Direct comparisons of lipoprotein(a) values, several established cardiovascular risk factors, and emerging markers in relation to within-person variability across 12 years (expressed as the regression dilution ratio [calculated using the Rosner multivariate regression method, adjusted for baseline age, sex, smoking history, diabetes mellitus history, total cholesterol, log triglycerides, systolic blood pressure, and body mass index]) (A) and odds ratios (top third vs bottom third) for coronary heart disease (CHD) (adjusted for established risk factors [age, sex, period of recruitment, smoking status, history of diabetes mellitus, total cholesterol, log triglycerides, systolic blood pressure, and body mass index]) (B). *Regression dilution ratios were calculated using the log-transformed variables. Error bars represent 95% confidence intervals.
Odds ratios for coronary heart disease by fifths (F) of baseline lipoprotein(a) levels adjusted for age, sex, period of recruitment, smoking status, and other established risk factors (total cholesterol, log triglycerides, systolic blood pressure, history of diabetes mellitus, and body mass index). The size of the data markers is proportional to the inverse of the variance of the odds ratios. Fifths were calculated on the basis of the distribution of controls. Geometric mean baseline lipoprotein(a) values in each F were as follows: F1, 3.49 mg/L; F2, 35.29 mg/L; F3, 84.23 mg/L; F4, 171.30 mg/L; and F5, 383.42 mg/L (to convert to micromoles per liter, multiply by 0.0357). Test for linear trend of odds ratios across fifths of lipoprotein(a) levels: P < .001. Error bars represent 95% confidence intervals (CIs) (calculated using floating variances).
Investigation of possible sources of heterogeneity in associations between lipoprotein(a) (Lp[a]) levels and risk of coronary heart disease (CHD) involving individual characteristics in the Reykjavik Study (A) and study-level characteristics in an updated meta-analysis of 31 studies (B). Values are adjusted for age, sex, smoking status, total cholesterol (to convert to millimoles per liter, multiply by 0.0259), log triglycerides (to convert to millimoles per liter, multiply by 0.0113), systolic blood pressure, body mass index, and history of diabetes mellitus. The size of the data markers is proportional to the inverse of the variance of the odds ratios. Thirds of systolic blood pressure, total cholesterol, triglycerides, and tissue plasminogen activator antigen were defined by their respective distributions in cases. Apart from heterogeneity for publication period (P = .004) and sample type (P = .003), there was no evidence of significant interaction between the different subgroups and Lp(a) levels. Error bars represent 95% confidence intervals. To convert C-reactive protein to nanomoles per liter, multiply by 9.524.
Odds ratios for coronary heart disease (CHD) (top third vs bottom third) in each of 31 published prospective studies of lipoprotein(a) in essentially general populations. Heterogeneity: χ230 = 52.6; P = .007: I2 = 43% (95% confidence interval [CI], 12%-63%). ARIC indicates Atherosclerosis Risk in Communities; BUPA, British United Provident Association; GRIPS, Göttingen Risk, Incidence and Prevalence Study; Lip Res Clin Prev Trial, Lipid Research Clinics Coronary Primary Prevention Trial; MONICA, Monitoring Trends and Determinants in Cardiovascular Disease; MRFIT, Multiple Risk Factor Intervention Trial; PRIME, Prospective Epidemiological Study of Myocardial Infarction; PROCAM, Prospective Cardiovascular Münster Study; VIP, Västerbotten Intervention Project; WHS, Women's Health Study; and WOSCOPS, West of Scotland Coronary Prevention Study. Error bars represent 95% CIs.
Bennet A, Di Angelantonio E, Erqou S, Eiriksdottir G, Sigurdsson G, Woodward M, Rumley A, Lowe GDO, Danesh J, Gudnason V. Lipoprotein(a) Levels and Risk of Future Coronary Heart DiseaseLarge-Scale Prospective Data. Arch Intern Med. 2008;168(6):598–608. doi:10.1001/archinte.168.6.598
Large-scale prospective data are needed to determine whether associations between lipoprotein(a) (Lp[a]) and coronary heart disease (CHD) risk are independent of established risk factors, to characterize the shape of this relationship, and to quantify associations in relevant subgroups.
Levels of Lp(a) were measured in samples obtained at baseline from 2047 patients who had first-ever nonfatal myocardial infarction or who died of CHD during the study and from 3921 control participants in the Reykjavik Study (n = 18 569), as well as in paired samples obtained 12 years apart from 372 participants to quantify within-person fluctuations.
Baseline Lp(a) levels had little or no correlation with known cardiovascular risk factors, such as age, sex, total cholesterol level, and blood pressure. The Lp(a) values were highly consistent from decade to decade, with a regression dilution ratio (calculated on the log scale) of 0.92 (95% confidence interval, 0.85-0.99). The odds ratio for CHD, unaltered after adjustment for several established risk factors (age, sex, smoking status, blood pressure, total cholesterol, triglycerides level, diabetes mellitus, and body mass index), was 1.60 (95% confidence interval, 1.38-1.85) in a comparison of extreme thirds of baseline Lp(a) levels. Odds ratios were progressively higher with increasing Lp(a) levels and did not vary materially by several individual- or study-level characteristics.
There are independent, continuous associations between Lp(a) levels and risk of future CHD in a broad range of individuals. Levels of Lp(a) are highly stable within individuals across many years and are only weakly correlated with known risk factors. Further assessment of their possible role in CHD prevention is warranted.
Lipoprotein(a) (Lp[a]) is a low-density lipoprotein–like particle synthesized by the liver that consists of an apolipoprotein B molecule covalently linked to a very large glycoprotein known as apolipoprotein(a) (Apo[a]).1,2 Several epidemiologic studies have assessed the association between circulating Lp(a) levels and cardiovascular diseases. By 2000, there were 18 population-based prospective studies3- 20 that had reported on Lp(a) levels and coronary heart disease (CHD) risk, with most, but not all, reporting positive associations. Few studies, however, have been adequately powered to examine potentially important aspects of the association, such as the shape of the Lp(a)-CHD relationship and the size of relative risks in clinically relevant subgroups (such as in men and women or at different levels of established risk factors). A previous review21 suggested a moderately strong overall association between Lp(a) levels and CHD risk, but because it analyzed only published data (rather than primary data) it did not address the uncertainties described in the preceding sentences. Furthermore, data on within-person variability are needed to help assess the long-term relevance of Lp(a) to CHD, but only 1 previous study22 has reported on it using a small subset of individuals.
We report new primary data on the largest single study of Lp(a) concentrations and CHD thus far, involving 2047 patients with either first-ever nonfatal myocardial infarction or coronary death and 3921 control subjects “nested” within a prospective population-based cohort of 18 569 participants. As recommended by an expert panel,23 we used an assay system that is not sensitive to Apo(a) isoform heterogeneity. Paired measurements were performed approximately 12 years apart in 372 participants to help quantify within-person variability in Lp(a) levels. We also report an updated review of previous prospective studies to help assess the comparability of associations of Lp(a) level with CHD risk reported in studies involving different blood handling, storage, and assay methods, particularly with assays affected by the variable affinity of antibodies to particular Apo(a) isoforms.24,25 The focus of the present report is on whether there is likely to be an etiologic association between Lp(a) levels and CHD (rather than the separate issue of risk prediction).
The Reykjavik Study, initiated in 1967, has been described in detail elsewhere.26 All men born between January 1, 1907, and December 31, 1934, and all women born between January 1, 1908, and December 31, 1935, who were residents of Reykjavik and its adjacent communities on December 1, 1966, were identified in the national population register and were invited to participate in the study. Five stages of recruitment, between 1967 and 1991, yielded 8888 male and 9681 female participants with no history of myocardial infarction (72% response rate). Nurses administered questionnaires, performed physical measurements, recorded electrocardiograms, and collected fasting venous blood samples. Serum was stored at −20°C until assay. All the participants were monitored by central registries for the occurrence of major cardiovascular morbidity (based on MONICA [Monitoring Trends and Determinants in Cardiovascular Disease] criteria) or cause-specific mortality (based on a death certificate with International Classification of Diseases, Ninth Revision codes 410-414), with loss to follow-up of only approximately 0.6% to date. A total of 2459 men and women recorded either nonfatal myocardial infarction or coronary death between study entry and the censoring date. One or 2 controls were frequency matched to cases by calendar year of recruitment, sex, and age (in 5-year age bands) from among all participants who did not develop CHD during follow-up, giving a total of 3969 controls. Because of random nonavailability of serum samples, the present study is restricted to 2418 incident CHD cases and 3921 controls with available Lp(a) measurements. The study protocol was approved by the National Bioethics Committee and the Data Protection Commission of Iceland. All the participants gave informed consent.
Levels of Lp(a) were measured in serum samples by laboratory staff unaware of participants' disease status using an enzyme immunoassay (ELITEST Lp[a]) and an assay standard (both from HYPHEN BioMed, Paris, France). This enzyme-linked immunosorbent assay–based system, which uses a monoclonal anti-Lp(a) antibody for capture and a polyclonal anti-Apo(B) antibody for detection, is not affected by Apo(a) isoform variation. The intra-assay and interassay coefficients of variation were 4.2% and 4.7%, respectively. The Lp(a) measurements were made in the 372 participants who provided paired samples at a mean interval of approximately 12 years. Lipid, biochemical, and hematologic measurements have been described previously.26,27
To minimize any impact of preexisting disease, principal analyses were restricted to the 2047 patients and 3921 controls without evidence of CHD or stroke at the baseline examination (ie, participants with electrocardiographic abnormalities or a history of myocardial infarction, angina, or stroke were excluded from the main analyses, although they were retained in subsidiary analyses). The Lp(a) values were natural log transformed to achieve an approximately symmetrical distribution. Unconditional logistic regression was used to calculate odds ratios (ORs) and 95% confidence intervals (CIs), progressively adjusted for possible confounding factors (Stata 9.2;StataCorp LP, College Station, Texas). The shape of the association between Lp(a) levels and CHD risk was investigated using groups defined by fifths of the baseline values of Lp(a) in controls; the corresponding 95% CIs were estimated from floated variances that reflect the amount of information underlying each group (including the reference group).28 Subgroup analyses by sex, smoking habits, systolic and diastolic blood pressure (BP), concentrations of serum lipids and C-reactive protein, and type of CHD outcome were also prespecified. To quantify within-person variability in levels of Lp(a) (and in other markers), regression dilution ratios were estimated from the available paired measurements by regressing repeated measures on baseline values. Regression dilution ratios for variables with skewed distributions (ie, Lp[a], C-reactive protein, and triglycerides) were calculated on the log scale.29 An updated meta-analysis was conducted of prospective studies published before December 1, 2006, with more than 1 year of follow-up in essentially general populations (ie, in cohorts not selected on the basis of preexisting disease).30 The analysis was restricted to nonfatal myocardial infarction or coronary death. To reduce potential biases, all the analyses involved only within-study comparisons (ie, cases and controls were directly compared only within each cohort). Data were combined using a random-effects model. Heterogeneity was assessed using standard χ2 tests and the I2 statistic.31 In the few studies that reported only 3 or 4 categories of Lp(a), rather than continuous values (owing to the use of semiquantitative assays based on reading of electrophoretic bands6,11), the highest category was taken to correspond to the upper third and the lowest category to the bottom third of baseline Lp(a) values. For studies reporting associations for men and women separately, a pooled estimate was calculated, weighted by their contributing proportions. Analyses involved formal tests of interaction to assess the effect of the following prespecified study characteristics: year of publication, study size, geographic location, ethnicity, sample storage features (ie, temperature and type), and features related to Lp(a) assay methods used.
As expected, levels of established cardiovascular risk factors at the baseline examination were higher in patients with CHD than in controls (Table 1). Baseline log-Lp(a) levels were higher in patients with CHD than in controls and were weakly, although significantly, correlated with levels of total cholesterol (r = 0.12; 95% CI, 0.09 to 0.15), log triglycerides (r = −0.12; 95% CI, −0.16 to −0.09), tissue plasminogen activator antigen (r = −0.09; 95% CI, −0.12 to −0.06), serum creatinine (r = −0.05; 95% CI, −0.08 to −0.02), and uric acid (r = −0.06; 95% CI, −0.09 to −0.02). There were no significant correlations between baseline log-Lp(a) levels and various established and emerging cardiovascular risk factors, such as age, sex, BP, body mass index, C-reactive protein, and albumin (data available on request). In the 372 participants who provided paired measurements at baseline and approximately12 years later, the regression dilution ratios were as follows: 0.92 (95% CI, 0.85-0.99) for log Lp(a), 0.54 (95% CI, 0.44-0.64) for log C-reactive protein, 0.57 (95% CI, 0.48-0.66) for log triglycerides, 0.55 (95% CI, 0.45-0.65) for von Willebrand factor, 0.59 (95% CI, 0.51-0.67) for total cholesterol, and 0.65 (95% CI, 0.54-0.77) for systolic BP (Figure 1A). Data on high-density lipoprotein cholesterol were unavailable.
In a comparison of individuals with baseline Lp(a) values in the top third vs the bottom third, the OR for CHD was 1.61 (95% CI, 1.41-1.84) after adjustment for age, sex, and calendar year of recruitment (Table 2). This OR changed little after further adjustment for several established cardiovascular risk factors (ie, smoking status, BP, total cholesterol, triglycerides, body mass index, and diabetes mellitus) and inflammatory markers (eg, C-reactive protein). Subsidiary analyses yielded adjusted ORs for CHD of 1.77 (95% CI, 1.57-1.99) in a comparison of extreme fifths and of 1.23 (95% CI, 1.16-1.31) for log-Lp(a) levels higher by 1 SD. Figure 1B shows that, in comparisons of several established and emerging markers in the same patients and controls, ORs for CHD with Lp(a) were smaller than those with total cholesterol. Figure 2 shows that the ORs for CHD increased continuously with increasing Lp(a) levels (P < .001, test for linear trend), although further work is needed to determine whether a straight line or a curvilinear line better describes the association. Figure 3A suggests that the association of Lp(a) levels with CHD risk did not vary materially in a range of subgroups based on individual characteristics, notably, sex, lipid concentrations, C-reactive protein, and fatal vs nonfatal CHD outcome (P > .10 for each test of heterogeneity).32- 47
Table 3 and Table 4 summarize the characteristics of 31 prospective studies of Lp(a), the first of which was published in 1990,19 with 14 studies reported since the publication of a meta-analysis in 2000.32- 41,43- 46 All the studies were based in continental North America or in Western Europe except 1.36 Most studies identified participants in population registers (eg, general practitioner lists or electoral rolls) or in occupational settings, involved middle-aged men of white European continental ancestry, and reported on incident myocardial infarction and coronary death outcomes. The interval between sample collection and assay performance varied from a few hours to approximately 20 years. Eighteen studies7,8,11,14,15,17,32- 35,39- 42,44,46- 48 measured Lp(a) levels in plasma and 13 studies (including the present study)5,6,10,12,13,18,19,36,37,45 in serum, with measurements generally performed in samples thawed after long-term storage at temperatures of −70°C or colder, whereas few studies conducted assays in samples stored at temperatures ranging from −70°C to −20°C12,17,18,39,41 or in freshly collected samples.6,10,35 Apart from 6 studies6,8,10,11,32,35 that used in-house Lp(a) assays, most of the studies used commercially available immunoassays. Assay results were generally reported as mass per volume, although 1 study8 used an analytical method that measured molarity, and 2 studies6,11 used semiquantitative assay methods. Detailed information on assay methods (such as the exact antibodies used and the existence of sensitivity to Lp[a] isoforms) was reported in only a subset of studies (Table 4). Reported mean or median levels of Lp(a) in controls varied substantially across studies, ranging from approximately 10 to 300 mg/L (to convert to micromoles per liter, multiply by 0.0357) (although, as in the present study, most were 50-200 mg/L). All but 3 studies10,32,45 reported adjustment of CHD ORs for at least age, sex, smoking status, BP, and lipid concentrations. Using only within-study comparisons, a combined analysis of published data from these studies (including the present study) involving a total of 9870 incident CHD cases yielded an adjusted OR of 1.45 (95% CI, 1.32-1.58) for individuals in the top third of the baseline Lp(a) distribution compared with those in the bottom third (Figure 4). There was moderate heterogeneity among these studies (χ230 = 52.6; P = .007; I2 = 43% [95% CI, 12%-63%]), some of which was explained by period of publication (P = .004) and sample type (P = .003) but only a small part by other characteristics prespecified for investigation, notably, study size, sample storage characteristics, and Lp(a) assay isoform sensitivity or standard used (P > .10 for each characteristic) (Figure 3B). A funnel plot did not show an excess of extreme findings in smaller studies (Egger test P = .23) (data available on request).
We demonstrated that the decade-to-decade consistency of Lp(a) levels in adults is very high, considerably higher than that of BP, serum lipid levels, and C-reactive protein concentration. Contrary to previous reports of no associations with CHD risk4,15,45 or of effects at only very high Lp(a) levels, the present, much larger-scale data indicate an approximately continuous relationship. In direct comparisons with several established and emerging markers, we showed that the OR for CHD with elevated Lp(a) levels is comparable to those with systolic BP and at least as strong as those with C-reactive protein27 and triglycerides.49 However, whereas ORs with C-reactive protein27 or triglycerides49 in this population attenuated considerably after adjustment for established risk factors (eg, smoking status, lipids, BP, diabetes mellitus, and body mass index), the OR with Lp(a) changed very little after such adjustment. This observation suggests that Lp(a) levels are associated with CHD risk independent of such factors. We showed that ORs for CHD with Lp(a) levels were similar in a range of clinically relevant subgroups, such as in men and women, or at different levels of established risk factors and under different blood handling, storage, and assay conditions.
These findings may have several implications for the development of CHD prevention strategies. First, the demonstration of high consistency of Lp(a) levels within individuals across many years emphasizes the lipoprotein's lack of substantial correlation with lifestyle characteristics or with several established risk factors (as shown in the cross-sectional analyses) and underscores the strong influence of the LPA locus on Lp(a) levels.50 Such high reproducibility suggests the simplifying conclusion that, unlike many other biomarkers, most of the effect of Lp(a) on CHD risk can be assessed using a single measurement. Second, by reliably showing that there are progressively higher ORs for CHD with increasing Lp(a) levels, we have renewed interest in existing and new strategies to modify Lp(a) levels. The demonstration of moderately strong ORs for CHD with elevated Lp(a) levels independent of several established risk factors should encourage studies that can help determine whether Lp(a) levels are causally involved in CHD. Large randomized trials of niacin in CHD prevention are already in progress (eg, HPS2-THRIVE [Heart Protection Study 2 Treatment of HDL (High-Density Lipoprotein) to Reduce the Incidence of Vascular Events]), although this agent raises high-density lipoprotein cholesterol levels and lowers low-density lipoprotein cholesterol and triglycerides in addition to lowering Lp(a) levels.51 Studies of CHD that use specific LPA genetic variants as proxies for circulating Lp(a) levels should also reduce potential biases, but they may need to be very large.50,52,53
The strengths and potential limitations of the present study merit careful consideration. These new data involve approximately 3 times as many incident CHD cases as the previous largest study32 that quantitatively assessed Lp(a) levels. We identified participants in population registers, had high response and follow-up rates, used robust methods to ascertain CHD outcomes, and minimized potential biases by excluding individuals with prevalent CHD or stroke. Concomitant measurements of several established and emerging markers enabled direct comparisons of ORs with different markers and allowed adjustment for a range of possible confounding factors, although the latter was somewhat limited owing to a lack of data on low- and high-density lipoprotein cholesterol (but previously published studies have reported only weak associations of Lp[a] with these lipid subfractions54- 56 and little effect on ORs for CHD after adjustment for them35,39). The present Lp(a) assay involved a detection antibody directed toward the Apo(B-100) component of the Lp(a) particle, and, hence, the measurement of Lp(a) levels was not sensitive to Apo(a) isoforms.25,57 The validity of the assay was confirmed by the observation of high decade-to-decade reproducibility in Lp(a) levels. However, because the present measurements did not provide specific information about Apo(a) isoforms (or record oxidized low-density lipoprotein), they could not test suggestions proposed in earlier studies of particularly strong associations with smaller-sized Lp(a) particles58 or in the presence of markers of oxidative damage.59 The present large-scale new data, reinforced by an updated meta-analysis of 31 long-term prospective studies, suggest only modest heterogeneity in OR for CHD with Lp(a) levels despite diversity of assay methods23 and variability in Lp(a) levels across populations. Despite substantial differences noted in Lp(a) levels among studies, none of the factors recorded (eg, features related to blood handling, storage, or assay conditions) in this updated review yielded important differences in ORs (apart from the possibility of more extreme results in studies involving serum rather than plasma, an exploratory finding that requires further investigation). A more detailed exploration of potential sources of heterogeneity requires collaborative pooling of individual participant data from prospective studies.60 Further studies are needed in racial groups, such as in people of African descent, in whom Lp(a) levels are particularly high.61
In conclusion, we observed, under various circumstances, continuous associations between Lp(a) levels and CHD risk apparently independent of the effect of several established cardiovascular risk factors. Levels of Lp(a) are highly stable within individuals across many years and are only weakly correlated with known risk factors. Further assessment of their role in CHD prevention is warranted.
Correspondence: John Danesh, FRCP, DPhil, Department of Public Health and Primary Care, University of Cambridge, Strangeways Research Laboratory, Wort's Causeway, Cambridge CB1 8RN, England.
Accepted for Publication: October 28, 2007.
Author Contributions: Drs Bennet and Di Angelantonio contributed equally to this work and are considered joint first authors. Study concept and design: Danesh and Gudnason. Acquisition of data: Eiriksdottir, Sigurdsson, Rumley, Lowe, Danesh, and Gudnason. Analysis and interpretation of data: Bennet, Di Angelantonio, Erqou, Sigurdsson, Woodward, and Danesh. Drafting of the manuscript: Bennet, Di Angelantonio, and Danesh. Critical revision of the manuscript for important intellectual content: Bennet, Di Angelantonio, Erqou, Eiriksdottir, Sigurdsson, Woodward, Rumley, Lowe, Danesh, and Gudnason. Statistical analysis: Bennet, Di Angelantonio, Erqou, and Woodward. Obtained funding: Sigurdsson and Danesh. Administrative, technical, and material support: Sigurdsson and Rumley. Study supervision: Lowe, Danesh, and Gudnason.
Financial Disclosure: None reported.
Funding/Support: This study was supported by a program grant from the British Heart Foundation (Drs Lowe, Danesh, and Gudnason) and by the Raymond and Beverly Sackler Research Award in the Medical Sciences (Dr Danesh). Aspects of the study were supported by an unrestricted educational grant from GlaxoSmithKline (Dr Danesh).
Additional Contributions: The following investigators provided additional information from their studies: Ian Ford, PhD, Barbara Howard, PhD, Paul Ridker, MD, Veikko Salomaa, MD, PhD, and Leon Simons, MD. Adam Butterworth, MSc, Philip Perry, MD, Mark B. Pepys, MD, PhD, FRCP, FRS, FMedSci, and Kausik Ray, MD, MPhil, commented helpfully. Estelle Poorhang and Paul Welsh provided technical assistance.
This article was corrected for typographical errors on 3/24/2008.