Non–High-Density Lipoprotein Cholesterol Levels Predict Five-Year Outcome in the Bypass Angioplasty Revascularization Investigation (BARI)
Background— Current National Cholesterol Education Program guidelines recommend that non–high-density lipoprotein cholesterol (non-HDL-C) be considered a secondary target of therapy among individuals with triglycerides >2.26 mmol/L. It is not known whether non-HDL-C relates to prognosis among patients with coronary heart disease.
Methods and Results— Lipid levels were available at baseline among 1514 patients (73% men; mean age, 61 years) enrolled in the Bypass Angioplasty Revascularization Investigation (BARI); all had multivessel coronary artery disease. Patients were followed for 5 years. Outcomes of death, nonfatal myocardial infarction, and death or myocardial infarction were modeled using univariate and multivariate time-dependent proportional hazards methods; angina pectoris at 5 years was modeled using univariate and multivariate logistic regression. Non-HDL-C was a strong and independent predictor of nonfatal myocardial infarction (multivariate relative risk, 1.049 [95% confidence intervals, 1.006 to 1.093] for every 0.26 mmol/L increase) and angina pectoris (multivariate odds ratio, 1.049 [95% confidence intervals, 1.004 to 1.096] for every 0.26 mmol/L increase), but it did not relate to mortality. HDL-C and LDL-C did not predict events during follow-up.
Conclusions— Among patients with lipid values in BARI, non-HDL-C is a strong and independent predictor of nonfatal myocardial infarction and angina pectoris at 5 years, even after consideration of powerful clinical variables. Our data suggest that non-HDL-C is an appropriate treatment target among patients with coronary heart disease.
Received June 21, 2002; revision received August 28, 2002; accepted September 2, 2002.
Although elevated low-density lipoprotein cholesterol (LDL-C) remains the main focus of therapy in the primary and secondary prevention of coronary heart disease, it is now widely recognized that some triglyceride-rich lipoproteins also contribute to atherogenesis.1 The recently revised National Cholesterol Education Program guidelines thus recommend that non–high-density lipoprotein cholesterol (non-HDL-C), calculated as total cholesterol minus HDL cholesterol, be considered a secondary target of therapy among individuals with triglycerides >2.26 mmol/L (200 mg/dL).2 Non-HDL-C goals are 30 points above the recommended LDL-C cutpoints.2 In contrast to LDL-C levels, which are traditionally calculated by the Friedewald formula and require a fasting blood sample, non-HDL-C can be measured in the nonfasting state, and its calculation does not make any assumptions about lipoprotein composition.1–3
See p 2526
Non-HDL-C has been shown to be a predictor of coronary heart disease and mortality in a number of prospective cohort studies that predominantly or exclusively enrolled individuals free of cardiovascular disease.4–11 To our knowledge, there are no prospective data to validate the use of non-HDL-C as a predictor of outcome among patients with prevalent coronary artery disease. The purpose of the current investigation was to determine the impact of baseline total cholesterol, triglycerides, HDL-C, LDL-C, and non-HDL-C on 5-year outcome in patients enrolled in the Bypass Angioplasty Revascularization Investigation (BARI).
Patients were eligible for enrollment in BARI if they had angiographically documented multivessel coronary artery disease with clinically severe angina or objective evidence of ischemia requiring revascularization and were suitable candidates for both coronary artery bypass grafting (CABG) and percutaneous transluminal coronary angioplasty (PTCA) as an initial revascularization strategy. The design of BARI has been published in detail previously.12 The study was approved by Institutional Review Boards at each participating institution; all study participants provided written, informed consent.
Briefly, between August 1988 and August 1991, 1829 patients were enrolled in 18 clinical centers and randomized to receive PTCA or CABG. Baseline data included demographic, clinical, and angiographic characteristics and information on risk factors, medications, quality of life, and functional status. Follow-up visits were conducted at the clinics at weeks 4 to 14 after study entry and at 1, 3, and 5 years, with interim phone contacts at 6 months and 2 and 4 years. The current analysis includes follow-up through June 1995 for an average follow-up of 5.4 years. Mortality from all causes was the primary BARI end point. Other end points included myocardial infarction, need for further revascularization, angina pectoris at 5 years, and functional status.
BARI Protocol for Lipid Management
In BARI, total cholesterol, HDL-C, and triglycerides were determined at clinic laboratories at baseline, 6 weeks (4 to 14 weeks) after randomization, and at 1, 3, and 5 years of follow-up. Lipid profiles were not obtained in patients who had suffered a myocardial infarction within 6 weeks of any study visit. LDL-C was calculated by the Friedewald equation among patients who had fasted before venipuncture and whose triglycerides were <4.52 mmol/L (<400 mg/dL).3 The importance of risk factor modification was emphasized throughout the study to the patients and their primary physicians. The protocol specified that all patients were to receive diet counseling and that lipid-lowering drugs were to be used if total cholesterol exceeded 6.22 mmol/L (240 mg/dL) or LDL-C exceeded 3.89 mmol/L (150 mg/dL). Niacin and bile acid–binding resins were considered preferred agents, whereas lovastatin, probucol, and gemfibrozil were acceptable alternates. Goals for lipid-lowering therapy included an LDL-C <2.85 mmol/L (<110 mg/dL), HDL-C >1.17 mmol/L (>45 mg/dL), triglycerides <1.36 mmol/L (<120 mg/dL), and total cholesterol/HDL-C ratio <3.5.
Design of the Current Study
Baseline and follow-up lipid profiles were reviewed by the investigators, and obviously erroneous values were set as missing (eg, follow-up HDL-C of 2.59 mmol/L when baseline HDL-C was 1.04 mmol/L). Lipid values obtained at randomization were not considered in subsequent analyses because most were obtained during hospitalization and might thus not reflect each patient’s true baseline values. Analyses subsequently referred to in this article as “baseline” analyses refer to lipid values measured at the early follow-up visit 4 to 14 weeks after randomization. Non-HDL-C was calculated as total cholesterol minus HDL-C. Outcomes considered included all-cause mortality, nonfatal myocardial infarction, the combined end point of death or myocardial infarction, angina pectoris at 5 years, and subsequent revascularization.
Outcomes of death, nonfatal myocardial infarction, and death or myocardial infarction were modeled using univariate and multivariate time-dependent proportional hazards methods. In the models of death and death or myocardial infarction, patients were censored at the last time the patient was known to be alive. Five-year mortality was assessed in 98% of the patients. In the model for myocardial infarction, patients were censored at death or at the last time the patient was known to be event-free. A stepwise regression was used that included covariates found to be predictive in the BARI population.13 Such variables included age, sex, race, coronary heart disease risk factors (excluding lipid values), comorbidities, and clinical coronary artery disease variables such as congestive heart failure, hypertension, abnormal left ventricular function, coronary dominance, and measures of the severity of coronary disease. Any variable that reached a statistical significance of 0.2 was selected for the final model. Baseline lipid variables were then added to the final model to assess if they added any additional information. Lipids were initially added individually to these models. Additional models with 2 lipid variables were also computed (non-HDL-C and HDL-C; triglycerides and HDL-C). For the end point of angina at 5 years, a logistic model was used, and only patients reporting angina status at 5 years were included. Again, the same stepwise approach was used to identify those demographic, clinical, and angiographic variables that predicted angina at 5 years.
Outcome over 5 years is likely to be a function not only of baseline lipid values, but also of “lipid burden over time” during the trial. The proportional hazards models for death, nonfatal myocardial infarction, and death or myocardial infarction were again used, but with time-dependent lipid covariates. The following 3 techniques were used to determine the values of the lipid covariate at each event time point: (1) the last nonmissing value, (2) the average of all nonmissing values up to that time point; and (3) the maximum of all nonmissing values up to that time point. For the logistic regression of angina at 5 years, similar techniques were used to define the lipid value, but the time point was always 5 years. All techniques yielded similar results. Only the analysis using the average of all nonmissing values up to a specified time point is presented in this article.
Availability of Lipoprotein Measurements
Among the 1829 patients randomized, 914 were assigned to CABG and 915 were assigned to PTCA as the initial revascularization strategy. Among the 914 CABG patients, 899 were alive at the time of the baseline lipid measurement, 745 patients (82%) had a total cholesterol measurement, and 712 (78%) had total cholesterol and HDL-C measured. Fasting lipid values were available in 672 CABG patients (74%), and LDL-C calculation was feasible in 621 (68%). Among the 915 patients randomized to PTCA, 901 were alive at the time of the baseline lipid measurement, 769 (84%) had a total cholesterol measurement, and 726 (79%) had total cholesterol and HDL-C measured. Fasting lipid values were available in 688 PTCA patients (75%), and LDL-C could be calculated in 642 (70%). LDL-C was thus available in 1263 patients overall, whereas non-HDL-C was available in 1438 patients.
Baseline characteristics of patients with and without an available baseline lipid measurement are shown in Table 1. Patients with lipid measurements were more likely to be white, were better educated, and were less likely to have diabetes, chronic obstructive pulmonary disease, or a history of congestive heart failure. The median number of coronary risk factors among patients with lipid measurements (as defined by the National Cholesterol Education Program) was 3 (range, −1 to 6).2 Patients with lipid measurements had better 5-year survival than those in whom cholesterol values were not measured at baseline (90% versus 80%; P<0.0001).
Lipid Levels at Baseline and Throughout the Study
Cholesterol and triglyceride values at baseline (ie, between weeks 4 and 14 after randomization) and years 1, 3, and 5 among patients randomized to PTCA and CABG are shown in Table 2. Among the 1514 patients who had baseline lipid determinations, 109 participants had only a baseline lipid profile, 253 patients had 2 lipid profiles measured during the trial, 584 patients had 3 measurements performed, and 568 patients had all 4 measurements performed. There were no significant differences between the groups at any of the time points except for HDL-C; patients with 3 or 4 lipid measures had lower baseline HDL-C than those with 1 or 2 measurements (0.98±0.26 mmol/L [38±10 mg/dL] versus 1.04±0.034 mmol/L [40±13 mg/dL]; P=0.016). The LDL-C and non-HDL-C values correspond approximately to the 50th to 60th percentile for men and women 60 years of age in the United States.14,15 At baseline, 9% of patients randomized to CABG and 16.1% of patients randomized to PTCA were on pharmacological lipid-lowering therapy. The proportion of treated patients increased over time to 40.5% and 45.5%, respectively. Despite increasing use of lipid-lowering agents over time, lipoprotein levels did not change significantly and remained well above currently recommended cut points throughout the study.2
Baseline Lipid Values and Outcome
Primary Outcome: Univariate Analyses
For every 0.26 mmol/L (10 mg/dL) increase in total cholesterol and non-HDL-C at baseline, there was a 5% increase in risk of subsequent nonfatal myocardial infarction and a 6% increase in angina at 5 years. Freedom from nonfatal myocardial infarction was highest in the lowest quintile of non-HDL-C (95%) and lowest in the highest quintile (88%). HDL-C, LDL-C, and triglycerides did not predict nonfatal myocardial infarction or angina. None of the baseline lipid measurements related to all-cause mortality, the combined end point of death or myocardial infarction, or revascularization during follow-up.
Nonfatal Myocardial Infarction: Multivariate Analyses
In previous BARI analyses, age, sex, race, chronic obstructive pulmonary disease, congestive heart failure, hypertension, history of prior myocardial infarction, smoking status, body surface area, number of significant coronary lesions, diabetes, and treatment assignment to CABG or PTCA were all found to be significant predictors of nonfatal myocardial infarction in this cohort and were thus considered important covariates for the current analysis. Table 3 shows the relative risks associated with a 10 mg/dL change in baseline lipid value when each lipid measurement was added individually to the BARI model for nonfatal myocardial infarction. Neither HDL-C nor LDL-C was a significant predictor of nonfatal myocardial infarction in this comprehensive model. Non-HDL-C was the strongest lipoprotein predictor, with a 4.9% increase in risk for each 0.26 mmol/L (10 mg/dL) increment, followed by total cholesterol (4.3% increase in risk) and triglycerides (1.6% increase in risk). The risk ratios for non-HDL-C and triglycerides did not change significantly when each was added to the BARI model together with HDL-C; both relative risks remained significant at 1.046 and 1.019, respectively.
Angina Pectoris at 5 Years: Multivariate Analyses
In previous BARI analyses, age, sex, race, history of hypertension, treatment assignment to PTCA or CABG, history of diabetes, history of peripheral vascular disease, and the patients’ subjective health assessment were all significant predictors of angina pectoris at 5 years. Table 4 shows the odds ratios associated with a 10 mg/dL change in baseline lipid value when each lipid measurement was added individually to this BARI model for angina pectoris. Only non-HDL-C was predictive of angina at 5 years, with a 4.9% increase in risk for each 0.26 mmol/L (10 mg/dL) increase in non-HDL-C. When both HDL-C and non-HDL-C were added to the previously validated model, the odds ratio for each 0.26 mmol/L (10 mg/dL) increase in non-HDL-C increased to 1.052 (95% confidence intervals, 1.003 to 1.091).
Diabetes and lipid-lowering therapy might affect the prognostic value of non-HDL-C. To explore this possibility, we added the respective interaction terms to the previously described models. None was significant. As detailed above, fewer patients had LDL-C measurements than non-HDL-C measurements, raising the possibility that the lack of predictive value of LDL-C was only due to smaller sample size. When we restricted the multivariate analyses to only those patients who had both LDL-C and non-HDL-C values available, our results were essentially the same, and the models are thus not reported separately.
Non-HDL-C Over Time and Outcome
Modeling of Lipid Burden Over Time
As shown in Table 5, non-HDL-C burden over time did not relate to death, the combined end point of death or myocardial infarction, or to revascularization in univariate or multivariate analyses, which used the average value to a given time point to interpolate missing data. Non-HDL-C burden during the study was, however, both a significant univariate and multivariate predictor of nonfatal myocardial infarction and angina pectoris at 5 years. For every 0.26-mmol/L (10-mg/dL) increment in average non-HDL-C, risk of nonfatal myocardial infarction increased by 5% and the odds of angina pectoris increased by 10%.
Among patients in the BARI cohort who had lipoprotein determinations performed at baseline, non-HDL-C was a strong and independent predictor of nonfatal myocardial infarction and angina pectoris during follow-up, even after adjustment for a multitude of other demographic and clinical characteristics known to influence outcome in this patient population. Non-HDL-C was a similarly strong predictor for nonfatal infarction and angina pectoris in our time-dependent analyses. In contrast, neither HDL-C nor LDL-C predicted outcome when added to the same previously validated models. Triglyceride values were predictive of nonfatal myocardial infarction but not angina pectoris.
To our knowledge, this study is the first to demonstrate a relationship between cardiovascular outcomes and non-HDL-C levels in a population of patients with prevalent coronary artery disease. Previously published studies have evaluated outcomes in populations free of cardiovascular disease at study entry or in mixed populations and demonstrated similarly strong and independent relationships between non-HDL-C and coronary heart disease incidence and, in some studies (primarily those with longer follow-up times), cardiovascular and all-cause mortality.4–11 We did not find any relationship between lipoprotein parameters and coronary heart disease death, which is possibly related to the shorter follow-up time and/or the high prevalence of other conditions strongly associated with fatal coronary heart disease (such as the severity of angiographic disease, left ventricular dysfunction, etc).
We did not find any association between cardiovascular outcomes and baseline HDL-C in our population. HDL-C levels were uniformly low in this BARI cohort (mean was below the 50th percentile for the US male population aged 55 to 64 years old).14 This lack of variation may not have allowed us to detect a relationship between HDL-C and cardiovascular end points. Our data are consistent with those of others, who also did not find a relationship between HDL-C and cardiovascular outcomes in middle aged and older cohorts without coronary heart disease at baseline.4,6,8
The role of triglycerides in atherosclerosis and clinical coronary heart disease remains controversial. We found that fasting triglycerides were independent predictors of nonfatal myocardial infarction in the BARI population. This finding further strengthens the argument for targeting non-HDL-C levels for treatment in coronary disease populations, because non-HDL-C encompasses not only LDL-C, intermediate density lipoprotein, and lipoprotein(a) (which are customarily estimated as LDL-C by the Friedewald formula), but also very low density lipoprotein cholesterol. Non-HDL-C as a variable for clinical risk stratification has several practical advantages as well. It can be calculated in the nonfasting state and can be calculated in the setting of hypertriglyceridemia, because the calculation makes no assumptions about composition of very low-density lipoprotein particles.
Our study has a number of limitations. Lipid levels were only available in a subset of participants, and many patients had only nonfasting data or fasting hypertriglyceridemia that precluded calculation of LDL-C. Our study may have lacked adequate power to assess the true predictive value of LDL-C compared with non-HDL-C, which was available in a much larger number of patients. It is also clear that patients in BARI who underwent venipuncture for lipoprotein determination were less sick than those in whom lipid measurements are not available. Lipids were not measured at a core laboratory, but rather at heterogeneous laboratories at the study sites with unknown quality control and different methodologies. However, we think that this heterogeneity with its inherent “measurement noise” would have biased our study toward the null rather than creating a spurious association between non-HDL-C and the outcome variables.
Among patients with lipid values in BARI, non-HDL-C is a strong and independent predictor of nonfatal myocardial infarction and angina pectoris at 5 years, even after consideration of powerful clinical variables. HDL-C and LDL-C did not relate to either outcome in this cohort. Our data suggest that non-HDL-C is an appropriate treatment target among patients with coronary heart disease.
This work was supported in part by the following grants from the National Heart, Lung, and Blood Institute: HL38493, HL38504, HL38509, HL 38512, HL38514-6, HL38518, HL38524, HL38529, HL38532, HL38556, HL38610, HL38642, and HL42145.
Presented in part at the 71st Scientific Sessions of the American Heart Association, Dallas, Tex, November 8–11, 1998, and published in abstract form (Circulation. 1998; 98(suppl I):I–412).
Dr Bittner has received research support and speaking honoraria from Merck and Pfizer.
Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma without use of the preparative ultracentrifuge. Clin Chem. 1972; 18: 499–502.
Keys A, Karvonen MJ, Pnsar S, et al. HDL serum cholesterol and 24-year mortality of men in Finland. Int J Epidemiol. 1984; 13: 428–435.
Pocock SJ, Shaper AG, Phillips AN, et al. High density lipoprotein cholesterol is not a major risk factor for ischaemic heart disease in British men. BMJ. 1986; 292: 515–519.
Frost PH, Davis BR, Burlando AJ, et al, for the Systolic Hypertension in the Elderly Research Group. Serum lipids and incidence of coronary heart disease. Circulation. 1996; 94: 2381–2388.
Lehto S, Roennemaa T, Haffner SM, et al. Dyslipidemia and hyperglycemia predict coronary heart disease events in middle-aged patients with NIDDM. Diabetes. 1997; 46: 1354–1359.
Protocol for the Bypass Angioplasty Revascularization Investigation. Circulation. 1991; 84 (suppl V): V1–V27.
Brooks MM, Jones RH, Bach RG, et al, for the BARI Investigators. Predictors of mortality and mortality from cardiac causes in the Bypass Angioplasty Revascularization Investigation (BARI) Randomized Trial and Registry. Circulation. 2000; 101: 2682–2689.
National Cholesterol Education Program. Second Report of the Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Washington, DC: US Department of Health and Human Services, Public Health Service, National Institutes of Health, NHLBI; 1993.NIH Publication No. 93-3095.