- Email: [email protected]
When is equal not equal?
Journal of Clinical Lipidology (2010) 4, 83–88
Original Articles
When is equal not equal? Allan D. Sniderman, MD*, Ken Williams, PhD, Matthew J. McQueen, MD, PhD, Curt D. Furberg, MD, PhD Mike Rosenbloom Laboratory for Cardiovascular Research, McGill University Health Centre, Room H7.22, Royal Victoria Hospital, 687 Pine Avenue West, Montreal, Quebec H3A 1A1, Canada (Dr. Sniderman); Department of Medicine, KenAnCo Biostatistics, University of Texas Health Science Center, San Antonio, TX, USA (Dr. Williams); Population Health Research Institute, McMaster University, Hamilton, ON, Canada (Dr. McQueen); and Division of Public Health Sciences, Wake Forest University School of Medicine, Winston–Salem, NC, USA (Dr. Furberg) KEYWORDS: ApoB; Non-HDL C; Hazard ratio; Meta-analysis; Cardiovascular risk
Abstract: The meta-analysis of the Emerging Risk Factor Collaboration demonstrated that the hazard ratios (HR) of the major cholesterol markers and the major apolipoproteins for vascular disease did not differ significantly in the studies they examined. Their conclusion was that they were functionally interchangeable. We believe there are important limitations in the execution of this study. Nevertheless, even if their findings are correct for groups, their conclusions do not follow for individuals. Conventionally, the HR expresses the increase in risk per standard deviation change for that parameter in a group. However, the predicted risk of vascular disease from an atherogenic parameter depends on its concentration within the individual. Depending on the composition of the apoB lipoproteins, individuals may have either concordant or discordant levels of cholesterol and apoB. For those who are concordant, the two markers predict equal risk. For those who are discordant, the predicted risks for the individual are different. We demonstrate that substantial discordance in the individual HR of non-high-density lipoprotein cholesterol and apoB is common. The result is that even with identical overall HR, apoB points to higher risk in a substantial number of individuals whereas the converse is the case for non- high-density lipoprotein cholesterol. Because we are concerned with risks in individuals, not groups, this discordance is important to appreciate and analyze. Our objective should be to learn how to combine the information from parameters rather than eliminate them and we need to focus on evaluation of risk in individuals and not just groups. Ó 2010 National Lipid Association. All rights reserved.
The Emerging Risk Factor Collaboration (ERFC) has reported that the hazard ratios (HRs) of low-density lipoprotein cholesterol (LDL-C), nonhigh-density lipoprotein cholesterol (HDL-C), and apoB for the risk of vascular disease are virtually identical.1 Moreover, the HR of the apoB/apoAI ratio was the same as the total cholesterol (TC)/HDL-C. The primary conclusion of this large meta-analysis was that what * Corresponding author. E-mail address: [email protected] Submitted December 18, 2009. Accepted for publication January 24, 2010.
we measure in clinical practice should be on the basis of factors other than predictive accuracy of the apolipoproteins versus the lipoprotein lipids. Because measurement of apolipoproteins represents additional costs, many might, understandably, conclude that these additional costs were not associated with additional benefits. The ERFC study is a valuable contribution to the challenge of improving our capacity to quantitate the atherogenic risk attributable to the plasma lipoproteins. To be sure, as with almost any study, there are limitations. Thus, the observation that directly measured LDL-C was as accurate as non-HDL-C at identifying risk is at variance with the
1933-2874/$ -see front matter Ó 2010 National Lipid Association. All rights reserved. doi:10.1016/j.jacl.2010.01.005
84 results of most recent studies2 and suggests the power of the ERFC study to discriminate amongst the markers may not be ideal. Moreover, the methodology to measure lipids and apolipoproteins was highly variable among the studies, many of which were performed early in the course of measuring apolipoproteins. Inaccuracies in methodology cannot be overcome simply by increasing the number of studies and may be the basis for the heterogeneity in the results noted by the authors. Finally, we do not agree with their reasoning that excluded the INTERHEART3 and the International Study of Infarct Surviva (ISIS)4 studies. On the contrary, we believe the case-control design is particularly suited to testing the relative value of closely correlated variables. Nevertheless, our primary focus is not on any limitations of this study. Rather, we will try to demonstrate that even if the primary findings are correct—namely that the HRs for the apolipoproteins and lipids are equal for the overall groups that were analyzed—this does not mean that lipids and apolipoproteins are clinically interchangeable in all individuals. After all, it is individuals, not groups, in whom risk is estimated and treated and therefore, it is the utility of risk markers within individuals that counts. Furthermore, the variance that is responsible for the discordance between the apolipoprotein and cholesterol risk markers in many individuals is due to differences in the composition of the major apoB lipoproteins—very lowdensity lipoprotein (VLDL) and LDL— and therefore, discordance among the risk markers is embedded in the pathophysiological mechanisms responsible for changes in the number as well as the composition of the apoB lipoproteins. The atherogenic risk to the individual from the apoB lipoproteins is driven by these differences in particle number and composition. Accordingly, if our objective is the best care for each patient, the challenge is to learn how to combine the information from cholesterol and apolipoproteins to enlarge our diagnostic accuracy rather than to continue only to pit one against the other. That is, our objective should be to add information not eliminate it.
Journal of Clinical Lipidology, Vol 4, No 2, April 2010 Plasma apoB, therefore, measures total atherogenic particle number. Except in very unusual circumstances, LDL particles account for the vast majority of the apoB particles.5 Non-HDL-C is the sum of the cholesterol in the apoB lipoproteins. Again, LDL particles account for most of the non-HDL-C, although the proportion is more variable than in the case of apoB.5
Population HRs versus individual HRs As is the usual practice, ERFC reports HR for a parameter as the increase in risk per standard deviation of that parameter within the study population. By normalizing risk over the range of values in the population, the overall risk for different parameters can be compared within groups. We will refer to this HR as the population HR (PHR). As noted previously, the PHR for apoB and nonHDL C were both equal in the ERFC study. However, the risk for the actual level of any parameter for an individual—the individual HR (IHR)—depends on the actual concentration of the atherogenic parameter in that individual, that is, the greater the concentration, the greater the risk. Thus, Figure 1 depicts the relation between IHR of non-HDL C/apoB on the basis of their concentrations expressed as a percentile of the American adult population. Because both parameters have the same HR, their overall relation to risk will be nearly identical. However, the relation between risk and concentration is curvilinear, demonstrating that absolute risk from any factor depends on the absolute level of that factor and that risk rises exponentially while concentration increases linearly. Any group is made up of individuals, each of which has a specific concentration of non-HDL-C and apoB. Although non-HDL-C and apoB are highly correlated, they are not
Non-HDL-C and apoB This analysis will focus on non-HDL-C and apoB, which have emerged as the two most powerful markers of the vascular risk attributable to the atherogenic plasma lipoproteins. Non-HDL-C and apoB are highly correlated variables, and on the basis of this property, it has been argued that they are clinically interchangeable. The equivalent HRs of the ERFC study would appear to support that contention. We will test whether that conclusion is correct when individuals are considered. Non-HDL-C and apoB measure different things. NonHDL-C is the sum of the masses of cholesterol within the atherogenic lipoproteins— or the purposes of this discussion, the mass of cholesterol within VLDL and LDL particles—whereas apoB equals the number of atherogenic particles, that is, the number of VLDL and LDL particles.
Figure 1 ‘‘Effectiveness curves’’: PHRs adjusted for age, systolic blood pressure, and smoking from the ERFC by percentile by use of the 5th percentile as the reference group. Estimates pertain to 190,977,311 nonpregnant, noninstitutionalized, civilian US adult residents .20 years of age represented by 1700 participants in the NHANES 2005 to 2006 study with both lipid measures. The published HRs were 1.63 per SD of 43 mg/dL of non-HDL-C and 1.62 per SD of 29 mg/dL of apoB.
Sniderman et al
When is equal not equal?
perfectly concordant, that is, for any given value of one variable, there is a range of values for the other. The precise degree to which they are interchangeable for an individual depends on how concordant they are. Figure 2 demonstrates the relation between non-HDL-C and apoB in the National Health and Nutrition Examination Survey (NHANES) 2005 to 2006 survey of the American adult population.6 There are three points to note. First, the overall correlation coefficient is very high: 0.95. Second, at the extremes, concordance is also very high. That is, when either non-HDL-C or apoB is very high or very low, the other marker is also very high or very low. Therefore, one marker conveys as much information about risk as the other. Moreover, because the total risk attributable to any parameter is determined to a major degree by the risk between the extreme values, this means that the overall risk attributable to either nonHDL-C or apoB must be similar. Third, from approximately the 10th to the 90th percentile, there is a definite range of values of one parameter versus the other as reflected in a fall in the overall correlation between the two parameters from 0.95 to 0.88. This is also evident from the scatter of the values around the line of identity (Fig. 3). The IHR for non-HDL C and apoB, that is, the HR for each on the basis of their actual values within an individual, can be calculated for all of the individuals between the 10th and 90th percentiles of the distribution and the results are demonstrated in Figure 4. Substantial discordance in the risk attributable to non-HDL-C versus the risk attributable to apoB is evident throughout this wide range of values for the population. In Table 1 we compare groups of individuals defining as discordant those individuals whose apoB and non-HDL-C percentiles are separated by one or more deciles. These comparisons yield several important points. More than 17% of US adults have values of these two different biological measures that are discordant by this definition. An estimated 8.8% have an individual apoB percentile value more than 10 percentiles points greater than their nonHDL–C value. Furthermore overall and among men
Figure 2 Scatter plot of percentile of apoB versus percentile of Non-HDL-C in NHANES 2005 to 2006.
85
Figure 3 Scatter plot of percentile of apoB versus percentile of non-HDL-C between the 10th and 90th non-HDL-C percentiles in NHANES 2005 to 2006.
10-year CHD risk, as assessed by Framingham equations,7,8 is significantly greater on average in the apoB discordant group than in the non-HDL-C discordant group.
Clinical interpretations of discordance in attributable-risk of cardiovascular disease between non-HDL-C and apoB within an individual When the IHR of non-HDL-C and apoB are the same or very close within the individual, they obviously convey
Figure 4 Scatter plot of individual apoB hazard ratio versus individual non-HDL-C HR in NHANES 2005 to 2006 subjects with non-HDL-C values between the 10th and 90th percentile.
86
Journal of Clinical Lipidology, Vol 4, No 2, April 2010
Table 1 Comparisons by concordance/discordance category in NHANES 2005 to 2006 representative of 190,877,311 nonpregnant, noninstitutionalized, civilian US adult residents $20 years of age
All Concordant Non-HDL-C %ile $ apoB %ile 1 10 ApoB %ile $ Non-HDL-C %ile 110 Men Concordant Non-HDL C %ile $ apoB %ile 1 10 ApoB %ile $ Non-HDL-C %ile 110 Women Concordant Non-HDL C %ile $ apoB %ile 1 10 ApoB %ile $ non-HDL-C %ile 110
Sample, n
People represented, n
Percent 6 SE, %
Risk, %*
970 126 175
110,845,840 16,532,859 16,836,517
58.1 6 2.5 8.7 6 2.5 8.8 6 1.5
3.2 2.7 5.3
.19 .002†
507 59 104
55,156,771 6,454,543 8,946,989
59.8 6 2.7 7.0 6 2.0 9.7 6 1.9
5.7 4.9 9.0
0.22 .006†
463 67 71
55,689,068 10,078,316 7,889,528
56.5 6 3.0 10.2 6 3.0 8.0 6 1.4
1.8 1.8 2.8
.94 .08
P vs row above
Individuals in the bottom and top deciles of either apoB or non-HDL-C (n 5 429 representing 24.4% of the population) were excluded. Individuals whose apoB and non-HDL C percentile (%ile) values differed by one decile or more were defined as discordant. All statistics were estimated using SAS 9.2 SURVEY procedures to account for the NHANES complex sample design. *Back-transformed mean of logit-transformed Framingham 10-year primary7 or secondary8 risk or coronary heart disease. †These values indicate statistical significance.
very similar information as to risk and pose no challenge as to interpretation. However, when they are substantially different within the individual, interpretation is more demanding. In many cases, however, we believe the discordance can be explained on the basis of wellrecognized differences in the metabolism of the major apoB lipoproteins—VLDL and LDL—that produce the differences in their composition that lead to the different IHRs. In these situations, the two measures contribute additive, not redundant, information. Two discordant IHR phenotypes need to be considered: (1) the apoB percentile . non-HDL-C percentile and (2) the non-HDL-C percentile . the apoB percentile.
apoB percentile. non-HDL C percentile This indicates that relative to their respective distributions in the population the total number of apoB particles is greater than the mass of cholesterol within them. Because LDL particles comprise the great majority of total apoB articles and because LDL-C makes up the great majority of non-HDL-C, this points to a relative excess of LDL particles compared with LDL-C. This specific pattern of discordance in the concentrations of LDL-C and LDL particle number is caused by an excess of small dense, cholesterol-depleted particles.5 This results in an apoB that is disproportionately high to LDL-C and to non-HDL-C in this instance as well. Many have argued that small dense LDL are more atherogenic than larger, cholesterol-enriched LDL and multiple in vitro studies2 have identified specific mechanisms that could explain this increased risk. These range from smaller size and therefore greater ease of entry into the arterial wall to increased binding affinity to the arterial glycosaminoglycans to greater risk of oxidation.5 Whether
the difference is clinically significant is not certain. Accordingly, we believe the secure conclusion for the moment is that there is no credible body of evidence to indicate that small dense LDL particles are associated with a sharply diminished risk on a per-particle basis than larger LDL particles and therefore, there is no justification that they should be ignored. It follows that in patients with disproportionately high apoB, a prudent physician would assess risk based on the apoB not on the non-HDL-C even assuming their PHR are identical. The Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER) study is an apposite example of how extreme the differences in non-HDL-C and apoB can be and how this can affect the clinical interpretation of events.9 On the basis of low levels of LDL C and non-HDL-C (approximately the 25th and 30th percentiles of the American population respectively),10 the risk from atherogenic apoB lipoproteins would be considered to be low. However, the level of apoB was considerably greater, approximately the 65th percentile of the American population.11 Given the level of apoB and the age of the subjects (average age 66 years) and therefore given the duration of the exposure to moderately elevated levels of atherogenic lipoprotein particles, the risk of vascular disease in the control group and the benefit from potent statin therapy should not be surprising.
Non-HDL C . apoB There are two pathophysiological explanations for the level of non-HDL-C to be greater than the concurrent apoB. The first are remnant VLDL and remnant chylomicron particles and the second, cholesterol-enriched LDL particles. Patients with familial dysbetalipoproteinemia (FDB) represent the extreme case of excessive accumulation of
Sniderman et al
When is equal not equal?
cholesterol-enriched remnant apoB48 and apoB100 lipoprotein particles, a syndrome that is unquestionably associated with an extreme elevation of the risk of vascular disease. FDB is characterized by high levels of total and non-HDL-C, high levels of plasma triglycerides, and normal levels of plasma apoB.12 Remnants are large cholesterol-rich, triglyceride-rich particles, which explains the hypercholesterolemia and hypertriglyceridemia. Failure to convert VLDL particles to LDL particles is another of the metabolic hallmarks of FDB. LDL apoB, accordingly, accounts, on average, for only half the total apoB particles in sharp contrast to the usual circumstance in which LDL particles make up 90% or more of the total apoB particles. Because the half-life of LDL particles is much longer than remnant particles, total plasma apoB is normal in these patients. Accordingly, the IHR for non-HDL C more accurately identifies risk than the IHR for apoB in those with FDB. Interpretation in these patients is not difficult because the overall pattern of marked hypertriglyceridemia and marked hypercholesterolemia with normal apoB plus the diagnostic ratios of TG/apoB and TC/apoB allow for specific diagnosis.13 Even so, FDB, as such, will account for only a small minority of circumstances in which nonHDL C is disproportionately greater than apoB. Remnants can be present in smaller amounts and could pose significant risk. At the present time, therefore, in hypertriglyceridemic patients in whom non-HDL C is greater than apoB, the prudent physician could base assessment of risk on nonHDL C rather than apoB. Cholesterol-enriched LDL particles constitute the alternative explanation for a level of non-HDL-C that is disproportionately elevated compared with apoB. Familial hypercholesterolemia represents a classical example of cholesterol-enriched LDL particles, but, in this instance, the discordance is of little significance because the values of LDL-C, non-HDL-C, and apoB are all extremely elevated. However, cholesterol-enriched LDL particles can occur with levels of apoB well within the normal distribution of values. The pathophysiological mechanism leading to cholesterol-enrichment of LDL particles has not been elucidated. Nevertheless, the clinical issue is whether the greater IHR from the non-HDL C based on cholesterolenriched LDL C is correct compared with the lower IHR based on the apoB. Unfortunately, the question cannot be answered definitively at the moment. However, the data from the Framingham Offspring Study demonstrate that risk in those with low LDL-C and low LDL particles (and therefore low apoB) is low and indistinguishable from those with high LDL-C and low LDL particles.14 That is, the risk of cardiovascular event was determined by the number of LDL particles, not by the mass of cholesterol within them. This suggests that in subjects in whom non-HDL-C is greater than apoB as the result of cholesterol-enriched LDL, apoB may be a more reliable marker of risk than nonHDL-C. Further work is necessary to test this conclusion.
87 Nevertheless, whether the target is LDL C, non-HDL-C or apoB, the target levels in terms of the population values should be equivalent. On the basis of the NHANES 2005 to 2006 survey, a non-HDL C of 125 mg/dL and an apoB of 90 mg/dL would be equivalent to an LDL-C of 100 mg/dL—the high-risk targets at the 40th percentile— whereas a non-HDL-C of 90 mg/dL and an apoB of 65 mg/dL would be equivalent to an LDL C of 70 mg/dL— the very high risk values at the 10th percentile.6
Summary Our objective has been to demonstrate that findings in groups do not necessarily apply to individuals. We do not devalue these overall findings from studies such as ERFC and the general relationships they establish. They provide the first irreplaceable level of evidence. However, they should not end enquiry because as physicians our concern is the outcome of individuals not of groups. Thus, even if the overall findings of the ERFC are correct, and we have reservations on that point as noted at the outset, the actual hazard that a factor poses to an individual depends on the actual level of that factor within that individual, not on the average hazard of the factor for the group. The distinction is not dissimilar to that between the average temperature in a locale versus the actual temperature on a particular day in a particular season. Acknowledging variance means acknowledging individuals. Accordingly, one lesson that should be drawn from the ERFC study is that the conventional statistical methods may not be as sensitive as desirable to compare closely correlated risk factors. An equally important one is that clinical decisions should be based on risk in the individual not in the group and therefore the findings of the ERFC study should not be blindly transferred to clinical practice. The challenge, we believe, is not to pit one number against another but to integrate the information from both. As noted at the outset, apoB and non-HDL C measure different things. ApoB accurately measures atherogenic particle number and is a more accurate guide to LDL particle number than cholesterol. Therefore, when apoB is disproportionately high, apoB should guide management. By contrast, apoB is insensitive to remnant particles whereas plasma lipids, in particular cholesterol and triglyceride, do reflect the impact of this abnormality. Accordingly, when non-HDL-C is disproportionately high, particularly in hypertriglyceridemic patients, non-HDL-C should guide management. Our overall conclusion, therefore, is that apolipoproteins and lipoprotein lipids should be used in a complementary rather than a mutually exclusive fashion and that we should move to amplify the diagnostic information in an individual rather than diminish it. We acknowledge that not all countries have the medical infrastructure and resources to apply this approach immediately. That does not exclude it as an objective. Moreover, in some settings, apolipoproteins may be easier and simpler
88 to incorporate in clinical practice just as cell phones were easier and simpler to adopt compared with land lines. Our point is that we can make the best choices when we have the most knowledge.
Journal of Clinical Lipidology, Vol 4, No 2, April 2010
7.
8.
References 9. 1. Danesh J. The Emerging Risk Factors Collaboration. Major lipids, apolipoproteins, and risk of vascular disease. J Am Med Assoc. 2009;302: 1993–2000. 2. Sniderman AD, de Groot E, Couture P. ApoB and the atherogenic ApoB dyslipoproteinemias. In: Kwiterovich PO Jr., editor. The Johns Hopkins Textbook of Dylipidemia. Philadelphia, PA: Lippincott Williams & Wilkins, 2009. p. 196–210. 3. McQueen MJ, Hawken S, Wang X, et al. The relative importance of lipids, lipoproteins, and apolipoproteins as risk markers associated with myocardial infarction in 52 countries. Lancet. 2008;372:224–233. 4. Parish S, Peto R, Palmer A, et al. The joint effects of apolipoprotein B, apolipoprotein A1, LDL cholesterol, and HDL cholesterol on risk: 3510 cases of acute myocardial infarction and 9805 controls. Eur Heart J. 2009;30:2137–2146. 5. Barter PJ, Ballantyne CM, Carmena R, et al. ApoB versus cholesterol in estimating cardiovascular risk and in guiding therapy: report of the thirty-person/ten-country panel. J Intern Med. 2006;259:247–258. 6. NHANES Investigators. NHANES Analytic Guidelines 2003–2004. Available at: http://www.cdc.gov/nchs/data/nhanes/nhanes_03_04/
10.
11.
12.
13.
14.
nhanes_analytic_guidelines_dec_2005.pdf. Accessed February 16, 2009. Wilson PWF, D’Agostino RB, Levy D, Belanger AM, Halit S, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation. 1998;97:1837–1847. D’Agostino RB, Russell MW, Huse DM, Ellison RC, Silbershatz H, Wilson PW, et al. Primary and subsequent coronary risk appraisal: new results from the Framingham study. Am Heart J. 2000;139:272–281. Ridker PM, Danielson E, Fonseca FA, et al. Rosuvastatin to prevent vascular events in men and women with elevated C-reactive protein. N Engl J Med. 2008;359:2195–2207. Sniderman A, Williams K, Cobbaert C. ApoB versus non-HDLC: What to do when they disagree. Curr Atheroscler Rep. 2009;11: 358–363. Ridker PM, Danielson E, Fonseca FA, et al. Reduction in C-reactive protein and LDL cholesterol and cardiovascular event rates after initiation of rosuvastatin: a prospective study of the JUPITER trial. Lancet. 2009;373:1175–1182. Sniderman AD, Tremblay A, Bergeron J, Gagne C, Couture P. Diagnosis of type III hyperlipoproteinemia from plasma total cholesterol, triglyceride, and apolipoprotein B. J Clin Lipidol. 2007;1:256–263. de Graaf J, Couture P, Sniderman A. A diagnostic algorithm for the atherogenic apolipoprotein B dyslipoproteinemias. Nat Clin Endocrinol Metab. 2008;4:608–618. Cromwell WC, Otvos JD, Keyes MJ, et al. LDL particle number and risk of future cardiovascular disease in the Framingham Offspring Study—Implications for LDL management. J Clin Lipidol. 2007;1: 583–592.
Original Articles
When is equal not equal? Allan D. Sniderman, MD*, Ken Williams, PhD, Matthew J. McQueen, MD, PhD, Curt D. Furberg, MD, PhD Mike Rosenbloom Laboratory for Cardiovascular Research, McGill University Health Centre, Room H7.22, Royal Victoria Hospital, 687 Pine Avenue West, Montreal, Quebec H3A 1A1, Canada (Dr. Sniderman); Department of Medicine, KenAnCo Biostatistics, University of Texas Health Science Center, San Antonio, TX, USA (Dr. Williams); Population Health Research Institute, McMaster University, Hamilton, ON, Canada (Dr. McQueen); and Division of Public Health Sciences, Wake Forest University School of Medicine, Winston–Salem, NC, USA (Dr. Furberg) KEYWORDS: ApoB; Non-HDL C; Hazard ratio; Meta-analysis; Cardiovascular risk
Abstract: The meta-analysis of the Emerging Risk Factor Collaboration demonstrated that the hazard ratios (HR) of the major cholesterol markers and the major apolipoproteins for vascular disease did not differ significantly in the studies they examined. Their conclusion was that they were functionally interchangeable. We believe there are important limitations in the execution of this study. Nevertheless, even if their findings are correct for groups, their conclusions do not follow for individuals. Conventionally, the HR expresses the increase in risk per standard deviation change for that parameter in a group. However, the predicted risk of vascular disease from an atherogenic parameter depends on its concentration within the individual. Depending on the composition of the apoB lipoproteins, individuals may have either concordant or discordant levels of cholesterol and apoB. For those who are concordant, the two markers predict equal risk. For those who are discordant, the predicted risks for the individual are different. We demonstrate that substantial discordance in the individual HR of non-high-density lipoprotein cholesterol and apoB is common. The result is that even with identical overall HR, apoB points to higher risk in a substantial number of individuals whereas the converse is the case for non- high-density lipoprotein cholesterol. Because we are concerned with risks in individuals, not groups, this discordance is important to appreciate and analyze. Our objective should be to learn how to combine the information from parameters rather than eliminate them and we need to focus on evaluation of risk in individuals and not just groups. Ó 2010 National Lipid Association. All rights reserved.
The Emerging Risk Factor Collaboration (ERFC) has reported that the hazard ratios (HRs) of low-density lipoprotein cholesterol (LDL-C), nonhigh-density lipoprotein cholesterol (HDL-C), and apoB for the risk of vascular disease are virtually identical.1 Moreover, the HR of the apoB/apoAI ratio was the same as the total cholesterol (TC)/HDL-C. The primary conclusion of this large meta-analysis was that what * Corresponding author. E-mail address: [email protected] Submitted December 18, 2009. Accepted for publication January 24, 2010.
we measure in clinical practice should be on the basis of factors other than predictive accuracy of the apolipoproteins versus the lipoprotein lipids. Because measurement of apolipoproteins represents additional costs, many might, understandably, conclude that these additional costs were not associated with additional benefits. The ERFC study is a valuable contribution to the challenge of improving our capacity to quantitate the atherogenic risk attributable to the plasma lipoproteins. To be sure, as with almost any study, there are limitations. Thus, the observation that directly measured LDL-C was as accurate as non-HDL-C at identifying risk is at variance with the
1933-2874/$ -see front matter Ó 2010 National Lipid Association. All rights reserved. doi:10.1016/j.jacl.2010.01.005
84 results of most recent studies2 and suggests the power of the ERFC study to discriminate amongst the markers may not be ideal. Moreover, the methodology to measure lipids and apolipoproteins was highly variable among the studies, many of which were performed early in the course of measuring apolipoproteins. Inaccuracies in methodology cannot be overcome simply by increasing the number of studies and may be the basis for the heterogeneity in the results noted by the authors. Finally, we do not agree with their reasoning that excluded the INTERHEART3 and the International Study of Infarct Surviva (ISIS)4 studies. On the contrary, we believe the case-control design is particularly suited to testing the relative value of closely correlated variables. Nevertheless, our primary focus is not on any limitations of this study. Rather, we will try to demonstrate that even if the primary findings are correct—namely that the HRs for the apolipoproteins and lipids are equal for the overall groups that were analyzed—this does not mean that lipids and apolipoproteins are clinically interchangeable in all individuals. After all, it is individuals, not groups, in whom risk is estimated and treated and therefore, it is the utility of risk markers within individuals that counts. Furthermore, the variance that is responsible for the discordance between the apolipoprotein and cholesterol risk markers in many individuals is due to differences in the composition of the major apoB lipoproteins—very lowdensity lipoprotein (VLDL) and LDL— and therefore, discordance among the risk markers is embedded in the pathophysiological mechanisms responsible for changes in the number as well as the composition of the apoB lipoproteins. The atherogenic risk to the individual from the apoB lipoproteins is driven by these differences in particle number and composition. Accordingly, if our objective is the best care for each patient, the challenge is to learn how to combine the information from cholesterol and apolipoproteins to enlarge our diagnostic accuracy rather than to continue only to pit one against the other. That is, our objective should be to add information not eliminate it.
Journal of Clinical Lipidology, Vol 4, No 2, April 2010 Plasma apoB, therefore, measures total atherogenic particle number. Except in very unusual circumstances, LDL particles account for the vast majority of the apoB particles.5 Non-HDL-C is the sum of the cholesterol in the apoB lipoproteins. Again, LDL particles account for most of the non-HDL-C, although the proportion is more variable than in the case of apoB.5
Population HRs versus individual HRs As is the usual practice, ERFC reports HR for a parameter as the increase in risk per standard deviation of that parameter within the study population. By normalizing risk over the range of values in the population, the overall risk for different parameters can be compared within groups. We will refer to this HR as the population HR (PHR). As noted previously, the PHR for apoB and nonHDL C were both equal in the ERFC study. However, the risk for the actual level of any parameter for an individual—the individual HR (IHR)—depends on the actual concentration of the atherogenic parameter in that individual, that is, the greater the concentration, the greater the risk. Thus, Figure 1 depicts the relation between IHR of non-HDL C/apoB on the basis of their concentrations expressed as a percentile of the American adult population. Because both parameters have the same HR, their overall relation to risk will be nearly identical. However, the relation between risk and concentration is curvilinear, demonstrating that absolute risk from any factor depends on the absolute level of that factor and that risk rises exponentially while concentration increases linearly. Any group is made up of individuals, each of which has a specific concentration of non-HDL-C and apoB. Although non-HDL-C and apoB are highly correlated, they are not
Non-HDL-C and apoB This analysis will focus on non-HDL-C and apoB, which have emerged as the two most powerful markers of the vascular risk attributable to the atherogenic plasma lipoproteins. Non-HDL-C and apoB are highly correlated variables, and on the basis of this property, it has been argued that they are clinically interchangeable. The equivalent HRs of the ERFC study would appear to support that contention. We will test whether that conclusion is correct when individuals are considered. Non-HDL-C and apoB measure different things. NonHDL-C is the sum of the masses of cholesterol within the atherogenic lipoproteins— or the purposes of this discussion, the mass of cholesterol within VLDL and LDL particles—whereas apoB equals the number of atherogenic particles, that is, the number of VLDL and LDL particles.
Figure 1 ‘‘Effectiveness curves’’: PHRs adjusted for age, systolic blood pressure, and smoking from the ERFC by percentile by use of the 5th percentile as the reference group. Estimates pertain to 190,977,311 nonpregnant, noninstitutionalized, civilian US adult residents .20 years of age represented by 1700 participants in the NHANES 2005 to 2006 study with both lipid measures. The published HRs were 1.63 per SD of 43 mg/dL of non-HDL-C and 1.62 per SD of 29 mg/dL of apoB.
Sniderman et al
When is equal not equal?
perfectly concordant, that is, for any given value of one variable, there is a range of values for the other. The precise degree to which they are interchangeable for an individual depends on how concordant they are. Figure 2 demonstrates the relation between non-HDL-C and apoB in the National Health and Nutrition Examination Survey (NHANES) 2005 to 2006 survey of the American adult population.6 There are three points to note. First, the overall correlation coefficient is very high: 0.95. Second, at the extremes, concordance is also very high. That is, when either non-HDL-C or apoB is very high or very low, the other marker is also very high or very low. Therefore, one marker conveys as much information about risk as the other. Moreover, because the total risk attributable to any parameter is determined to a major degree by the risk between the extreme values, this means that the overall risk attributable to either nonHDL-C or apoB must be similar. Third, from approximately the 10th to the 90th percentile, there is a definite range of values of one parameter versus the other as reflected in a fall in the overall correlation between the two parameters from 0.95 to 0.88. This is also evident from the scatter of the values around the line of identity (Fig. 3). The IHR for non-HDL C and apoB, that is, the HR for each on the basis of their actual values within an individual, can be calculated for all of the individuals between the 10th and 90th percentiles of the distribution and the results are demonstrated in Figure 4. Substantial discordance in the risk attributable to non-HDL-C versus the risk attributable to apoB is evident throughout this wide range of values for the population. In Table 1 we compare groups of individuals defining as discordant those individuals whose apoB and non-HDL-C percentiles are separated by one or more deciles. These comparisons yield several important points. More than 17% of US adults have values of these two different biological measures that are discordant by this definition. An estimated 8.8% have an individual apoB percentile value more than 10 percentiles points greater than their nonHDL–C value. Furthermore overall and among men
Figure 2 Scatter plot of percentile of apoB versus percentile of Non-HDL-C in NHANES 2005 to 2006.
85
Figure 3 Scatter plot of percentile of apoB versus percentile of non-HDL-C between the 10th and 90th non-HDL-C percentiles in NHANES 2005 to 2006.
10-year CHD risk, as assessed by Framingham equations,7,8 is significantly greater on average in the apoB discordant group than in the non-HDL-C discordant group.
Clinical interpretations of discordance in attributable-risk of cardiovascular disease between non-HDL-C and apoB within an individual When the IHR of non-HDL-C and apoB are the same or very close within the individual, they obviously convey
Figure 4 Scatter plot of individual apoB hazard ratio versus individual non-HDL-C HR in NHANES 2005 to 2006 subjects with non-HDL-C values between the 10th and 90th percentile.
86
Journal of Clinical Lipidology, Vol 4, No 2, April 2010
Table 1 Comparisons by concordance/discordance category in NHANES 2005 to 2006 representative of 190,877,311 nonpregnant, noninstitutionalized, civilian US adult residents $20 years of age
All Concordant Non-HDL-C %ile $ apoB %ile 1 10 ApoB %ile $ Non-HDL-C %ile 110 Men Concordant Non-HDL C %ile $ apoB %ile 1 10 ApoB %ile $ Non-HDL-C %ile 110 Women Concordant Non-HDL C %ile $ apoB %ile 1 10 ApoB %ile $ non-HDL-C %ile 110
Sample, n
People represented, n
Percent 6 SE, %
Risk, %*
970 126 175
110,845,840 16,532,859 16,836,517
58.1 6 2.5 8.7 6 2.5 8.8 6 1.5
3.2 2.7 5.3
.19 .002†
507 59 104
55,156,771 6,454,543 8,946,989
59.8 6 2.7 7.0 6 2.0 9.7 6 1.9
5.7 4.9 9.0
0.22 .006†
463 67 71
55,689,068 10,078,316 7,889,528
56.5 6 3.0 10.2 6 3.0 8.0 6 1.4
1.8 1.8 2.8
.94 .08
P vs row above
Individuals in the bottom and top deciles of either apoB or non-HDL-C (n 5 429 representing 24.4% of the population) were excluded. Individuals whose apoB and non-HDL C percentile (%ile) values differed by one decile or more were defined as discordant. All statistics were estimated using SAS 9.2 SURVEY procedures to account for the NHANES complex sample design. *Back-transformed mean of logit-transformed Framingham 10-year primary7 or secondary8 risk or coronary heart disease. †These values indicate statistical significance.
very similar information as to risk and pose no challenge as to interpretation. However, when they are substantially different within the individual, interpretation is more demanding. In many cases, however, we believe the discordance can be explained on the basis of wellrecognized differences in the metabolism of the major apoB lipoproteins—VLDL and LDL—that produce the differences in their composition that lead to the different IHRs. In these situations, the two measures contribute additive, not redundant, information. Two discordant IHR phenotypes need to be considered: (1) the apoB percentile . non-HDL-C percentile and (2) the non-HDL-C percentile . the apoB percentile.
apoB percentile. non-HDL C percentile This indicates that relative to their respective distributions in the population the total number of apoB particles is greater than the mass of cholesterol within them. Because LDL particles comprise the great majority of total apoB articles and because LDL-C makes up the great majority of non-HDL-C, this points to a relative excess of LDL particles compared with LDL-C. This specific pattern of discordance in the concentrations of LDL-C and LDL particle number is caused by an excess of small dense, cholesterol-depleted particles.5 This results in an apoB that is disproportionately high to LDL-C and to non-HDL-C in this instance as well. Many have argued that small dense LDL are more atherogenic than larger, cholesterol-enriched LDL and multiple in vitro studies2 have identified specific mechanisms that could explain this increased risk. These range from smaller size and therefore greater ease of entry into the arterial wall to increased binding affinity to the arterial glycosaminoglycans to greater risk of oxidation.5 Whether
the difference is clinically significant is not certain. Accordingly, we believe the secure conclusion for the moment is that there is no credible body of evidence to indicate that small dense LDL particles are associated with a sharply diminished risk on a per-particle basis than larger LDL particles and therefore, there is no justification that they should be ignored. It follows that in patients with disproportionately high apoB, a prudent physician would assess risk based on the apoB not on the non-HDL-C even assuming their PHR are identical. The Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER) study is an apposite example of how extreme the differences in non-HDL-C and apoB can be and how this can affect the clinical interpretation of events.9 On the basis of low levels of LDL C and non-HDL-C (approximately the 25th and 30th percentiles of the American population respectively),10 the risk from atherogenic apoB lipoproteins would be considered to be low. However, the level of apoB was considerably greater, approximately the 65th percentile of the American population.11 Given the level of apoB and the age of the subjects (average age 66 years) and therefore given the duration of the exposure to moderately elevated levels of atherogenic lipoprotein particles, the risk of vascular disease in the control group and the benefit from potent statin therapy should not be surprising.
Non-HDL C . apoB There are two pathophysiological explanations for the level of non-HDL-C to be greater than the concurrent apoB. The first are remnant VLDL and remnant chylomicron particles and the second, cholesterol-enriched LDL particles. Patients with familial dysbetalipoproteinemia (FDB) represent the extreme case of excessive accumulation of
Sniderman et al
When is equal not equal?
cholesterol-enriched remnant apoB48 and apoB100 lipoprotein particles, a syndrome that is unquestionably associated with an extreme elevation of the risk of vascular disease. FDB is characterized by high levels of total and non-HDL-C, high levels of plasma triglycerides, and normal levels of plasma apoB.12 Remnants are large cholesterol-rich, triglyceride-rich particles, which explains the hypercholesterolemia and hypertriglyceridemia. Failure to convert VLDL particles to LDL particles is another of the metabolic hallmarks of FDB. LDL apoB, accordingly, accounts, on average, for only half the total apoB particles in sharp contrast to the usual circumstance in which LDL particles make up 90% or more of the total apoB particles. Because the half-life of LDL particles is much longer than remnant particles, total plasma apoB is normal in these patients. Accordingly, the IHR for non-HDL C more accurately identifies risk than the IHR for apoB in those with FDB. Interpretation in these patients is not difficult because the overall pattern of marked hypertriglyceridemia and marked hypercholesterolemia with normal apoB plus the diagnostic ratios of TG/apoB and TC/apoB allow for specific diagnosis.13 Even so, FDB, as such, will account for only a small minority of circumstances in which nonHDL C is disproportionately greater than apoB. Remnants can be present in smaller amounts and could pose significant risk. At the present time, therefore, in hypertriglyceridemic patients in whom non-HDL C is greater than apoB, the prudent physician could base assessment of risk on nonHDL C rather than apoB. Cholesterol-enriched LDL particles constitute the alternative explanation for a level of non-HDL-C that is disproportionately elevated compared with apoB. Familial hypercholesterolemia represents a classical example of cholesterol-enriched LDL particles, but, in this instance, the discordance is of little significance because the values of LDL-C, non-HDL-C, and apoB are all extremely elevated. However, cholesterol-enriched LDL particles can occur with levels of apoB well within the normal distribution of values. The pathophysiological mechanism leading to cholesterol-enrichment of LDL particles has not been elucidated. Nevertheless, the clinical issue is whether the greater IHR from the non-HDL C based on cholesterolenriched LDL C is correct compared with the lower IHR based on the apoB. Unfortunately, the question cannot be answered definitively at the moment. However, the data from the Framingham Offspring Study demonstrate that risk in those with low LDL-C and low LDL particles (and therefore low apoB) is low and indistinguishable from those with high LDL-C and low LDL particles.14 That is, the risk of cardiovascular event was determined by the number of LDL particles, not by the mass of cholesterol within them. This suggests that in subjects in whom non-HDL-C is greater than apoB as the result of cholesterol-enriched LDL, apoB may be a more reliable marker of risk than nonHDL-C. Further work is necessary to test this conclusion.
87 Nevertheless, whether the target is LDL C, non-HDL-C or apoB, the target levels in terms of the population values should be equivalent. On the basis of the NHANES 2005 to 2006 survey, a non-HDL C of 125 mg/dL and an apoB of 90 mg/dL would be equivalent to an LDL-C of 100 mg/dL—the high-risk targets at the 40th percentile— whereas a non-HDL-C of 90 mg/dL and an apoB of 65 mg/dL would be equivalent to an LDL C of 70 mg/dL— the very high risk values at the 10th percentile.6
Summary Our objective has been to demonstrate that findings in groups do not necessarily apply to individuals. We do not devalue these overall findings from studies such as ERFC and the general relationships they establish. They provide the first irreplaceable level of evidence. However, they should not end enquiry because as physicians our concern is the outcome of individuals not of groups. Thus, even if the overall findings of the ERFC are correct, and we have reservations on that point as noted at the outset, the actual hazard that a factor poses to an individual depends on the actual level of that factor within that individual, not on the average hazard of the factor for the group. The distinction is not dissimilar to that between the average temperature in a locale versus the actual temperature on a particular day in a particular season. Acknowledging variance means acknowledging individuals. Accordingly, one lesson that should be drawn from the ERFC study is that the conventional statistical methods may not be as sensitive as desirable to compare closely correlated risk factors. An equally important one is that clinical decisions should be based on risk in the individual not in the group and therefore the findings of the ERFC study should not be blindly transferred to clinical practice. The challenge, we believe, is not to pit one number against another but to integrate the information from both. As noted at the outset, apoB and non-HDL C measure different things. ApoB accurately measures atherogenic particle number and is a more accurate guide to LDL particle number than cholesterol. Therefore, when apoB is disproportionately high, apoB should guide management. By contrast, apoB is insensitive to remnant particles whereas plasma lipids, in particular cholesterol and triglyceride, do reflect the impact of this abnormality. Accordingly, when non-HDL-C is disproportionately high, particularly in hypertriglyceridemic patients, non-HDL-C should guide management. Our overall conclusion, therefore, is that apolipoproteins and lipoprotein lipids should be used in a complementary rather than a mutually exclusive fashion and that we should move to amplify the diagnostic information in an individual rather than diminish it. We acknowledge that not all countries have the medical infrastructure and resources to apply this approach immediately. That does not exclude it as an objective. Moreover, in some settings, apolipoproteins may be easier and simpler
88 to incorporate in clinical practice just as cell phones were easier and simpler to adopt compared with land lines. Our point is that we can make the best choices when we have the most knowledge.
Journal of Clinical Lipidology, Vol 4, No 2, April 2010
7.
8.
References 9. 1. Danesh J. The Emerging Risk Factors Collaboration. Major lipids, apolipoproteins, and risk of vascular disease. J Am Med Assoc. 2009;302: 1993–2000. 2. Sniderman AD, de Groot E, Couture P. ApoB and the atherogenic ApoB dyslipoproteinemias. In: Kwiterovich PO Jr., editor. The Johns Hopkins Textbook of Dylipidemia. Philadelphia, PA: Lippincott Williams & Wilkins, 2009. p. 196–210. 3. McQueen MJ, Hawken S, Wang X, et al. The relative importance of lipids, lipoproteins, and apolipoproteins as risk markers associated with myocardial infarction in 52 countries. Lancet. 2008;372:224–233. 4. Parish S, Peto R, Palmer A, et al. The joint effects of apolipoprotein B, apolipoprotein A1, LDL cholesterol, and HDL cholesterol on risk: 3510 cases of acute myocardial infarction and 9805 controls. Eur Heart J. 2009;30:2137–2146. 5. Barter PJ, Ballantyne CM, Carmena R, et al. ApoB versus cholesterol in estimating cardiovascular risk and in guiding therapy: report of the thirty-person/ten-country panel. J Intern Med. 2006;259:247–258. 6. NHANES Investigators. NHANES Analytic Guidelines 2003–2004. Available at: http://www.cdc.gov/nchs/data/nhanes/nhanes_03_04/
10.
11.
12.
13.
14.
nhanes_analytic_guidelines_dec_2005.pdf. Accessed February 16, 2009. Wilson PWF, D’Agostino RB, Levy D, Belanger AM, Halit S, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation. 1998;97:1837–1847. D’Agostino RB, Russell MW, Huse DM, Ellison RC, Silbershatz H, Wilson PW, et al. Primary and subsequent coronary risk appraisal: new results from the Framingham study. Am Heart J. 2000;139:272–281. Ridker PM, Danielson E, Fonseca FA, et al. Rosuvastatin to prevent vascular events in men and women with elevated C-reactive protein. N Engl J Med. 2008;359:2195–2207. Sniderman A, Williams K, Cobbaert C. ApoB versus non-HDLC: What to do when they disagree. Curr Atheroscler Rep. 2009;11: 358–363. Ridker PM, Danielson E, Fonseca FA, et al. Reduction in C-reactive protein and LDL cholesterol and cardiovascular event rates after initiation of rosuvastatin: a prospective study of the JUPITER trial. Lancet. 2009;373:1175–1182. Sniderman AD, Tremblay A, Bergeron J, Gagne C, Couture P. Diagnosis of type III hyperlipoproteinemia from plasma total cholesterol, triglyceride, and apolipoprotein B. J Clin Lipidol. 2007;1:256–263. de Graaf J, Couture P, Sniderman A. A diagnostic algorithm for the atherogenic apolipoprotein B dyslipoproteinemias. Nat Clin Endocrinol Metab. 2008;4:608–618. Cromwell WC, Otvos JD, Keyes MJ, et al. LDL particle number and risk of future cardiovascular disease in the Framingham Offspring Study—Implications for LDL management. J Clin Lipidol. 2007;1: 583–592.