemia, the interpretation of Si (ef- large proportion of individuals with IGT CAD case subjects had the most extensive fectiveness of insulin on glucose kinetics) [15]), some forms of hyperinsulinemia carotid atherosclerosis. Thus, low insulin is easier than that of fasting insulin levels. (e.g., insulinoma), and disorders of insu- sensitivity is associated with both subclin- Fasting insulin levels increase with insulin clearance (e.g., cirrhosis). The major ical carotid atherosclerosis and clinical CAD. The association between Si and carotid wall thickness (17) or CAD (current report) was independent of and much stronger than the associations with fasting or 2-h insulin levels. The exact contribu- tion of the proposed atherogenic effect of insulin (13) to the association between insulin resistance and CAD is difficult to quantify in this cross-sectional analysis but appears to be relatively small (Table 2, model 1a versus 1d). This is consistent with the variable and generally weak associations between insulin levels and CAD reported previously (30). On the other hand, our data confirm that hypertension (31), dyslipidemia (32), and dia- Figure 2—Mean internal carotid artery intima-media wall thickness among CAD case (n ϭ 91) ϭ betes (acting through hyperglycemia or and control (n 1,391) subjects by quintiles of insulin sensitivity (Si, adjusting for age, sex, other risk factors [33,34]) mediate a sig- ethnicity, and clinic); IRAS 1992–1995.

784 DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 Rewers and Associates advantage of the IRAS protocol was the followed prospectively with major cardio- sulin sensitivity, consistent with that re- ability to measure insulin sensitivity in in- vascular disease end points ascertained ported here. dividuals with diabetes who are at a two- through annual participant interviews Fifth, the minimal model measure- to fourfold increased risk of CAD (12,35). and committee review of medical records ment of insulin sensitivity is technically They have typically been excluded from of reported fatal and nonfatal events. A difficult in clinical practice. In search for a previous studies (6–11), yet diabetes af- 10-year follow-up of the study cohort will simpler solution, we substituted Si with the fects, in the U.S., 6–14% of people aged be completed in 2005. homeostasis model assessment (HOMA) 30–64 years and 18–32% of those over Second, the IRAS cohort is not strictly measurement of insulin sensitivity that 64 years (36). population based. The study participants can be derived from the FSIGT (39). In Despite the advantages of the mini- were drawn from two existing popula- none of the models, except for the sim- mal model analysis in assessment of insu- tion-based epidemiological studies and plest model 1a, was HOMA associated lin sensitivity, the method resulted in a from two health maintenance organiza- with CAD. Although easier to obtain than ϳ “zero Si” estimate in 16% of IRAS par- tion populations; however, individuals Si, the HOMA estimate of insulin sensitiv- ticipants (in 2% of those with normal, with IGT and diabetes were over-sampled ity appears to be insufficiently precise for 13% with impaired, and 36% of those by design. On the other hand, demanding studies of IRAS size. with diabetic glucose tolerance). “Zero in- protocol and specific exclusion criteria re- Finally, Si and insulin levels display sulin sensitivity” is a difficult concept to moved from the study population indi- significant variability, partially related to accept; however, we have demonstrated viduals with the most severe diabetes or precision of measurements and partially ϭ that IRAS participants with Si 0 had CAD. Less than expected carotid artery due to acute day-to-day and diurnal more features of the metabolic syndrome atherosclerosis among the most insulin- changes (40). The interclass correlation than other insulin-resistant IRAS partici- resistant IRAS participants reported pre- for Si measured twice within 1 week in 58 Ͼ pants with Si 0 (37). The phenomenon viously (17) and lower than expected IRAS participants was 0.67 compared has been recently explained (38) as an ar- CAD prevalence found in the current with 0.76 for fasting insulin. Thus, it is tifact of a single compartment glucose dis- study in that group could be due to a “sur- unlikely that we measured Si with more tribution assumption underlying the vivor bias.” This could occur if individu- precision than fasting insulin levels and minimal model estimation of Si, which als with the most severe CAD have died, that this could account for the stronger does not include insulin action on hepatic elected not to participate, or were ex- association of CAD with Si than with fast- glucose metabolism. A more exact two- cluded. This potential selection bias ing insulin. We confirmed that by using compartment modeling is not suitable for would tend to underestimate the true as- the average of two fasting insulin mea- field studies due to complexity and use of sociation between Si and CAD. surements (on the oral glucose tolerance a radiolabeled tracer. However, allowing Third, the study population included test day and on the FSIGT day) instead of Si to assume apparently negative values Hispanic and non-Hispanic whites as well a single measurement in alternative mod- could partly correct the deviation and im- as African Americans, but relatively few els 1, 2, and 5. Although some of the ORs prove the correlation with euglycemic end points in each of these subgroups for fasting insulin increased slightly, the clamp derived measure of insulin sensi- limited our ability to detect any ethnic dif- ORs for Si and the associated P values vir- tivity (39). When we recalculated Si,al- ferences in the relation between low Si tually did not change. We did not esti- lowing negative values, the rank of Si and CAD. There were no clear interac- mate the reproducibility of 2-h insulin Ͼ values was virtually unchanged. The ORs tions between Si and ethnicity (P 0.4, levels in IRAS, but they may vary by Ͼ for CAD by quintile of such calculated Si data not shown), and the present analyses 30% in normal subjects studied 48 h (data not shown) looked nearly identical were adjusted for, but not stratified by, apart (41), which is comparable with the to those shown in Fig. 1, which were cal- ethnicity. reproducibility of Si and fasting insulin. culated using traditional Si values. This Fourth, there could have been some Therefore, differential measurement pre- could be expected because Si estimates misclassification of the CAD status using cision of Si and insulin levels is unlikely to from the two-compartment model corre- the study criteria. Only 91 participants explain the apparent independence and late perfectly with the one-compartment with most severe clinical or ECG manifes- greater strength of the association be- model Si estimates (38). Therefore, tations of CAD were classified as “case tween Si and CAD compared with that whereas the minimal model systemati- subjects,” whereas obviously many more between insulin levels and CAD. cally underestimated insulin sensitivity, had some degree of CAD but were classi- In middle-aged women and men rep- compared with the euglycemic clamp or a fied as “control subjects.” More precise resentative of the three major U.S. ethnic two-compartment model, it provided a procedures to document CAD, such as groups and including individuals with dependable, cost-efficient, and minimally coronary angiography or electron beam normal, impaired, and diabetic glucose invasive way to measure insulin sensitiv- tomography for coronary calcification, tolerance, we found that CAD was cross- ity in a large free-living population. were too invasive or expensive for this sectionally associated with low insulin The present study has several limita- large study. Our definition of CAD most sensitivity. This association was indepen- tions. First, the relation between Si, insu- likely underestimated the true associa- dent of and stronger than that between lin levels, and CAD were assessed cross- tions between CAD and risk factors, in- CAD and fasting or postload insulin lev- sectionally, and the proposed role of low cluding Si. Recently, a study of just 13 els. Dyslipidemia, hypertension, diabetes, insulin sensitivity as one of the causes of case subjects with arteriographically doc- obesity, and fat centrality explained part CAD needs to be confirmed in longitu- umented CAD and 10 control subjects (3) of the association between low insulin dinal studies. The IRAS cohort is being found a significant difference in their in- sensitivity and CAD.

DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 785 The Insulin Resistance Atherosclerosis Study

10. Perry IJ, Wannamethee SG, Whincup PH, LB, Mayer EJ, Orleans M, Murphy JR, Acknowledgments— This study was sup- Shaper AG, Walker MK, Alberti KGMM: Lezotte DC: Methods and prevalence of ported by the National Heart, Lung, and Blood Serum insulin and incident coronary non-insulin dependent diabetes mellitus Institute Awards U01-HL-47887, HL-47889, heart disease in middle-aged British men. in a biethnic Colorado population: the HL-47890, HL-47892, and HL-47902. In ad- Am J Epidemiol 144:224–234, 1996 San Luis Valley Diabetes Study. Am J Epi- dition, the Los Angeles center was supported 11. Despre´s J-P, Lamarche B, Maurie`ge P, demiol 129:295–311, 1989 by grant M01-RR-43 from the National Center Cantin B, Dagenais GR, Moorjani S, Lu- 23. World Health Organization: Diabetes Mel- for Research Resources/National Institutes of pien PJ: Hyperinsulinemia as an indepen- litus: Report of a WHO Study Group. Ge- Health. dent risk factor for ischemic heart disease. neva, World Health Org., 1985 (Tech. The authors would like to thank the women N Engl J Med 334:952–957, 1996 Rep. Ser., no. 727) and men who participated in this study. We 12. Folsom AR, Liao F, Szklo M, Smith R, 24. Herbert V, Lau K, Gottlieb C, Bleicher S: would also like to acknowledge the valuable Stevens J, Eckfeldt JH: A prospective Coated charcoal immunoassay of insulin. contributions of additional IRAS investigators, study of coronary heart disease in relation J Endocrinol Metab 25:1375–1384, 1965 clinical, and technical staff. to fasting insulin, glucose, and diabetes. 25. Bergman RN, Finegood DT, Ader M: As- Diabetes Care 20:935–942, 1997 sessment of insulin sensitivity in vivo. En- 13. Stout RW: Insulin and atheroma: 20-year docr Rev 6:45–86, 1985 References perspective. Diabetes Care 13:631–655, 26. Welch S, Gebhart SSP, Bergman RN, Phil- 1. Reaven GM: Role of insulin resistance in 1990 lips LS: Minimal model analysis of intra- human disease. Diabetes 37:326–337, 14. Kahn SSE, Prigeon RL, McCulloch DK, venous glucose tolerance derived insulin 1988 Boyko EJ, Bergman RN, Schwartz MW, sensitivity in diabetic subjects. J Endocri- 2. DeFronzo RA, Ferrannini E: Insulin resis- Neiﬁng JL, Ward WK, Beard JC, Palmer nol Metab 71:1508–1518, 1990 tance: a multifaceted syndrome responsi- JP, Porte D Jr: Quantiﬁcation of the rela- 27. Saad MF, Anderson RL, Laws A, Wa- ble for NIDDM, obesity, hypertension, tionship between insulin sensitivity and tanabe RM, Kades WW, Chen Y-DI, Sands dyslipidemia, and atherosclerotic cardio- ␤-cell function in human subjects: evi- RE, Pei D, Savage PJ, Bergman RN: A com- vascular disease. Diabetes Care 14:173– dence for a hyperbolic function. Diabetes parison between the minimal model and 194, 1991 42:1663–1672, 1993 glucose clamp in the assessment of insulin 3. Young MH, Jeng CY, Sheu WH, Shieh SM, 15. Laakso M: How good a marker is insulin sensitivity across the spectrum of glucose Fuh MM, Chen YD, Reaven GM: Insulin level for insulin resistance? Am J Epidemiol tolerance. Diabetes 43:1114–1121, 1994 resistance, glucose intolerance, hyperin- 137:959–965, 1993 28. The ARIC Investigators: The Atheroscle- sulinemia and dyslipidemia in patients 16. Ludvik B, Nolan JJ, Baloga J, Sacks D, rosis Risk in Communities (ARIC) Study: with angiographically demonstrated cor- Olefsky J: Effect of obesity on insulin re- design and objectives. Am J Epidemiol 129: onary artery disease. Am J Cardiol 72: sistance in normal subjects and patients 687–702, 1989 458–460, 1993 with NIDDM. Diabetes 44:1121–1125, 29. Prineas RJ, Crow RS, Blackburn H: The 4. Shinozaki K, Suzuki M, Ikebuchi M, Hara 1995 Minnesota Code: Manual of Electrographic Y, Harano Y: Demonstration of insulin re- 17. Howard G, O’Leary DH, Zaccaro D, Findings. Boston, John Wright-PGS, 1982 sistance in coronary artery disease docu- Haffner S, Rewers M, Hamman R, Selby 30. Wingard DL, Ferrara A, Barrett-Connor mented with angiography. Diabetes Care JV, Saad MF, Savage PJ, Bergman R: Insu- EL: Is insulin really a heart disease risk 19:1–7, 1996 lin sensitivity and atherosclerosis. Circu- factor? Diabetes Care 18:1299–1304, 1995 5. Bressler P, Bailey SR, Matsuda M, De- lation 93:1809–1817, 1996 31. Ferrannini E, Buzzigoli G, Bonadonna R, Fronzo RA: Insulin resistance and coro- 18. Watarai T, Yamasaki Y, Ikeda M, Kubota Giorco MA, Oleggini M, Graziadei L, nary artery disease. Diabetologia 39:1345– M, Kodama M, Tsujino T, Kishimoto M, Pedrinnelli R, Brandi L, Bevilacqua S: In- 1350, 1996 Kawamori R, Hori M: Insulin resistance sulin resistance in essential hypertension. 6. Welborn TA, Wearne K: Coronary heart contributes to carotid arterial wall thick- N Engl J Med 317:350–357, 1987 disease incidence and cardiovascular ness in patients with non-insulin-depen- 32. Orchard TJ, Becker DJ, Bates M, Kuller mortality in Busselton with reference to dent-diabetes mellitus. Endocr J 46:629– LH, Drash AL: Plasma insulin and li- glucose and insulin concentrations. Dia- 638, 1999 poprotein concentrations: an atherogenic betes Care 2:154–160, 1979 19. Wohlin M, Sundstrom J, Arnlov J, Andren association? Am J Epidemiol 118:326– 7. Pyo¨ra¨la¨ M, Miettinen H, Laakso M, B, Zethelius B, Lind L: Impaired insulin 337, 1983 Pyo¨ra¨la¨ K: Hyperinsulinemia predicts sensitivity is an independent predictor of 33. Barrett-Connor E: Does hyperglycemia coronary heart disease risk in healthy common carotid intima-media thickness really cause coronary heart disease? Dia- middle-aged men: the 22-year follow-up in a population sample of elderly men. betes Care 20:1620–1623, 1997 results of the Helsinki Policemen Study. Atherosclerosis 170:181–185, 2003 34. Lehto S, Ro¨nnemaa T, Haffner SM, Circulation 98:398–404, 1998 20. Wagenknecht LE, Mayer EJ, Rewers MR, Pyo¨ra¨la¨ K, Kallio V, Laakso M: Dyslipide- 8. Eschwege E, Ducimetiere P, Thibult N, Haffner S, Selby J, Borok GM, Henkin L, mia and hyperglycemia predict coronary Richard JL, Claude JR, Rosselin GE: Cor- Howard G, Savage PJ, Saad MF, Bergman heart disease events in middle-aged pa- onary heart disease mortality in relation RN, Hamman R: The Insulin Resistance tients with NIDDM. Diabetes 46:1354– with diabetes, blood glucose and plasma Atherosclerosis Study (IRAS): Objectives, 1359, 1997 insulin levels: the Paris prospective study design, and recruitment results. Ann Epi- 35. Wingard DL, Barrett-Connor E: Heart dis- 10 years later. Horm Metab Res 15 demiol 5:464–472, 1995 ease and diabetes. In Diabetes in America. (Suppl.):41–45, 1985 21. Stern MP, Rosenthal M, Haffner SM, Ha- Ed 2. Bethesda, MD, NIH, NIDDK, 1995, 9. Yarnell JWG, Sweetnam PM, Marks V, zuda HP, Franco LJ: Sex differences in the p. 429–548 (NIH publ. no. 95–1468) Teale JD, Bolton CH: Insulin in ischaemic effects of sociocultural status on diabetes 36. King H, Rewers M, for WHO Ad Hoc Di- heart disease: are associations explained and cardiovascular risk factors in Mexican abetes Reporting Group: Global estimates by triglyceride concentrations: the Caer- Americans: the San Antonio Heart Study. for prevalence of diabetes and impaired pholly prospective study. Br Heart J 171: Am J Epidemiol 120:834–851, 1984 glucose tolerance in adults. Diabetes Care 293–296, 1994 22. Hamman RF, Marshall JA, Baxter J, Kahn 16:157–177, 1993

786 DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 Rewers and Associates

37. Haffner SM, D’Agostino R Jr, Festa A, ment of glucose effectiveness and insulin tions in man. Diabetologia 28:412–419, Bergman RN, Mykkanen L, Karter A, Saad sensitivity from minimal model analysis: 1985 MF, Wagenknecht LE: Low insulin sensi- the effect of single compartment glucose 40. Van Cauter E, Polonsky KS, Scheen AJ: ϭ tivity (SI 0) in diabetic and nondiabetic distribution assumption. Diabetes 46: Roles of circadian rhythmicity and sleep subjects in the insulin resistance athero- 1813–1821, 1997 in human glucose regulation. Endocr Rev sclerosis study: is it associated with com- 39. Matthews DR, Hosker JP, Rudenski AS, 18:716–738, 1997 ponents of the metabolic syndrome and Naylor BA, Treacher DF, Turner R: Ho- 41. Olefsky JM, Reaven GM: Insulin and glu- nontraditional risk factors? Diabetes Care meostasis model assessment: insulin re- cose responses to identical oral glucose 26:2796–2803, 2003 sistance and ␤-cell function from fasting tolerance test performed 48 hours apart. 38. Ni T-C, Ader M, Bergman RN: Reassess- plasma glucose and insulin concentra- Diabetes 23:449–453, 1974

DIABETES CARE, VOLUME 27, NUMBER 3, MARCH 2004 787