Although the Adult Treatment Panel III (ATP III) of the National Cholesterol Education Program recommends obtaining a full fasting lipoprotein profile, including total cholesterol, low-density-lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and triglyceride levels,1 the most significant measurement is considered to be LDL cholesterol, and the reduction of LDL cholesterol is the chief goal of treatment to reduce the risk of coronary artery disease (CAD).
It is recognized that increased LDL cholesterol levels lead to the development of atherosclerosis1; however, the use of LDL cholesterol levels to predict the risk of CAD is controversial.2 LDL particles can be heterogeneous in size and number,3 and smaller LDL particles may be more atherogenic.4 The use of another measurement has been proposed as a predictor of CAD risk—total apolipoprotein B100 (apoB100).5
Over the past decades, lipid measurements have been significantly improved and standardized, and the primary predictor in clinical guidelines may be reconsidered. Because they provide assessment of protective lipid fractions and atherogenic factors, better determination of risk may be made using such ratios as apoB100/apoA-I ratio, LDL cholesterol/HDL cholesterol, or total cholesterol/HDL cholesterol. For patients with increased triglyceride levels, non-HDL cholesterol levels were put forth by ATP III as an alternative predictor to LDL cholesterol1; in clinical practice, however, lipid indexes were not suggested as a favored measurement tool.
We assessed the usefulness of various lipid measurements in predicting CAD events in a nested case-control study among participants in the Nurses’ Health Study (NHS).
Patients and methods
Blood samples were taken from 32,826 female nurses enrolled in the NHS from 1989 to 1990. Incident fatal CAD and nonfatal myocardial infarction (MI) between the time of blood collection and May 31, 1998, was the end point. Physicians established the diagnosis of MI using criteria proposed by the World Health Organization.6 Determination of fatal MI was made by autopsy or hospital records. Subjects with a previous report of cancer or CAD before the blood collection were excluded. Each case was matched by age and smoking history to two controls who were free of CAD at the time of the case diagnosis. The lipid analyses were performed in a certified laboratory.
A total of 449 control subjects and 234 CAD cases (35 with fatal CAD and 199 with nonfatal CAD) participated in the study. Lower HDL cholesterol levels, higher levels of all other lipid measurements, higher body mass index (BMI), parental MI, hypertension, and diabetes occurred more commonly in the cases compared with the control subjects (data not shown).
Following are the relative risks (RRs) associated with an increase of approximately 1 standard deviation (SD; mg/dL) after adjustment for homocysteine, BMI, family history, hypertension, diabetes, C-reactive protein, postmenopausal hormone use, physical activity, alcohol intake, and blood draw parameters in multivariate models: triglycerides, RR = 1.3 (1.0—
1.5), SD = 80; LDL cholesterol, RR = 1.4 (1.1—1.7), SD = 36; apoB100, RR = 1.7 (1.4–2.1), SD = 32; HDL cholesterol, RR = 0.6 (0.5–0.8), SD = 171; and total cholesterol, RR = 1.4 (1.1–
1.6), SD = 40. The RRs for the lipid indexes were as follows: total cholesterol/HDL cholesterol, RR = 1.6 (1.3—
1.9), SD = 1.3; non-HDL cholesterol, RR = 1.6 (1.3—1.9), SD = 42 mg/dL; apoB100/HDL cholesterol, RR = 1.7 (1.4–2.1), SD = 1.0; and LDL cholesterol/HDL cholesterol, RR = 1.5 (1.3–
1.9), SD = 1.0.
In the Figure, A shows that HDL cholesterol was the principal contributor to the prediction model, increasing the receiver operating characteristic (ROC) to 0.63 from 0.51 in the stepwise model, which was adjusted for smoking status and age only. ApoB100 was the second contributor, raising the ROC to 0.66. The regression was performed again with a multivariable model, as shown in B of the Figure. HDL cholesterol was still the principal contributor to prediction of the model, increasing the ROC to 0.70 from 0.69. LDL cholesterol was brought into the second step, and the ROC increased to 0.72. In the Figure, D shows that additional information was not provided by triglyceride or apoB100 levels over that provided by LDL and HDL cholesterol levels. Also in the Figure, C shows that apoB100 levels did not provide additional information for a model in which we forced the other clinical parameters, whereas HDL levels did. Significant additional information was provided to the multivariate intercept by all of the lipid indexes. The ROC was increased to 0.72 from 0.69 by the ratios containing HDL cholesterol: apoB100/HDL cholesterol, LDL cholesterol/HDL cholesterol, and total cholesterol/HDL cholesterol. E through G in the Figure show that there were no other lipid biomarkers that contributed significantly; H shows that the ROC was increased to 0.71 by non-HDL cholesterol, and the model was additionally improved when HDL cholesterol was added.
Over a period of 8 years, taking into consideration a number of lipid and nonlipid risk factors for CAD, we assessed the usefulness of a number of plasma lipid parameters in predicting future CAD in women. The principal lipid predictor for postmenopausal women was HDL cholesterol, using a multivariate model that was adjusted for various lipids. The total cholesterol/HDL cholesterol ratio is a single measurement that can predict CAD risk apart from other proven risk factors, supplying a strong predictive model that can be used in clinical practice.
We showed that HDL cholesterol rather than LDL cholesterol was the primary discriminator among the women in the study, but both values are needed to predict CAD in a multivariate model. Although HDL cholesterol level is a defined risk factor in the ATP III algorithm for primary prevention,1 with an HDL cholesterol below 40 mg/dL counted as a risk factor and an HDL cholesterol above or equal to 60 mg/dL counted as a protective factor, the superiority of HDL cholesterol level as a major predictor among women is not well established. The reasons might be historical, driven by unstandardized HDL cholesterol measurements in the past and the lack of established HDL-raising drugs.
ApoB100 was the lipid fraction that was most significantly related to CAD risk. The liver synthesizes apoB100, which is secreted with very-low-density lipoprotein (VLDL). Because there is one apoB100 molecule per lipoprotein particle, total apoB100 reflects the total particle numbers in LDL, VLDL, and intermediate-density lipoprotein (IDL).7 Because the half-life of LDL cholesterol particles is nine times the half-life of IDL and VLDL particles,8 the number of LDL particles can be approximated by the apoB100 concentration. Our results are consistent with previous epidemiologic studies,9,10 and the Canadian Cardiovascular Society has targeted levels of apoB100 below 90 mg/dL for patients at high risk for CAD.11 Although apoB100 remained the superior lipid-biomarker predictor in a multivariate model, its association was appreciably attenuated by other CAD risk factors.
For patients with triglyceride levels above or equal to 200 mg/dL, a secondary aim of therapy is treatment of non-HDL cholesterol, which includes cholesterol in triglyceride-rich and LDL-rich lipoproteins.1 It has been suggested that non-HDL cholesterol is a less expensive and more practical par-ameter for CAD risk assessment that is already available from the lipid panel.12 However, among lipid indexes, non-HDL cholesterol provided the exclusive unsaturated model, for which HDL cholesterol was needed to provide additional information (H in the Figure). These measurements reflect different biological elements, in spite of the fact that there is a high association between non-HDL cholesterol and apoB100. Animal models have shown that the atherogenicity of cholesterol
in VLDL is weaker than that of cholesterol in LDL.13 Non-HDL cholesterol, therefore, may not be a good replacement for apoB100.
All the HDL-cholesterol—related ratios, which reflect the proportion of atherogenic to antiatherogenic lipid fractions, showed superior predictive value. Ratios of lipids or apolipoproteins, or both, have been better predictors of CAD risk than have levels of any lipid fraction.14 By taking into account the triglyceride-rich lipoproteins, the total cholesterol/HDL cholesterol ratio appears to be a better predictor15 of CAD than LDL cholesterol/HDL cholesterol. In our study, the apoB100/HDL cholesterol ratio showed the strongest and most linear association with MI. However, clinical usefulness of this ratio must be judged against the cost of its measurement, the present lack of a direct therapeutic option, and the very small gain in overall predictive value compared with that of the available total cholesterol/HDL cholesterol ratio.
Because of the powerful association of low HDL cholesterol levels with CAD among women, HDL-cholesterol—related ratios, such as the total cholesterol/HDL ratio, should be considered as the primary target tool in measures to prevent CAD.