Author + information
- Received March 12, 1998
- Revision received June 8, 1999
- Accepted June 30, 1999
- Published online November 1, 1999.
- Keijiro Saku, MD, PhD, FACCa,∗,†,* (, )
- Bo Zhang, MS, PhDa,∗,†,
- Kazuyuki Shirai, MD, PhDa,∗,†,
- Shiro Jimi, PhD∗,
- Kazuhiko Yoshinaga, BSa,∗,† and
- Kikuo Arakawa, MD, PhD, FACCa,∗,†
- ↵*Reprint requests and correspondence: Dr. Keijiro Saku, Department of Internal Medicine, Fukuoka University School of Medicine, 7-45-1 Nanakuma Jonan-ku, Fukuoka 814-0180, Japan
Part of this work was presented at the 63rd Annual Scientific Session of the Japanese Circulation Society, March 27th, 1999, Tokyo, Japan.
The purpose of this study was to investigate the association among insulin resistance, high density lipoprotein cholesterol (HDL-C) and coronary heart disease (CHD), and to test the hypothesis that HDL-C may ameliorate the adverse effects of insulin.
Serum low HDL-C (hypoalphalipoproteinemia) and hyperinsulinemia are independent predictors for CHD, but a strong negative correlation exists between them, as in patients with syndrome X.
Fifty-four pairs of cases (M/F: 49/5), defined as patients with angiographically proved CHD, and control subjects (M/F: 49/5) matched with cases with regard to gender and age were included. Insulin resistance was assessed by the homeostasis model assessment (HOMA).
Cases had increased HOMA insulin resistance and lower serum levels of HDL-C than controls. A receiver operating characteristic (ROC) curve analysis indicated that HDL-C and insulin resistance were significant discriminators of CHD (area under ROC curve: 0.72 and 0.69, respectively). The interaction between HDL-C and the association of insulin resistance with CHD was significant: subjects with hyperinsulinemia and high HDL-C had no increased risk of CHD. Multivariate conditional logistic regression analysis showed that hyperinsulinemic hypoalphalipoproteinemia was a stronger indicator for CHD than either HDL-C or insulin resistance alone (−2 log likelihood: 19.0 vs. 12.6 or 15.7).
Hyperinsulinemic hypoalphalipoproteinemia was a more potent indicator for CHD than either insulin resistance or low serum HDL-C levels alone, and the adverse effects of hyperinsulinemia seem to be ameliorated by high HDL-C levels.
An inverse relationship between plasma levels of high density lipoprotein cholesterol (HDL-C) and coronary heart disease (CHD) has been well established (1–3). Direct evidence for the antiatherogenic effects of HDL has recently been obtained in studies of the over- or underexpression of apolipoprotein (apo) A-I using genetic animal models of reverse cholesterol transport (4–8). Several large population-based prospective studies have identified hyperinsulinemia as an independent risk factor of CHD (9–13). Hyperinsulinemia, as a compensatory response to insulin resistance, has been strongly associated with decreased HDL-C and other metabolic abnormalities that are secondary to hyperinsulinemia, as in the multiple metabolic syndrome and insulin resistance syndrome (“syndrome X”) (14).
Because hyperinsulinemia and low HDL (hypoalphalipoproteinemia) strongly correlate but predict the risk of CHD independently, it is of great interest whether or not these two variables interact in their association with CHD. However, this point has not been investigated previously. Therefore, in the present case-control study, we assessed the influence of serum levels of HDL-C on the association between insulin resistance as determined by the homeostasis model assessment (HOMA) model (15)and CHD.
Cases (patients with CHD) were selected from patients who underwent diagnostic coronary angiography for suspected or known coronary atherosclerosis or for other reasons (mostly atypical chest pain) at Fukuoka University Hospital from 1994 to 1997 and were defined as those who had one, two or three stenosed (>50% luminal narrowing) epicardial coronary arteries (n = 85, F/M: 12/73, age: 62.1 ± 10.6 years). Control subjects (those subjects without CHD, n = 111, F/M: 52/59, age: 57.5 ± 12 years) were selected from those who had come to our heart clinic for a medical check up or because they had various symptoms (without episodes of chest pain) and were confirmed to be free of CHD based on their medical history, exercise-stress electrocardiogram or negative results of single photon emission computed tomography (SPECT). This study was approved by the ethics committee of Fukuoka University Hospital and informed consent was obtained from each patient. Patients with diabetes mellitus (DM), acute myocardial infarction (MI) (within three weeks after onset), heart failure (Killip class ≥2 after MI), vascular disease (aortitis treated by prednisolone) or hepatic dysfunction (virus and nonvirus, transaminases more than three times the normal value) were excluded from the study. Patients with systolic or diastolic blood pressure >140 mm Hg or 90 mm Hg or who were receiving antihypertensive treatment were considered to have hypertension (HT). Patients who were being treated for DM or who had symptoms of DM and a fasting glucose concentration ≥126 mg/dl were considered to have DM. Otherwise, the results of a 75-g oral glucose tolerance test were used to diagnose DM. None was receiving hormone replacement therapy.
Of the 85 patients with CHD and 111 control subjects, 54 pairs of cases and controls were selected by matching them on a one-to-one basis with regard to gender and age. The categoric variable, gender, was matched perfectly. Age was matched by applying the minimum-distance case-control matching method (16,17). The largest difference between a case and a control was three years (control subjects were ±1 year of the case’s age). Thirty-one cases and 53 control subjects were rejected because they did not match.
Coronary arteries were cannulated by the Judkins technique (18)with 5F catheters and recorded on Kodak 35 mm cinefilm at a rate of 25 frames/s. Coronary arteries were divided into 15 segments according to the classification of the American Heart Association Grading Committee. Coronary artery segments were carefully selected by two expert cardiologists on the basis of smooth luminal borders and the absence of stenotic changes. The presence of stenoses was determined using a computer-assisted coronary angiography analysis system (Micron 1; Kontron Co., Tokyo, Japan) after the direct intracoronary injection of isosorbide dinitrate (ISDN) (2.5 mg/5 ml solution), as described previously (19–21). Arterial stenosis that produced more than 50% luminal narrowing was considered significant.
Determination of lipids, lipoproteins and apolipoproteins
Blood was drawn in the morning after an overnight fast. Serum total cholesterol (TC), triglyceride (TG), HDL-C, apo A-I, apo A-II and apo B were determined as described previously (21).
Estimation of insulin resistance
Plasma glucose concentrations were measured by the glucose oxidase method, and plasma immunoreactive insulin concentrations were measured by a standard radioactive immunoassay. Insulin resistance was estimated with the HOMA model (15), using fasting plasma insulin and glucose concentrations. The HOMA insulin resistance has been shown to correlate well with insulin resistance measured using a euglycemic clamp (r = 0.88, p < 0.0001) (15).
Statistical analysis was performed using the SAS Software Package (Version 6.12, Statistical Analysis System, SAS Institute Inc., Cary, North Carolina) at Fukuoka University. Categoric variables (such as gender) and frequency distribution patterns of insulin resistance and serum HDL-C in cases and controls were compared by a chi-square analysis (22). The distribution of continuous variables was examined by the Shapiro-Wilk test (22). Differences in continuous variables between pairs of cases and control subjects were examined by a repeated measures analysis of variance (ANOVA) (23), and other variables, such as body mass index (BMI), were adjusted for by a repeated measures analysis of covariance (ANCOVA) (23). Correlations between HOMA insulin resistance and lipid variables were examined by the Pearson correlation (23). Linear relationships between continuous variables were examined by linear regression models with least-squares fitting (24). Interaction effects between regression lines were examined by the general linear model (24).
The predictive value of HOMA insulin resistance and HDL-C as continuous variables was assessed using a receiver operating characteristic (ROC) curve analysis (25,26). Sensitivity was calculated as true positives/(true positives + false negatives) and specificity was calculated as true negatives/(true negatives + false positives) (27). A ROC curve (plot of sensitivity vs. 1−specificity) analysis is a powerful tool for assessing a test’s ability to discriminate between two groups of subjects and does not depend on the threshold (cutoff) value selected. The area under the ROC curve represents the probability for a randomly chosen disease-free subject (control) to exhibit a value lower (such as HOMA insulin resistance) or higher (such as HDL-C) than the level observed among randomly chosen diseased subjects (cases). A value of 0.5 means that the distributions of the values in the two groups are similar; conversely, a value of 1 means that distributions of values in the two groups do not overlap. We determined the area under the ROC curve by the trapezoidal rule, as in a linear logistic regression model (28), and evaluated its significance by the Wald chi-square test (28). The ROC curve also allows variable thresholds to be determined based on a target sensitivity or specificity. We set the cutoff value for each continuous variable at 0.70 (70%) specificity by using a modified ROC analysis in which we constructed and applied a “two-graph-ROC” plot (29,30). Dummy variables were produced using these cutoff values. A dummy variable was given a value of 0 for subjects with a value less than (<) or equal to (=) the cutoff value (low-value group) and a value of 1 for those with a value higher than (>) the cutoff value (high-value group).
The interaction effects of HDL-C with the association between HOMA insulin resistance and CHD were examined both: 1) by plotting the differences in HOMA insulin resistance between cases and controls versus HDL-C levels (continuous) of cases and testing the significance of the linear trend, and 2) by stratifying all of the subjects according to their HDL-C levels (two levels) and then testing significance of the difference in HOMA insulin resistance between cases and controls for the low-HDL-C and high HDL-C strata (groups) by the Wilcoxon rank-sum test (24).
The associations of HOMA insulin resistance and HDL-C with the risk of CHD in 54 pairs of cases and gender- and age-matched control subjects were examined by a conditional logistic regression analysis with the proportional hazards regression (PHREG) procedure (28), by using the discrete logistic model and forming a stratum for each matched set. For each continuous variable, the cutoff value upon which to base dummy variables was chosen to give a specificity of 70%. We examined the interdependence of the associations of HOMA insulin resistance and HDL-C with CHD and adjusted for other lipid variables and conventional risk factors by a multivariate conditional logistic regression analysis (28). Independent indicators for the presence of CHD were also selected by a stepwise conditional logistic regression analysis (28). For all of the adjusted odds ratios, we calculated 95% confidence intervals (CIs). For regression coefficients in the conditional logistic regression analysis, we showed the standard error. All p values are two-tailed. The significance level was considered to be 5% unless otherwise indicated.
Table 1shows the characteristics of 54 pairs of cases and control subjects matched with regard to gender and age. No statistically significant differences in age, BMI, HT or smoking status were found between cases and control subjects.
Figure 1shows the frequency distributions of insulin resistance as assessed by the HOMA model (15), as well as serum levels of HDL-C in cases and control subjects. As shown in Figure 1A, distribution of the HOMA insulin resistance was significantly different between cases and controls; the distribution of HOMA insulin resistance in cases was shifted toward higher values compared with that in control subjects. By definition, one third of control subjects were found in each tertile of HOMA insulin resistance. In contrast, a smaller proportion of cases (14.8%) was found within the first tertile, and a larger proportion of cases (64.8%) was found within the third tertile. As shown in Figure 1B, the distribution of serum HDL-C levels was also significantly different between cases and controls; the distribution of HDL-C in cases shifted toward lower values compared with that in control subjects. A larger proportion of cases (68.5%) compared with controls (35.2%) was found within the first tertile of HDL-C, and a smaller proportion of cases (5.6%) was found within the third tertile.
Table 2compares fasting levels of plasma glucose and insulin, HOMA insulin resistance, and serum levels of lipids, HDL-C and apolipoproteins in 54 pairs of cases and controls matched by gender and age. Cases had significantly higher values of fasting insulin (10.0 ± 6.3 μU/ml vs. 6.1 ± 3.4 μU/ml, p < 0.05) and HOMA insulin resistance (2.3 ± 1.5 vs. 1.4 ± 0.8, p < 0.05) and lower levels of serum HDL-C (42.0 ± 10.8 mg/dl vs. 52.9 ± 17.1 mg/dl, p < 0.05), apo A-l and apo A-II than control subjects, as assessed by a repeated measures ANOVA (23). No significant differences in fasting levels of plasma glucose or serum levels of TC and TG were observed between cases and controls. Therefore, cases, as compared with controls, were characterized by hyperinsulinemia that compensated for insulin resistance.
Because cases and control subjects were matched for age by the minimum-distance matching method (17), we tested the interaction effects of age with the effects of HOMA insulin resistance and HDL-C on the risk of CHD by plotting the differences in these two variables between cases and control subjects versus the age of the cases, respectively (data not shown), and assessing the significance of the linear trend by a regression analysis. In both cases, the null hypothesis of no linear trend could not be rejected with p values of 0.91 and 0.17, respectively, which shows that the matching variable, age, had no interaction effects with the relation between HOMA insulin resistance, HDL-C and CHD.
Figure 2Ashows the ROC curves (25,26)of HOMA insulin resistance. As shown, HOMA insulin resistance showed a moderate ability to discriminate cases from controls. Table 3(left and middle columns) shows the areas under the ROC curves for continuous variables. As shown, fasting plasma insulin levels (area under ROC curve, 0.701, p < 0.01), HOMA insulin resistance (0.694, p < 0.01) and serum levels of HDL-C (0.716, p < 0.01), apo A-I and apo A-II showed significant discriminative ability for CHD. The predicative value of insulin resistance and HDL-C for CHD were similar as indicated by their area under the ROC curve. Figure 2Bshows the two-graph-ROC plots (29)for HOMA insulin resistance, to demonstrate how the cutoff value for making dummy variables was determined. As shown, cutoff values were those that gave a specificity of 0.7 (70%). Table 3(right column) shows the cutoff values of continuous variables (1.48 for HOMA insulin resistance and 43.6 mg/dl for HDL-C).
In all of the subjects, HOMA insulin resistance showed strong correlations with BMI (r = 0.27, p < 0.01), fasting plasma levels of glucose and insulin (r = 0.28 and 0.98, p < 0.01) and serum levels of HDL-C and apo A-I (r = −0.36 and −0.30, p < 0.01). Correlation coefficients were essentially similar in cases and control subjects, except for HDL-C. As shown in Figure 3A, the regression lines of HOMA insulin resistance versus HDL-C in the cases (regression coefficient ± SE = −0.046 ± 0.018, t = −2.46, p < 0.05) and control subjects (regression coefficient ± SE = −0.014 ± 0.007, t = −2.05, p < 0.05) interacted at a significance level of less than 0.1 (F value = 3.23, p = 0.075), as assessed by an ANCOVA (23); for example, HOMA insulin resistance in cases increased more than that in controls with decreasing HDL-C levels (Fig. 3A).
Because a significant inverse correlation was found between HOMA insulin resistance and HDL-C (Fig. 3A), the interaction of HDL-C with the association between HOMA insulin resistance and CHD was tested in two ways: by treating HDL-C as a continuous (Fig. 3B)or a category variable (Fig. 4), respectively. As illustrated in Figure 3B, the differences in HOMA insulin resistance between cases and control subjects were greater for cases with lower HDL-C than for those with higher HDL-C. In fact, the average difference approached zero at an HDL-C level of 50 mg/dl. This interaction was significant as tested by assessing the linear trend (regression coefficient ± SE = −0.056 ± 0.026, t = −2.15, p < 0.05). As shown in Figure 4, when HDL-C was used as a stratification variable (two levels), HOMA insulin resistance appeared to be more strongly related to CHD for the low HDL-C stratum than for the high HDL-C stratum. These results suggest that HDL-C may ameliorate the adverse effects of HOMA insulin resistance.
To examine the contribution of HOMA insulin resistance to the risk of CHD and its interdependence with HDL-C, a series of multivariate conditional logistic analyses to predict the case-control status were performed (Table 4). As illustrated in model 1, the risk of CHD for subjects with high HOMA insulin resistance was 3.7-fold (95% CI, 1.6 to 8.4) higher (p < 0.01) than that in subjects with low HOMA insulin resistance. As shown in model 2, the risk of CHD for high HDL-C was 3.1 (1/0.32)-fold (95% CI, 1.4 [1/0.69] to 6.7 [1/0.15]) lower (p < 0.01) than that for low HDL-C. As shown in model 3, inclusion of both HOMA insulin resistance and HDL-C in the same model attenuated to some extent the associations between HOMA insulin resistance and CHD (odds ratio [95% CI]: 3.0 [1.2 to 7.0], p < 0.05) and between low HDL-C and CHD (2.4 [1/0.42], 1.1 [1/0.95] to 5.3 [1/0.19], p < 0.05), but these associations remained significant (Table 4). Further adjustment for TC (model 4), TG (model 5) and BMI, HT and smoking status together (model 6) had very little influence on the relationships between HOMA insulin resistance, HDL-C and CHD (Table 4). These results suggest that high HOMA insulin resistance (hyperinsulinemia) and low HDL are independent indicators for CHD.
Because the association between HOMA insulin resistance and CHD varied with the HDL-C level, as shown in Figure 4, we combined HOMA insulin resistance and low HDL-C into one variable (hyperinsulinemic hypoalphalipoproteinemia), which was given a value of 1 if both high HOMA insulin resistance and low HDL-C were present and a value of 0 for all other cases (high-HOMA insulin resistance and high-HDL-C or low-HOMA insulin resistance and high HDL-C or low-HOMA insulin resistance and low HDL-C) and compared the predictive value of hyperinsulinemic hypoalphalipoproteinemia with those of hyperinsulinemia and hypoalphalipoproteinemia, after adjusting for conventional risk factors (Table 5). As shown in Table 5, the presence of hyperinsulinemic hypoalphalipoproteinemia was more strongly associated with the risk of CHD than either hyperinsulinemia or hypoalphalipoproteinemia alone, as judged by the model fitting criterion, −2 log likelihood (28). The interdependence of these variables based on their association with CHD was examined by a stepwise conditional logistic regression analysis (Table 6). As shown in the upper panel of Table 6, both hyperinsulinemia and hypoalphalipoproteinemia were selected as independent indicators for CHD when hyperinsulinemia, hypoalphalipoproteinemia, TC, TG, BMI, HT and smoking status were included in the model as independent variables (model 1). As shown in the lower panel of Table 6, only hyperinsulinemic hypoalphalipoproteinemia was selected as an independent indicator for CHD, when hyperinsulinemic hypoalphalipoproteinemia was added to model 1 as an independent variable (model 2), which agrees with the results in Table 5. These results suggest that the combination of HOMA insulin resistance and low HDL-C was a better indicator for CHD than either insulin resistance or HDL-C alone.
Figure 4also shows the odds ratios for the combination of HOMA insulin resistance and HDL-C. As shown, the respective risk of CHD in subjects with high HOMA insulin resistance and high HDL-C (odds ratio [95% CI]: 2.2 [0.65 to 7.5], ns.) and in those with low HOMA insulin resistance and low HDL-C (1.6 [0.41 to 6.3], ns.) were not significantly increased relative to low-HOMA insulin resistance high-HDL-C subjects (1.0). However, subjects with high HOMA insulin resistance and low HDL-C had an 8.7-fold higher (2.5 to 30, p < 0.001) risk of CHD than subjects with low-HOMA insulin resistance and high-HDL-C, which is much higher than could be expected by a multiplicative model (3.5 = 2.2 × 1.6) (31), suggesting that hyperinsulinemia and hypoalphalipoproteinemia synergistically increase the risk of CHD.
In summary, HOMA insulin resistance and HDL-C are independently associated with CHD. However, the adverse effects of hyperinsulinemia may be ameliorated by high HDL-C levels. Hyperinsulinemic hypoalphalipoproteinemia was a more potent indicator for CHD than either increased insulin resistance or low HDL-C alone. Increased HOMA insulin resistance and low HDL-C increase the risk of CHD synergistically.
Hyperinsulinemia and insulin resistance are main characteristics of syndrome X and result in secondary syndrome X features, including hyperglycemia, increased very low density lipoprotein (VLDL) concentrations, decreased HDL-C and HT (14). The antiatherogenic effects of HDL have been directly documented by studies using transgenic animal models (4–8). Therefore, we tested the hypothesis that HDL-C levels may ameliorate the adverse effects of insulin resistance.
ROC analysis of the ability of insulin resistance and HDL-C to predict CHD
In the present case-control study, cases were patients with angiographically proved CHD, and non-CHD controls were matched with cases for gender (perfect matching) and age (minimum-distance matching), because gender and age are strongly correlated with both the prevalence of CHD and the predicative variables, i.e., HOMA insulin resistance and HDL-C.
Cases, compared with control subjects, were characterized by hyperinsulinemia, as indicated by higher fasting levels of insulin, and normal glucose levels (Table 2). This is consistent with the findings of several large population studies that have investigated the association between hyperinsulinemia and CHD (10–13). Our finding that cases had lower serum levels of HDL-C, apo A-I and apo A-II (Table 2)agrees with the established inverse relationship between HDL and CHD (1–3).
The results of our ROC curve analysis (25,26), which did not depend on the cutoff value, showed that HDL-C and insulin resistance are significant discriminators for CHD and have a similar discriminatory ability. Although the associations between hyperinsulinemia and CHD and between HDL-C and CHD have been extensively studied, the discriminatory abilities of these two predictors have not been previously compared.
Relationship between insulin resistance and HDL-C
Our finding that insulin resistance (hyperinsulinemia) was inversely correlated with HDL-C is consistent with the results of other studies (32–34). Although the mechanism of this association is not quite clear, enhanced hepatic lipase activity and the impaired activity of the insulin-dependent enzyme lipoprotein lipase observed in type II DM have been suggested to play a role (35). We also observed a strong positive correlation between insulin resistance (hyperinsulinemia) and serum TG levels in controls (data not shown). A direct relationship between insulin and hypertriglyceridemia has also been suggested by other studies (32–34). Increased fatty acid production and impaired VLDL clearance in type II DM have been proposed as possible underlying mechanisms (35).
Conditional logistic regression analysis of the association among insulin resistance, HDL-C and CHD
Because a logistic regression analysis using dummy variables does not depend on the distribution of predicative variables, a conditional logistic regression analysis was used to examine the association among the prevalence of CHD, HDL-C and insulin resistance in pairs of cases and gender- and age-matched control subjects. Our finding that HDL-C and insulin resistance are independent indicators of CHD after considering confounding variables (Table 4)is consistent with many previous studies (1–3,9–13).
Influence of HDL-C on the association between insulin resistance and CHD
Our findings that insulin resistance was inversely correlated with HDL-C levels in both cases and control subjects, yet with different regression coefficients (Fig. 3A), and was associated with CHD independent of HDL-C, encouraged us to test the interaction of HDL-C with the association between insulin resistance and CHD. Our hypothesis that the adverse effects of hyperinsulinemia on CHD may be ameliorated by high HDL-C levels was supported by our finding of a significant interaction effect (Fig. 3B and 4; the association between insulin resistance and CHD was significant when HDL-C was low but not significant when HDL-C was high), and by our finding that hyperinsulinemic hypoalphalipoproteinemia was a stronger indicator for CHD than either insulin resistance or low HDL-C alone, as assessed by a multivariate conditional logistic regression analysis (Tables 5 and 6). In addition, our results also suggest that hyperinsulinemia and hypoalphalipoproteinemia increase the risk of CHD synergistically (Fig. 4). The interaction between insulin resistance and HDL-C in their association with CHD has not been investigated previously.
There is some evidence that insulin may promote atherogenesis by directly affecting the arterial wall. In animal studies, insulin enhances the development of atherosclerotic plaque (36), and in vitro, insulin has been shown to cause smooth muscle cell proliferation (37), stimulate low density lipoprotein binding to smooth muscle cells (38), fibroblasts (39)and monocytes (40); it also stimulates cholesterol synthesis in monocytes (41). It has been shown that high levels of HDL protect against CHD, possibly due to its role in reverse cholesterol transport (42). Recent studies in transgenic mice have indicated that HDLs are directly antiatherogenic (4). New evidence suggests that HDLs inhibit the expression of cell adhesion molecules that are required for the interaction between leukocytes and the endothelium in the early stage of the development of early atherosclerosis (43). Therefore, the adverse effects of hyperinsulinemia on CHD may be ameliorated by HDLs.
We examined the association among insulin resistance, HDL-C and CHD in a case-control study. However, whether or not HDL-C has a causal effect on the association of hyperinsulinemia with CHD cannot be determined from a case-control study.
Insulin resistance (hyperinsulinemia) as assessed by the HOMA model and serum levels of HDL-C are independently associated with CHD. However, the adverse effects of hyperinsulinemia seem to be ameliorated by high HDL-C levels. This report is the first to note that insulin resistance and low HDL-C levels may synergistically increase the risk of CHD and that hyperinsulinemic hypoalphalipoproteinemia was a more potent indicator for CHD than either insulin resistance or serum HDL-C levels alone.
☆ This work was supported by grants-in-aid from the Ministry of Education, Science and Culture of Japan (Nos. 07670827, 09670773, 10670221, 10670693, and 11670724), by research grants from the Ministry of Health and Welfare and by research grants from the Fukuoka University Research Fund.
- analysis of covariance
- analysis of variance
- body mass index
- coronary heart disease
- confidence interval
- diabetes mellitus
- high density lipoprotein cholesterol
- homeostasis model assessment
- isosorbide dinitrate
- myocardial infarction
- proportional hazards regression
- receiver operating characteristic
- single photon emission computed tomography
- total cholesterol
- very low density lipoprotein
- Received March 12, 1998.
- Revision received June 8, 1999.
- Accepted June 30, 1999.
- American College of Cardiology
- Gordon D.J,
- Probstfield J.L,
- Garrison R.J,
- et al.
- Plump A.S,
- Scott C.J,
- Breslow J.L
- Garcia M.J,
- McNamara P.M,
- Gordon T,
- Kannel W.B
- Howard G,
- O’Leary D.H,
- Zaccaro D,
- et al.
- Baillie G.M,
- Sherer J.T,
- Weart C.W
- Wynder E.L,
- Stellman S.D
- Smith A.H,
- Kark J.D,
- Cassel J.C,
- Spears G.F
- Saku K,
- Zhang B,
- Ohta T,
- Arakawa K
- ↵(1990) SAS Procedures Guide, Version 6 (SAS Institute Inc, Cary, NC), Third Edition.
- ↵(1990) SAS/STAT User’s Guide, Version 6, Volume 1 (SAS Institute Inc, Cary, NC), Fourth Edition.
- ↵(1990) SAS/STAT User’s Guide, Version 6, Volume 2 (SAS Institute Inc, Cary, NC), Fourth Edition.
- ↵SAS/STAT Software: changes and enhancements through release 6.11. Cary, NC: SAS Institute Inc., 1996:317–314.
- Hamashima N
- Stout R.W
- Young I.R,
- Stout R.W
- Oppenheimer M.J,
- Sundquist K,
- Bierman E.L
- Krone W,
- Nagele H,
- Behnke B,
- Greten H
- Shepherd J,
- Packard C.J
- Cockerill G.W,
- Rye K.A,
- Gamble J.R,
- et al.