Comparison of Apolipoprotein B and Non-High-Density Lipoprotein Cholesterol
Comparison of Apolipoprotein B and Non-High-Density Lipoprotein Cholesterol
Whether non--high-density lipoprotein cholesterol is equivalent to apolipoprotein B (apo B) for screening remains controversial. One reason for continued controversy is that most studies express results as relative risk/hazard or odds ratios based on P values that reflect diagnostic values poorly. Apo B and lipoprotein lipids were compared in 437 men. The results were evaluated by multivariate techniques and by receiver operating characteristic (ROC) curves. When analyzed by ROC curves, the difference between apo B and lipoprotein lipids proved to be less than would be anticipated from the odds ratios. Although, after adjustment, the difference was about 14% by odds ratios, ROC analysis showed only a small difference of about 1%. These data show that clinical studies should analyze the data using an absolute measure of risk such as ROC curves rather than just relative indexes. Such a small absolute difference may also explain discrepancies between studies.
Evidence indicates that apolipoprotein B-100 (apo B) is a better predictor of coronary risk than total cholesterol or low-density lipoprotein cholesterol (LDLC). This would be expected because a single apo B molecule is found in each molecule of all b-lipoproteins, making it a better measure of the number of potentially atherogenic particles, especially small, dense LDL.
Serum nonhigh-density lipoprotein cholesterol (non-HDLC) is better correlated with apo B than is LDLC. Non-HDLC has been proposed as a surrogate for apo B. In fact, on the basis of relative hazard ratios, a recent article concluded that non-HDLC was as good or a better predictor of future cardiovascular events as apo B in women, although another study, on the basis of P values, found that apo B was superior in both men and women. And, based on relative odds and risk ratios, other recent studies concluded that both were strong predictors of coronary heart disease but that apo B was the better marker in men. Nevertheless, it is difficult to assess possible clinical usefulness of one test vs the other from these studies because relative risk, hazard and odds ratios, or other correlations based on statistically significant P values do not translate into diagnostic sensitivity and specificity values that are required for defining clinically useful discrimination.
In the present study, the ability of apo B to discriminate between disease and nondisease was compared with non-HDLC and other lipoprotein lipids in a group of 437 men who had undergone coronary angiography and were classified as having severe coronary artery disease (CAD), intermediate CAD, or no appreciable disease (normal). The ability of each test to discriminate was evaluated by 3 statistical techniques. The first technique was stepwise discriminant analysis to determine how apo B compared with lipoprotein lipids across the entire group. Second, multivariate logistic regression was used to determine whether, when compared with data for subjects having no disease, data for subjects with only severe disease gave rise to the same conclusions as for the entire group. Logistic regression also provides a way for the present study to be compared with other studies, most of which used logistic techniques for analysis. Finally, these data were analyzed by receiver operating characteristic (ROC) curves that provide a means for definitive comparisons for clinical purposes, before and after multivariate adjustment for nonlipoprotein risk factors.
The results of discriminant analysis and logistic regression agreed well with one another and with prior studies in men, indicating apo B was a superior marker to lipoprotein lipids. Nevertheless, the ROC curve analysis showed that on average, there was not a great difference between apo B and non-HDLC and that the difference was mitigated after adjustment for traditional nonlipoprotein risk factors. This example illustrates why studies presuming to assess risk for diagnostic purposes should express the data in absolute form as well as relative indexes.
Abstract and Introduction
Abstract
Whether non--high-density lipoprotein cholesterol is equivalent to apolipoprotein B (apo B) for screening remains controversial. One reason for continued controversy is that most studies express results as relative risk/hazard or odds ratios based on P values that reflect diagnostic values poorly. Apo B and lipoprotein lipids were compared in 437 men. The results were evaluated by multivariate techniques and by receiver operating characteristic (ROC) curves. When analyzed by ROC curves, the difference between apo B and lipoprotein lipids proved to be less than would be anticipated from the odds ratios. Although, after adjustment, the difference was about 14% by odds ratios, ROC analysis showed only a small difference of about 1%. These data show that clinical studies should analyze the data using an absolute measure of risk such as ROC curves rather than just relative indexes. Such a small absolute difference may also explain discrepancies between studies.
Introduction
Evidence indicates that apolipoprotein B-100 (apo B) is a better predictor of coronary risk than total cholesterol or low-density lipoprotein cholesterol (LDLC). This would be expected because a single apo B molecule is found in each molecule of all b-lipoproteins, making it a better measure of the number of potentially atherogenic particles, especially small, dense LDL.
Serum nonhigh-density lipoprotein cholesterol (non-HDLC) is better correlated with apo B than is LDLC. Non-HDLC has been proposed as a surrogate for apo B. In fact, on the basis of relative hazard ratios, a recent article concluded that non-HDLC was as good or a better predictor of future cardiovascular events as apo B in women, although another study, on the basis of P values, found that apo B was superior in both men and women. And, based on relative odds and risk ratios, other recent studies concluded that both were strong predictors of coronary heart disease but that apo B was the better marker in men. Nevertheless, it is difficult to assess possible clinical usefulness of one test vs the other from these studies because relative risk, hazard and odds ratios, or other correlations based on statistically significant P values do not translate into diagnostic sensitivity and specificity values that are required for defining clinically useful discrimination.
In the present study, the ability of apo B to discriminate between disease and nondisease was compared with non-HDLC and other lipoprotein lipids in a group of 437 men who had undergone coronary angiography and were classified as having severe coronary artery disease (CAD), intermediate CAD, or no appreciable disease (normal). The ability of each test to discriminate was evaluated by 3 statistical techniques. The first technique was stepwise discriminant analysis to determine how apo B compared with lipoprotein lipids across the entire group. Second, multivariate logistic regression was used to determine whether, when compared with data for subjects having no disease, data for subjects with only severe disease gave rise to the same conclusions as for the entire group. Logistic regression also provides a way for the present study to be compared with other studies, most of which used logistic techniques for analysis. Finally, these data were analyzed by receiver operating characteristic (ROC) curves that provide a means for definitive comparisons for clinical purposes, before and after multivariate adjustment for nonlipoprotein risk factors.
The results of discriminant analysis and logistic regression agreed well with one another and with prior studies in men, indicating apo B was a superior marker to lipoprotein lipids. Nevertheless, the ROC curve analysis showed that on average, there was not a great difference between apo B and non-HDLC and that the difference was mitigated after adjustment for traditional nonlipoprotein risk factors. This example illustrates why studies presuming to assess risk for diagnostic purposes should express the data in absolute form as well as relative indexes.
Source...