Supplementary Online Content

Transcription

Supplementary Online Content
Supplementary Online Content
The Emerging Risk Factors Collaboration. Lipoprotein(a) Concentration and the Risk of Coronary
Heart Disease, Stroke, and Nonvascular Mortality. JAMA. 2009;302(4):412-423.
eAppendix 1. Description of Methods Used in the Emerging Risk Factors Collaboration
eAppendix 2. Rationale for Using Non-HDL Cholesterol Rather Than Calculated LDL Cholesterol
in the Current Analyses
eAppendix 3. List of Acronyms for Studies Included in the Current Report
eTable 1. Baseline Characteristics, Blood Handling, Storage and Assay Characteristics of Studies
Contributing Data to the Current Analysis
eTable 2. Characterisation of Baseline and Incident Cardiovascular Disease Outcomes in Studies
Contributing Data to the Current Analysis
eTable 3. Results From Quadratic Models for the Association Between Usual Lp(a) Levels and the
Risk of CHD
eTable 4. Risk Ratios for CHD for 3.5-fold (ie, 1-SD) Higher Usual Levels of Lp(a), Further
Adjusted for Usual Values of Various Potential Confounding Factors
eTable 5. Parallel Analyses of the Association of Lp(a) With Disease Risk: (a) for Comparison of
Individuals in Extreme Thirds of Baseline Level Distributions, and (b) per 3.5-fold (ie, 1-SD)
Higher Baseline Level
eFigure 1. Study Specific Adjusted Risk Ratios for CHD, Corresponding to the Adjusted Risk
Ratio in Table 2
eFigure 2. Mean Lp(a) Levels by Cohort and (a) Assay Method Principle, or (b) Whether the Assay
Method Used Was Sensitive to Apolipoprotein(a) Isoform Variation
eFigure 3. Within-Person Variability in Lipoprotein(a) Levels (Regression Dilution Ratios†
Stratified by Study and Repeat)
eFigure 4. Age- and Sex-Adjusted Risk Ratios for Various Vascular and Non-Vascular Endpoints
per 3.5 Fold
Higher Usual Lp(a) Levels
eFigure 5. Risk Ratios for Coronary Heart Disease by Fifths of Usual Lp(a) Levels, After
Excluding the First 5 Years of Follow-up
eFigure 6. Risk Ratios for Coronary Heart Disease per 3.5-fold Higher Usual Lp(a) Levels, by
Strata of Various Study Level Characteristics
eFigure 7. Direct Comparison of Adjusted Risk Ratios for CHD Between Lp(a) and Non-HDL-C,
for a 1-SD Higher† Baseline or Usual Levels
eFigure 8. Association of Lp(a) With Fatal Vascular and Non-Vascular Outcomes in Analyses That
Did Not Censor for Nonfatal
Events† – RRs are per 3.5-fold Higher Usual Lp(a) Levels Adjusted for Cardiovascular Risk
Factors
This supplementary material has been provided by the authors to give readers additional
information about their work.
© 2007 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eAppendix 1. Description of Methods Used in the Emerging Risk Factors Collaboration
Study selection criteria Eligible prospective studies (reported as observational cohort studies, nested case-control or case-cohort subsets) had: (1) data on relevant exposure variables
available from baseline measurements; (2) at least 1 year of follow-up; (3) participants not selected on the basis of having previous cardiovascular disease; and (4) information on causespecific mortality and/or major cardiovascular morbidity collected during follow-up.1 Studies were prioritised for inclusion if they were known to have recorded >20,000 person years at
risk and were conducted in Western populations. Most studies were identified in previously published meta-analyses,2-11 with additional studies identified through updated computerassisted literature searches of databases, scanning of reference lists, hand-searching of relevant journals and correspondence with authors of relevant reports. As only a handful of
relevant studies could not provide data,12-15 >90% of relevant incident CHD cases in known Western studies are estimated to have been included.
Data collection Anonymised data were sought from collaborators on many characteristics recorded at the baseline and subsequent surveys during follow-up.1 Information on categorical
variables, such as alcohol consumption status, physical activity and smoking status, was systematically re-coded to maximise comparability among studies. For each individual, data were
sought on the following outcomes and on their dates of occurrence: non-fatal CHD; non-fatal stroke; cause-specific mortality (or at least fatal CHD and fatal stroke) and other
cardiovascular outcomes. Precise details of the diagnostic criteria used for the definition of incident cases were sought from each study (as were data on the completeness of follow-up).
Principal analyses were based on events classified according to the International Classification of Diseases (ICD) or, where this was not available, on study-specific classification systems.
Attribution of death referred to the primary cause provided (or, in its absence, the underlying cause provided). The influence on the findings of differential cause of death coding (across
cohorts, calendar year and age at death) was small. Data obtained from each participating study were checked for internal consistency and any queries then referred back, in confidence,
to the study collaborator(s), before harmonisation to a standard format. The content of the data was unchanged by this process, and computer-generated detailed summary tabulations
based on the converted data were reviewed and confirmed by collaborators. Data have been stored securely and anonymously at the coordinating centre.
Statistical methods 95% confidence intervals (CIs) and two-sided p-values were used. Studies contributing 10 or fewer outcomes to any particular analysis were excluded.
Regression analyses−The main analyses were based on Cox proportional hazards (PH) models16 estimated for each study separately, with logistic regression used for “nested” casecontrol studies (see below). The PH models were stratified by sex and, if applicable, randomised group. So for each study s=1…S, with strata k=1…Ks (for most studies Ks=2 just for the
two sexes) and individuals i=1…ns with exposure of interest Esi and other covariates Xsi, the hazard at time t after baseline were modelled as:
log(hski (t | Esi , X si )) = log h0sk (t ) + β s Esi + γ s X si .
(1)
The evolution of risk over time was thus modelled independently for each stratum in each study, as represented by the non-parametric baseline hazards h0sk(t). The βs were the
parameters of interest, being the log hazard ratios per unit increase in the exposure in study s, adjusted for the confounding effects of the covariates Xsi. These estimated log hazard
ratios were combined over studies using random-effects meta-analysis. Parallel analyses involved fixed-effect models,17-19 writing the variance of the estimated βs as vs, the randomeffects meta-analysis model is:
βˆs = β s + ε s ; where ε s ~ N(0,υs )
β s = β + ηs ; where η s ~ N(0,τ 2 ).
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
(2)
Here β is the average log hazard ratio, whose estimate combines within-study information on the relationship between exposure and risk, while allowing for heterogeneity between studies
as represented by the variance τ2, although potential sources of heterogeneity were specifically investigated (see below). A standard moment estimator of τ2 was used.20 Nested casecontrol studies were analysed with similar methods to those described above, but involved logistic regression.21 For individually-matched studies, conditional logistic regression was used,
whereas unconditional logistic regression was used in frequency-matched studies, including matching factors as covariates. Such analyses either provided estimates of hazard ratios (if
matched controls were selected to be disease-free at the time the case had an event), or odds ratios (if the selected controls were disease-free at the end of the study). Provided the
disease is relatively rare (say fewer than 10% of the study’s participants), odds ratios approximate hazard ratios and it is reasonable to combine these estimates. For nested case-cohort
studies, weighted analyses allowed for the fact that, by design, cases who were not in the randomly selected sub-cohort were also included in the analyses.22 A modified PH regression
model then provided estimates of log hazard ratios with robust standard errors.23
To investigate shape of relationships, exposure variables were divided into quantile groups based on overall distribution across studies. The hazard or odds ratios in each quantile group,
compared to the lowest group, were estimated using Cox PH regression or logistic regression in each study separately. These risk ratios were pooled across studies using multivariate
random-effects meta-analysis,24,25 and floating absolute risks were estimated,26,27 which were then plotted against the mean exposure level in each quantile group. Estimation of floating
absolute risks does not alter their values, but ascribes an appropriate variance to the log of the risk ratio for each group, including even the reference group with a risk ratio of 1 (rather
than having one group arbitrarily chosen to have a relative risk of 1 with no associated variation). This allows the values to be compared informatively (ie, with known variance) between
any pair of exposure categories, rather than only between each exposure category and the arbitrarily chosen reference group. Subsidiary analyses involved use of fractional polynomials28
to investigate curvature.
To investigate confounding, adjustment was made progressively for increasing numbers of potential confounding factors (including correction for measurement error and within-person
variation in these covariates: see below). Use of simple linear terms for age at baseline was generally sufficient, but empirical comparisons were made of alternatives as sensitivity
analyses (eg, adjustment or stratification by age categories at baseline, and inclusion of polynomial terms and interactions with other covariates, especially sex). Similar considerations
applied to adjustment for other covariates. The change in the Wald χ2 statistic provides an indication of the reduction in the evidence of association and/or increase in uncertainty
following adjustment.18,29,30
Measurement error and within-person variation−Measurement error and within-person variability in an exposure variable can cause any association of disease with the current usual level
of the exposure to be underestimated.31-33 The degree of underestimation, or regression dilution bias,31,32 was quantified by regressing serial measurements of the exposure on baseline
exposure and confounder values34 to provide a regression dilution ratio (RDR). Within-person variability and measurement error in confounders can also affect the observed exposuredisease association, and this bias may either exaggerate or obscure associations. A regression calibration technique adapted to the context of multiple studies was used to correct for
multivariate measurement error.35 For each variable measured with error, a regression calibration model incorporating between-study and between-individual heterogeneity was
constructed amongst studies s=1…S with individuals i=1…,ns providing repeat measurements r=1…rsi as:
Esir = α + (β + us )Esi + λ X si + wsi + ε sir ;
where us ~ N(0, σ u2 ), wsi ~ N(0, σ w2 ) and ε sir ~ N(0, σ e2 ),
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
(3)
and where Esir and Esi denote repeat and baseline measurements of the error-prone variable respectively, Xsi are other baseline covariates, σ u2 denotes between-study heterogeneity on the
estimated RDR value β, and σ w2 and σ e2 denote individual-specific and residual variation respectively. Empirical Bayes conditional expectations (ie, expected usual levels) for all individuals
were estimated from model (3) and used in the main analyses. The regression calibration model (3) was extended to allow for measurement error time trends, resulting in timedependent expected usual levels which were entered into a time-dependent Cox PH model. These methods assume disease risk depends on the current usual exposure (and confounder)
levels.36
Joint effects−Potential effect modifiers measured at the individual level, such as age or other risk markers, were assessed using within-study information.37,38 A 2-stage procedure was
adopted. Study-specific estimates of interaction terms δs for the potential effect modifier Xsi were estimated from model (4) and subsequently combined using random-effects metaanalysis, as in (2):
log(hski (t | E si , X si )) = log h0 sk (t ) + β s Esi + γ s X si + δ s E si X si .
(4)
The overall interaction was then based only on within-study information. Model (4) was extended to include adjustments for other confounders, and indeed their interactions with the
exposure of interest; this enabled investigation of whether a particular interaction was confounded by other main effects or interactions. Potential effect modifiers measured at the study
level, such as population type or the laboratory methods, were assessed entirely on between-study comparisons using random effects meta-regression.39 Using the estimates of βs from
(1), model (2) has been extended to include a study level covariate Xs by writing:
βˆs = β s + ε s ; where ε s ~ N(0,υs )
β s = β + δ B X s + ηs ; where η s ~ N(0,τ 2 )
(5)
where δB is the between-study interaction term, with statistical significance assessed allowing for the residual between-study heterogeneity τ 2.
Proportionality of hazards−This was evaluated in each study separately by including an interaction between the exposure and time, or by the commonly used diagnostic based on
Schoenfeld residuals.
40
which gives a χ12 statistic for each study. These independent χ12 statistics were summed across the S studies, yielding a χ s2 statistic testing the hypothesis that PH
holds in each study. However, because this approach is not a powerful test against the plausible alternative hypothesis that hazard ratios tend to decline with time in all studies, the
interaction terms between the exposure and time were pooled over studies using random-effects meta-analysis. This provided an “average” interaction term and corresponding test
statistic.
Heterogeneity and reporting biases−In addition to the standard χ 2 test for heterogeneity,41 the impact of heterogeneity was expressed in terms of I2,42 the percentage of variance in the
estimated log hazard ratios from each study that is attributable to between-study variation as opposed to sampling variation. The potential effect of study size on risk estimates was
investigated by: (i) χ 2 tests comparing studies with ≥500 cases vs those with <500 cases; (ii) meta-regression; and (iii) visual tests (forest and/or funnel plots).43
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
Assessment of cross-sectional correlates−Unadjusted Pearson correlation coefficients were pooled across cohorts by random-effects meta-analysis of Fisher’s Z transformation of cohortand sex-specific correlation coefficients.44 Associations of exposure variables with various characteristics were then assessed using a linear mixed model that included a study-level
random gradient for the relationship between the correlate and exposure of interest, but a fixed constant for each study. The main effect of cohort was modelled as a separate fixed
effect. Continuous variables were divided into tenths based on the overall distribution in males and females combined, allowing assessment of shape of associations without imposing a
priori any particular shape. Natural logarithms were used to achieve approximately symmetrical distributions for positively skewed variables. Categorical variables were modelled similarly
to the risk-factor tenths, except dummy variables were also used in the random effects equation since there was no natural monotonic ordering of the categories. From each fitted mixed
model, overall adjusted means and 95% CI by sex within the tenths of continuous markers, or category for categorical variables, were obtained. These adjusted mean values were used
to assess the shape of the association by plotting the mean (95% CI) of the exposure variable against the mean marker value within each tenth. An inverse-variance weighted polynomial
was superimposed across the adjusted means to assess whether the overall association was consistent with a linear or a quadratic shape.
Censoring of outcomes−For participants who had multiple events (eg, two CHD events at separate time points, or a CHD event followed by another type of event such as a stroke or
death from cancer), analyses in the ERFC focused on first events.1 Thus, in an analysis of CHD events, participants were followed until their first CHD event, or censored at the time of
other non-fatal cardiovascular events, such as stroke, or death from other causes. Individuals were not censored at the time of cardiovascular investigations or interventions, such as
angiography or coronary bypass operations, or at the diagnosis of angina. The rationale for this was that major cardiovascular events, such as first non-fatal MI or stroke, may disrupt the
association between baseline risk factors and subsequent disease risk. The incidence of angina and coronary interventions was, however, not recorded reliably enough in sufficient studies
to consider censoring for them. The potential biases that arise through these decisions on censoring were addressed through sensitivity analyses and by implementing alternative
censoring criteria. In general, such changes had only minimal effects.
eAppendix 1 References
1.
The Emerging Risk Factors Collaboration. The Emerging Risk Factors Collaboration: analysis of individual data on lipid, inflammatory and other markers in over 1.1 million
participants in 104 prospective studies of cardiovascular diseases. Eur J Epidemiol 2007; 22:839-869.
2.
Hokanson JE, Austin MA. Plasma triglyceride level is a risk factor for cardiovascular disease independent of high-density lipoprotein cholesterol level: a meta-analysis of populationbased prospective studies. J Cardiovasc Risk 1996; 3(2):213-219.
3.
Sarwar N, Danesh J, Eiriksdottir G et al. Triglycerides and the Risk of Coronary Heart Disease. 10 158 Incident Cases Among 262 525 Participants in 29 Western Prospective
Studies. Circulation 2006; 115(4):450-458.
4.
Thompson A, Danesh J. Associations between apolipoprotein B, apolipoprotein AI, the apolipoprotein B/AI ratio and coronary heart disease: a literature-based meta-analysis of
prospective studies. J Intern Med 2006; 259(5):481-492.
5.
Craig WY, Neveux LM, Palomaki GE, Cleveland MM, Haddow JE. Lipoprotein(a) as a risk factor for ischemic heart disease: metaanalysis of prospective studies. Clin Chem 1998;
44(11):2301-2306.
6.
Danesh J, Collins R, Peto R. Lipoprotein(a) and coronary heart disease. Meta-analysis of prospective studies. Circulation 2000; 102(10):1082-1085.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
7.
Danesh J, Whincup P, Walker M et al. Low grade inflammation and coronary heart disease: prospective study and updated meta-analyses. BMJ 2000; 321(7255):199-204.
8.
Danesh J, Wheeler JG, Hirschfield GM et al. C-reactive protein and other circulating markers of inflammation in the prediction of coronary heart disease. N Engl J Med 2004;
350(14):1387-1397.
9.
Danesh J, Collins R, Appleby P, Peto R. Association of fibrinogen, C-reactive protein, albumin, or leukocyte count with coronary heart disease: meta-analyses of prospective studies.
JAMA 1998; 279(18):1477-1482.
10.
Wheeler JG, Mussolino ME, Gillum RF, Danesh J. Associations between differential leucocyte count and incident coronary heart disease: 1764 incident cases from seven prospective
studies of 30,374 individuals. Eur Heart J 2004; 25(15):1287-1292.
11.
Gordon DJ, Probstfield JL, Garrison RJ et al. High-density lipoprotein cholesterol and cardiovascular disease. Four prospective American studies. Circulation 1989; 79(1):8-15.
12.
Nguyen TT, Ellefson RD, Hodge DO, Bailey KR, Kottke TE, bu-Lebdeh HS. Predictive value of electrophoretically detected lipoprotein(a) for coronary heart disease and
cerebrovascular disease in a community-based cohort of 9936 men and women. Circulation 1997; 96(5):1390-1397.
13.
Farquhar JW, Fortmann SP, Maccoby N et al. The Stanford Five-City Project: design and methods. Am J Epidemiol 1985; 122(2):323-334.
14.
Schaefer EJ, Lamon-Fava S, Jenner JL et al. Lipoprotein(a) levels and risk of coronary heart disease in men. The lipid Research Clinics Coronary Primary Prevention Trial. JAMA
1994; 271(13):999-1003.
15.
Frick MH, Elo O, Haapa K et al. Helsinki Heart Study: primary-prevention trial with gemfibrozil in middle-aged men with dyslipidemia. Safety of treatment, changes in risk factors,
and incidence of coronary heart disease. N Engl J Med 1987; 317(20):1237-1245.
16.
Cox DR. Regression Models and Life-Tables. JRSSB 1972; 34(2):187-220.
17.
Prospective Studies Collaboration. Collaborative overview ('meta-analysis') of prospective observational studies of the associations of usual blood pressure and usual cholesterol
levels with common causes of death: protocol for the second cycle of the Prospective Studies Collaboration. J Cardiovasc Risk 1999; 6(5):315-320.
18.
Prospective Studies Collaboration. Age-specific relevance of usual blood pressure to vascular mortality: a meta-analysis of individual data for one million adults in 61 prospective
studies. Lancet 2002; 360(9349):1903-1913.
19.
Prospective Studies Collaboration. Blood cholesterol and vascular mortality by age, sex, and blood pressure: a meta-analysis of individual data from 61 prospective studies with
55,000 vascular deaths. Lancet 2007; 370(9602):1829-1839.
20.
DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7(3):177-188.
21.
Breslow NE, Day NE. Statistical methods in cancer research. Volume I - The analysis of case-control studies. IARC Sci Publ 1980;(32):5-338.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
22.
Prentice R. A case-cohort design for epidemiologic cohort studies and disease prevention trials. Biometrika 1986; 73(1):1-11.
23.
Barlow WE, Ichikawa L, Rosner D, Izumi S. Analysis of case-cohort designs. J Clin Epidemiol 1999; 52(12):1165-1172.
24.
White IR. Multivariate random-effects meta-analysis. Stata Journal 2009; 9(1):40-56
25.
Berkey CS, Hoaglin DC, Antczak-Bouckoms A, Mosteller F, Colditz GA. Meta-analysis of multiple outcomes by regression with random effects. Stat Med 1998; 17(22):2537-2550.
26.
Easton DF, Peto J, Babiker AG. Floating absolute risk: an alternative to relative risk in survival and case-control analysis avoiding an arbitrary reference group. Stat Med 1991;
10(7):1025-1035.
27.
Plummer M. Improved estimates of floating absolute risk. Stat Med 2004; 23(1):93-104.
28.
Royston P, Ambler G, Sauerbrei W. The use of fractional polynomials to model continuous risk variables in epidemiology. Int J Epidemiol 1999; 28(5):964-974.
29.
The Fibrinogen Studies Collaboration. Plasma fibrinogen level and the risk of major cardiovascular diseases and nonvascular mortality: an individual participant meta-analysis. JAMA
2005; 294(14):1799-1809.
30.
The Fibrinogen Studies Collaboration. Correcting for multivariate measurement error by regression calibration in meta-analyses of epidemiological studies. Statistics in Medicine
2009;28:1067-92
31.
Clarke R, Shipley M, Lewington S et al. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol 1999;
150(4):341-353.
32.
MacMahon S, Peto R, Cutler J et al. Blood pressure, stroke, and coronary heart disease. Part 1, Prolonged differences in blood pressure: prospective observational studies corrected
for the regression dilution bias. Lancet 1990; 335(8692):765-774.
33.
The Fibrinogen Studies Collaboration. Regression dilution methods for meta-analysis: assessing long-term variability in plasma fibrinogen among 27 247 adults in 15 prospective
studies. Int J Epidemiol 2006; 35(6):1570-1578.
34.
Rosner B, Willett WC, Spiegelman D. Correction of logistic regression relative risk estimates and confidence intervals for systematic within-person measurement error. Stat Med
1989; 8(9):1051-1069.
35.
Rosner B, Spiegelman D, Willett WC. Correction of logistic regression relative risk estimates and confidence intervals for measurement error: the case of multiple covariates
measured with error. Am J Epidemiol 1990; 132(4):734-745.
36.
Frost C, White IR. The effect of measurement error in risk factors that change over time in cohort studies: do simple methods overcorrect for 'regression dilution'? Int J Epidemiol
2005; 34(6):1359-1368.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eAppendix 2. Rationale for Using Non-HDL Cholesterol Rather Than Calculated LDL Cholesterol in the Current Analyses
Because direct measurement of low density lipoprotein cholesterol (LDL-C) has been relatively uncommon in long-term prospective studies, most studies have tended to use the
Friedewald equation1 to estimate LDL-C values from the measured concentrations of total cholesterol (TC), high density lipoprotein cholesterol (HDL-C), and triglycerides (divided by a
constant):
Calculated LDL-C = TC – HDL-C – (triglycerides/2.2)
(1)
where triglycerides/2.2 approximates the concentration of cholesterol carried in very low density lipoprotein (VLDL-C) when the units of measurement are mmol/l (if mg/dl are used, the
constant is 5.0).
Non-HDL-C (calculated as the difference between TC and HDL-C) can be substituted into equation 1:
Calculated LDL-C = non-HDL-C – (triglycerides/2.2)
(2)
As a consequence, any regression model (eg, Cox proportional hazards model for survival data, or logistic regression model for case-control data) that concomitantly includes calculated
LDL-C, HDL-C and triglycerides, is simply a mathematical rearrangement of a model that includes non-HDL-C, HDL-C and triglycerides, as shown below.
Consider a survival model for the log hazard ratio (HR) including non-HDL-C, HDL-C and triglycerides (TG), where the mutually adjusted coefficients for each term are given by β1, β2 and
β3, respectively:
logHR = β1 non-HDL-C + β2 HDL-C + β3 TG
(3)
Adding and subtracting (β1/2.2) TG and simplifying yields:
logHR = β1 non-HDL-C + β2 HDL-C + β3 TG + (β1/2.2 – β1/2.2) TG
= β1 (non-HDL-C – TG/2.2) + β2 HDL-C + (β3 + β1/2.2) TG
Substituting calculated LDL-C for (non-HDL-C – TG/2.2) as in equation 2 gives:
logHR = β1 calculated LDL-C + β2 HDL-C + (β3 + β1/2.2) TG
(4)
Comparing equations 3 and 4 demonstrates that:
•
the calculated LDL-C parameter (β1 in equation 4) equals the non-HDL-C parameter (β1 in equation 3) in any model that also includes HDL-C and triglycerides. Indeed, any of
the coefficients in equation 4 can be calculated from those in equation 3, should the need arise.
•
when calculated LDL-C, HDL-C and triglycerides are included in the same model (equation 4), the coefficient for triglycerides is biased by β1/2.2 (or β1/5.0 if mg/dl are used)
compared to equation 3. This means that even if triglycerides concentration was not associated with the outcome of interest in equation 3 (ie. β3 = 0), then it would appear to
have an association of (β1/2.2) when adjusted for calculated LDL-C and HDL-C in equation 4.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eAppendix 3. List of Acronyms for Studies Included in the Current Report
AFTCAPS (Air Force/Texas Coronary Atherosclerosis Prevention Study)
ARIC (Atherosclerosis Risk in Communities Study)
ATTICA (Attica Study)
BRHS (British Regional Heart Study)
BRUN (Bruneck Study)
BUPA (British Union Provident Association)
CHARL (Charleston Heart Study)
CHS (Cardiovascular Health Study)
COPEN (Copenhagen City Heart Study)
DUBBO (Dubbo Study of the Elderly)
EAS (Edinburgh Artery Study)
FIA (First Myocardial Infarction in Northern Sweden)
FINRISK 92 (Finrisk Cohort 1992)
FLETCHER (Fletcher Challenge Blood Study)
FRAMOFF (Framingham Offspring Cohort)
GOH (The Glucose Intolerance, Obesity and Hypertension Study)
GOTO33 (Göteborg Study 1933)
GRIPS (Göttingen Risk Incidence and Prevalence Study)
HPFS (Health Professionals Follow-up Study)
KIHD (Kuopio Ischaemic Heart Disease Study)
MRFIT (Multiple Risk Factor Intervention Trial 1)
NHANES III (National Health and Nutrition Examination Survey III)
NHS (Nurses’ Health Study)
NPHS II (Northwick Park Heart Study II)
PRIME (Prospective Epidemiological Study of Myocardial Infarction)
PROCAM (Prospective Cardiovascular Münster Study)
QUEBEC (Quebec Cardiovascular Study)
REYK (Reykjavik Study)
SHS (Strong Heart Study)
TARFS (Turkish Adult Risk Factor Study)
ULSAM (Uppsala Longitudinal Study of Adult Men)
USPHS (U.S. Physicians Health Study)
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
WHITE II (Whitehall II Study)
WHS (Women’s Health Study)
WOSCOPS (West of Scotland Coronary Prevention Study)
ZUTE (Zutphen Elderly Study)
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eAppendix 3. List of Acronyms for Studies Included in the Current Report
AFTCAPS (Air Force/Texas Coronary Atherosclerosis Prevention Study)
ARIC (Atherosclerosis Risk in Communities Study)
ATTICA (Attica Study)
BRHS (British Regional Heart Study)
BRUN (Bruneck Study)
BUPA (British Union Provident Association)
CHARL (Charleston Heart Study)
CHS (Cardiovascular Health Study)
COPEN (Copenhagen City Heart Study)
DUBBO (Dubbo Study of the Elderly)
EAS (Edinburgh Artery Study)
FIA (First Myocardial Infarction in Northern Sweden)
FINRISK 92 (Finrisk Cohort 1992)
FLETCHER (Fletcher Challenge Blood Study)
FRAMOFF (Framingham Offspring Cohort)
GOH (The Glucose Intolerance, Obesity and Hypertension Study)
GOTO33 (Göteborg Study 1933)
GRIPS (Göttingen Risk Incidence and Prevalence Study)
HPFS (Health Professionals Follow-up Study)
KIHD (Kuopio Ischaemic Heart Disease Study)
MRFIT (Multiple Risk Factor Intervention Trial 1)
NHANES III (National Health and Nutrition Examination Survey III)
NHS (Nurses’ Health Study)
NPHS II (Northwick Park Heart Study II)
PRIME (Prospective Epidemiological Study of Myocardial Infarction)
PROCAM (Prospective Cardiovascular Münster Study)
QUEBEC (Quebec Cardiovascular Study)
REYK (Reykjavik Study)
SHS (Strong Heart Study)
TARFS (Turkish Adult Risk Factor Study)
ULSAM (Uppsala Longitudinal Study of Adult Men)
USPHS (U.S. Physicians Health Study)
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
WHITE II (Whitehall II Study)
WHS (Women’s Health Study)
WOSCOPS (West of Scotland Coronary Prevention Study)
ZUTE (Zutphen Elderly Study)
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eTable 1. Baseline Characteristics, Blood Handling, Storage and Assay Characteristics of Studies Contributing Data to the Current Analysis
Study
Country
Isoform
sensitivity
Population source / Baseline
sampling
year
Fasting
status†/
duration
Blood
sample
Storage
duration
Storage
temperature
Assay method
(source)
Assay
Standard
Antibody used
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Fated /
> 8 hrs
Fasted /
> 8 hrs
NS
serum
NS
NS
NS
NS
NS
NS
ELISA
(In-house)
ITA (NS)
In-house
Anti-apo(a) PAb
No
NS
NS
NS
ELISA
(Immuno)
ELISA
(Terumo)
ELISA
(Genetech)
ITA
(DAKO)
ELISA
(Biopool)
ELISA
(Biopool)
IRMA
(Pharmacia)
ITA
(DiaSorin SPQIII)
ITA
(K-Assay)
ELISA
(Immuno)
IRMA
(Pharmacia)
ELISA (Strategic
Diagnostics)
ELISA
(Biopool)
ELISA
(In-house)
EID
(Behringwerke)
ELISA
(Biopool)
ELISA
(Terumo)
INA
(Behring)
IRMA (Pharmacia)
In-house
No
Manufacturer
C: Anti-apo(a) PAb
D: Anti-apo(a) Ab
D: anti-Lp(a) MAb
Yes
In-house
D: Anti apo(a) MAb
Yes
Manufacturer
Rabbit anti-Lp(a) Pab
Yes
Manufacturer
C: Anti-Lp(a) PAb
D: Anti-Lp(a) MAb
C: Anti-Lp(a) PAb
D: Anti-Lp(a) MAb
Two site anti-apo(a)
MAb
NS
No
Cohort studies
AFTCAPS
USA
ARIC 2001
USA
ATTICA
Greece
Popln. screening /
1990-93
complete
Household
1987-89
listings/Random
Popln. register / Random 2000-1
BRUN 1999
Italy
Popln. register / Random 1990
CHARL
USA
CHS1 2003
USA
Household listing /
Random
Medicare lists / Random
COPEN 2008 Denmark
2
DUBBO 2002 Australia
27
EAS
2001
Scotland
1960-61
1989-93
Popln. register / Random 1991-94
Electoral roll / Complete
GP list / Random
1988-89
1987-88
FINRISK92
2005
FRAMOFF
1996
GOH
Finland
Israel
Offspring & spouse to
1991-95
FHS / Complete
Popln. register / Random 1969-73
GRIPS 1997
Germany
Occupational / Complete 1982
KIHD
Finland
Popln. register / Random 1984-89
NHANES3
USA
Census list / Cluster
USA
Popln. register / Random 1992
1988-1994
NPHSII 2001 UK
GP list / Complete
1989-94
PRIME 2002 France /
Ireland‡
PROCAM
Germany
1996
QUEBEC
Canada
1998
SHS 2002
USA
General Popln. / Quota
1991-94
TARFS
Turkey
ULSAM
Sweden
WHITE2
WHS 2006
UK
USA
WOSCOPS
2000
UK
Occupational / Complete 1975-2001
Popln. register / Random 1985-86
Tribal rolls / Complete
1989-92
Household listings /
Random
Popln. screening /
complete
Civil servant / Complete
Health professionals /
Complete
Heart screening clinic /
Complete
1990
1970-74
1985-88
1993-2004
1989-91
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
0
Plasma
1 week -1 yr
Frozen, -70 C
Serum
< 1 week
Fresh
Plasma
<1 week
Fresh
NS
1-5 yrs
Frozen, -70 0C
Fasted /
> 8 hrs
Non-fasted
Plasma
1 week-1 yr
Serum
1 week-2 yrs
Frozen, -80 0C
Fasted /
> 8 hrs
Fasted/
> 8 hrs
Fasted /
4-8 hrs
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Fasted /
4-8 hrs
Fasted /
> 8 hrs
Fasted /
> 6 hrs
Non-fasted
Serum
< 1 week
Fresh
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Fasted /
. 8 hrs
Fasted / NS
3/4 Fasted / >8
hrs
Fasted /
> 8 hrs
Serum
Serum
5-10 yr
1 week- 1 yr
0
Frozen, -50 C
0
Frozen, -70 C
0
Plasma
1–5 yrs
Frozen, -80 C
Plasma
<1 week
Fresh
Serum
5-10 yrs
Frozen, -90 0C
Serum
2-6 yrs
Frozen, -20 0C
Serum
1 week–1 yr
Frozen, -20 0C
0
Serum
1 week-1 yr
Frozen, -80 C
Plasma
<1 week
Fresh
Serum
<1 week
Fresh
0
Plasma
5-10 yrs
Frozen, -70 C
Plasma
NS
NS
Plasma
NS
Frozen, -75 0C
Serum
> 10 yrs
Frozen, -150 0C
Serum
Plasma
NS
>10 yrs
Frozen, -80 0C
Frozen, -150 0C
Plasma
1-5 yrs
Frozen, -70 0C
ITA (NS)
ITA
(Denka Seiken)
ELISA (Innogenetics)
Manufacturer
Manufacturer
Manufacturer
Manufacturer
No
No
Yes
Yes
Manufacturer
Goat anti-Lp(a)
antisera
C: Anti-apo(a) PAb
D: Anti-apo(a) MAb
Two site anti-apo(a)
MAb
C: Anti-Lp(a) MAb
D: Anti-Lp(a) PAb
C: Anti-Lp(a) MAb
D: Anti-Lp(a) PAb
C: Anti-Apo(a) Mab
D: Anti-ApoB MAb
Rabbit anti-Lp(a)
antisera
C: Anti-Lp(a) MAb
D: Anti-Lp(a) PAb
C: MAb; D: Pab
NS
NS
No
Manufacturer
Two site anti-apo(a)
MAb
NS
Anti-Lp(a) PAb
No
Manufacturer
Manufacturer
Manufacturer
Manufacturer
NS
Immuno
CDC
NS
Manufacturer
NS
C: Anti-apo(a) Mab
D: Anti-apoB PAb
No
No
Yes
No
No
No
No
Yes
Yes
No
No
ZUTE
The
Popln. register / Random 1990
Netherlands
Non-fasted
Serum
NS
Frozen, -20 0C
NS
Fasted / NS
Serum
>10 yrs
Frozen, -40 0C
ELISA
Nested case-control studies (individually matched)
BUPA 1994
FIA 1998
UK
Sweden
Medical center list /
1975-82
Complete
Popln. register / Random 1985-99
FLETCHER
2007
HPFS
New Zealand Occupational , electoral
1992-94
roll / complete, random
USA
Occupational / Complete 1994
MRFIT 2001
USA
NHS 2005
USA
Popln. screening /
1973-76
Complete
Occupational / Complete 1990
Fasted /
4 hrs
Non-fasted
Plasma
Plasma
6-19 yrs
UK
GOTO33 1993 Sweden
REYK 2008
Iceland
USPHS 1993
USA
GP lists / Random
1978-80
Popln. register /
1983-84
Complete
Popln. register /
1967-91
Complete
Occupational / Complete 1982
Frozen, -80 C
0
>10 yrs
Frozen, -70 C
2/3 Fasted / NS Plasma
5 -10 yrs
Frozen , -130 0C
Fasted /
> 8 hrs
Fasted /
Variable
Plasma
>10 yrs
Frozen, -50 0C
Plasma
5-10 yrs
Frozen , -130 0C
Serum
>10 yrs
Frozen, -20 0C
Nested case-control studies (frequency matched)
BRHS
0
Non-fasted
Fasted /
> 8 hrs
Fasted /
> 8 hrs
Non-fasted
0
Serum
5-10 yrs
Frozen, -70 C
Serum
>10 yrs
Frozen, -20 0C
Plasma
>10 yrs
Frozen, -80 0C
†Fasting status at blood sampling; ‡Northern Ireland; Popln. indicates Population; NS indicates not stated
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
NS
(Biopool) NS
ELISA
(Hypehn Biomed)
ELISA
(Hyphen Biomed)
ITA
(Denka Seiken)
ELISA (Strategic
Diagnostics)
ITA
(Denka Seiken)
ELISA
(Hyphen Biomed)
ELISA
(Biopool)
ELISA
(Hyphen Biomed)
ELISA
(Biopool)
In-house
Manufacturer
Manufacturer
NS
NS
C: Anti-Lp(a) Pab
D: Anti-Lp(a) MAb
Mono-specific antiapo(a) PAb
C: Anti-apo(a) MAb
D: Anti-apoB PAb
Anti-Lp(a) PAb
No
No
No
No
Yes
Manufacturer
C: Anti-Lp(a) Mab
D: Anti-Lp(a) PAb
Anti-Lp(a) PAb
Manufacturer
NS
No
Manufacturer
C: Anti-Lp(a) PAb
D: Anti-Lp(a) MAb
C: Anti-apo(a) MAb
D: Anti-apoB PAb
C: Anti-Lp(a) Mab
D: Anti-Lp(a) Pab
No
Manufacturer
Manufacturer
Manufacturer
No
No
No
eFigure 6. Risk ratios for coronary heart disease per 3.5 fold higher usual Lp(a) levels, by strata of various study level
characteristics.
Variable/
subgroup
No. of
studies
No. of
individuals
RR (95% CI) Heterogeneity
p-value
No. of
cases
Ethnicity†
White
Black
Other
29
3
2
95753
4546
4207
7540
261
426
1.14 (1.09, 1.19)
1.05 (0.90, 1.23)
1.36 (0.88, 2.09)
0.52
18
10
2
54301
49981
2363
5138
2819
405
1.16 (1.10, 1.23)
1.08 (1.02, 1.15)
1.14 (1.02, 1.28)
0.39
17
13
45650
60995
4631
3731
1.17 (1.10, 1.24)
1.09 (1.04, 1.15)
0.18
11
19
24517
82128
2332
6030
1.17 (1.07, 1.28)
1.11 (1.07, 1.15)
0.26
4
6
16
13409
32128
48530
533
2009
5182
1.18 (1.07, 1.30)
1.07 (1.02, 1.13)
1.14 (1.07, 1.22)
0.48
4
15
8
13409
71359
13486
533
3710
3103
1.18 (1.07, 1.30)
1.16 (1.08, 1.24)
1.09 (1.02, 1.17)
0.50
20
4
4
58455
38284
8776
6483
861
955
1.13 (1.07, 1.20)
1.18 (1.06, 1.32)
1.11 (1.03, 1.19)
0.75
22
6
2
79749
25766
1130
6561
1738
63
1.14 (1.09, 1.20)
1.11 (1.01, 1.22)
1.10 (0.82, 1.46)
0.83
Geographical region
Western Europe
North America
Other
Blood sample
Serum
Plasma
Fasting status
Non-fasted
Fasted
Storage duration†
< 1 wk
1wk-1yr
> 1yr
Storage temperature†
Fresh
-70 or less
-20 to -70
Assay method†
ELISA
ITA/INA
OTHER
Isoform sensitivity‡
insensitive
sensitive
unknown
.75
1
1.5
2.5
RR per 3.5 fold higher Lp(a) levels
CI indicates confidence interval. Sizes of data markers are proportional to the inverse of the variance of the risk ratios. Risk ratios are adjusted for age, and usual levels of systolic
blood pressure, smoking status, history of diabetes, body mass index and total cholesterol, and stratified, where appropriate, by sex and study group. †Although a total of 30 studies
have contributed to the analyses, for different characteristics different number of studies had relevant data; for Storage duration 4 studies, for Storage temperature 3 studies, and for
Assay method 2 studies did not have relevant data. For Ethnicity, the no. of studies add to 34 because 4 studies contributed to 2 categories. ‡Isoform sensitivity refers to whether the
result of an assay is affected by apolipoprotein(a) isoform variation.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eTable 3. Results From Quadratic Models for the Association Between Usual Lp(a) Levels and
the Risk of CHD
Study
No. of
observations
No. of
cases
Association with CHD risk:
RR (95% CI) per 3.5 fold higher level
Main effect
Quadratic term
Cohort studies
902
21
1.15(0.72,1.84)
14033
850
1.15(1.05,1.25)
798
53
1.10(0.74,1.63)
165
19
0.73(0.37,1.45)
3860
592
1.01(0.91,1.13)
7487
283
1.16(1.03,1.32)
2008
273
1.26(1.07,1.47)
637
54
1.41(0.97,2.07)
2201
92
1.03(0.80,1.33)
2850
109
1.35(1.03,1.76)
5784
299
1.31(1.13,1.51)
1996
386
1.01(0.90,1.15)
4496
107
1.26(0.99,1.61)
2375
157
1.12(0.92,1.37)
7441
115
1.44(1.12,1.86)
3198
94
1.67(1.07,2.60)
2012
53
1.15(0.81,1.64)
3837
416
1.11(0.97,1.27)
1400
3
0.82(0.21,3.12)
1866
485
1.14(1.03,1.26)
7903
170
1.27(1.03,1.57)
27791
227
0.92(0.78,1.09)
4617
299
1.02(0.89,1.18)
305
42
1.11(0.75,1.66)
Nested case-control studies (individually matched)
BUPA 1994
1184
208
1.55(1.24,1.93)
FIA 1998
1454
510
1.34(1.10,1.64)
FLETCHER 2007
372
134
1.31(0.94,1.82)
HPFS
691
220
1.15(0.94,1.40)
MRFIT 2001
736
246
0.90(0.73,1.10)
NHS 2005
687
234
1.27(1.00,1.61)
Nested case-control studies (frequency matched)
BRHS
1561
461
1.06(0.91,1.23)
GOTO33 1993
128
16
1.28(0.66,2.46)
REYK 2008
5771
1850
1.30(1.21,1.39)
USPHS 1993
652
243
1.19(0.97,1.46)
Overall†
123,198 ‡
9321
1.16 (1.11,1.21)
AFTCAPS
ARIC 2001
BRUN 1999
CHARL
CHS1 2003
COPEN 2008
DUBBO 2002
EAS 2001
FINRISK92 2005
FRAMOFF 1996
GRIPS 1997
KIHD
NHANES3
NPHSII 2001
PRIME 2002
PROCAM 1996
QUEBEC 1998
SHS 2002
TARFS
ULSAM
WHITE2
WHS 2006
WOSCOPS 2000
ZUTE
1.04(0.73,1.47)
1.05(0.96,1.14)
1.05(0.76,1.43)
0.66(0.29,1.52)
0.98(0.87,1.10)
1.12(1.01,1.23)
0.91(0.79,1.05)
0.80(0.51,1.25)
1.05(0.82,1.33)
0.98(0.76,1.27)
1.14(1.04,1.23)
1.18(1.06,1.32)
1.02(0.79,1.32)
1.04(0.81,1.35)
0.90(0.74,1.10)
0.84(0.60,1.18)
0.95(0.66,1.38)
1.01(0.94,1.07)
1.49(0.31,7.16)
1.09(1.00,1.19)
1.03(0.89,1.20)
1.33(1.15,1.53)
1.00(0.87,1.16)
1.06(0.72,1.55)
1.03(0.82,1.28)
1.27(0.95,1.69)
1.02(0.76,1.38)
0.93(0.76,1.14)
0.97(0.80,1.17)
1.00(0.83,1.20)
1.06(0.97,1.16)
2.01(0.95,4.22)
1.00(0.93,1.09)
1.06(0.84,1.33)
1.05 (1.02,1.08)*
Risk ratios are adjusted for baseline age and, where appropriate, stratified by sex and study group. † Overall effect
calculated by combining study specific estimates for main effect and quadratic terms for the loge Lp(a) using
multivariate random-effects meta-analysis. ‡Overall number is less than study total because 2 studies (ATTICA and
GOH) did not contribute CHD endpoints.*P-value for comparison of the linear versus the quadratic model = 0.003.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eTable 4. Risk Ratios for CHD for 3.5-fold (ie, 1-SD) Higher Usual Levels of Lp(a), Further Adjusted for Usual Values of Various Potential
Confounding Factors
Subset
No. of
studies
No. of
subjects
No. of
cases
26
96675
5728
1.13
1.15
1.14
1.17
1.19)
1.21)
1.20)
1.24)
23
25
25
30
Plus Non-HDL-C & HDL-C
Plus Apo-B & Apo-AI
15
75560
3540
1.21 (1.12 , 1.31)
1.18 (1.09 , 1.27)
22
18
Fibrinogen†
Basic ‡ Plus Non-HDL-C & HDL-C
Further adjustment for fibrinogen
19
87708
4227
1.16 (1.09 , 1.24)
1.13 (1.06 , 1.21)
20
12
CRP†
Basic ‡ Plus Non-HDL-C & HDL
Further adjustment for CRP
19
55146
3375
1.09 (1.04 , 1.15)
1.09 (1.04 , 1.13)
11
13
Lipid markers†
Adjustment
Basic ‡ Plus total cholesterol
Basic ‡ Plus Non-HDL-C & HDL-C
Basic ‡ Plus Non-HDL-C, HDL-C & log-triglycerides
Basic ‡ Plus total cholesterol corrected for Lp(a)
cholesterol§
Apolipoproteins† Basic
Basic
‡
‡
RR (95% CI)
Wald
χ12
‡
(1.08
(1.09
(1.08
(1.11
,
,
,
,
†Analysis was restricted to participants with complete information on sex, study group and respective confounding variables for each subset. Basic adjustment includes age and
usual values of systolic blood pressure, smoking, history of diabetes and body mass index. Risk ratios are stratified by sex and study group where appropriate. Studies with fewer
than 10 events were excluded from these analyses. § Correction for the cholesterol content of Lp(a) particles was made by subtracting estimated Lp(a) cholesterol values from
total cholesterol; Lp(a) cholesterol was estimated from Lp(a) total mass using the following equation: Lp(a) cholesterol (mg/dl) = 0.15*Lp(a) (mg/dl)+1.24 (Clinical Chemistry
1998; 44(8):1629-40)
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eTable 5. Parallel Analyses of the Association of Lp(a) With Disease Risk: (a) for Comparison of
Individuals in Extreme Thirds of Baseline Level Distributions, and (b) per 3.5-fold (ie, 1-SD) Higher
Baseline Level
Coronary heart disease
106645 individuals 8362 cases 30 cohorts
I2
Risk ratio
2
Wald χ1
(95% CI)
(95% CI)
Ischaemic stroke
69539 individuals 1684 cases 13 cohorts
I2
Risk ratio
2
Wald χ1
(95% CI)
(95% CI)
a) Top vs. bottom thirds of Lp(a) distribution
With adjustment for...
Age & sex
1.33 (1.23 , 1.45)
Plus systolic blood pressure
1.34 (1.23 , 1.45)
Plus smoking status
1.33 (1.23 , 1.45)
Plus history of diabetes
1.35 (1.24 , 1.47)
Plus body mass index
1.35 (1.25 , 1.47)
Plus total cholesterol
1.27 (1.17 , 1.38)
46
47
45
49
50
33
42
42
44
45
42
40
(11 , 63)
(11 , 63)
(13 , 63)
(15 , 64)
(10 , 63)
(6 , 62)
1.19
1.17
1.17
1.19
1.19
1.18
(1.01
(1.01
(1.01
(1.02
(1.03
(1.02
,
,
,
,
,
,
1.41)
1.36)
1.36)
1.38)
1.38)
1.37)
5
4
4
5
6
5
17 (0 , 56)
0 (0 , 57)
0 (0 , 57)
0 (0 , 57)
0 (0 , 57)
0 (0 , 57)
b) Per 3.5 fold higher baseline
With adjustment for...
Age & sex
Plus systolic blood pressure
Plus smoking status
Plus history of diabetes
Plus body mass index
Plus total cholesterol
33
34
32
35
38
25
62
62
63
64
63
59
(43
(44
(45
(46
(46
(39
1.08
1.07
1.08
1.08
1.09
1.08
(1.00
(1.00
(1.00
(1.01
(1.02
(1.01
,
,
,
,
,
,
1.17)
1.15)
1.15)
1.16)
1.16)
1.16)
4
4
4
6
6
5
52
38
37
33
31
35
Lp(a) level
1.13
1.14
1.13
1.14
1.15
1.11
(1.09
(1.09
(1.09
(1.09
(1.10
(1.07
,
,
,
,
,
,
1.18)
1.19)
1.18)
1.19)
1.20)
1.16)
,
,
,
,
,
,
74)
75)
75)
75)
75)
73)
Analyses were restricted to participants with complete information on sex and all confounding variables. Risk ratios are stratified by sex
and study group where appropriate. Studies with less than 10 events were excluded from analysis. Note: I2 is a measure of consistency
across studies: the percentage of variance in estimated log RRs that is attributable to between study variations as opposed to sampling
variation. Values of I2 close to 0 indicate lack of evidence of heterogeneity
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
(9
(0
(0
(0
(0
(0
,
,
,
,
,
,
74)
68)
67)
65)
64)
66)
eFigure 1. Study specific adjusted risk ratios for CHD, corresponding the adjusted risk ratio in Table 2.
Cohort
No. of
individuals
GOTO33
BUPA
QUEBEC
AFTCAPS
ZUTE
EAS
BRUN
NHANES3
FINRISK
PROCAM
PRIME
FLETCHER
NPHSII
WHITE2
WHS
USPHS
NHS
MRFIT
COPEN
DUBBO
GRIPS
WOSCOPS
KIHD
ULSAM
FIA
SHS
BRHS
CHS1
ARIC
REYKCON
126
110
617
826
304
622
798
2457
2190
3185
7431
368
2367
7720
22667
515
609
736
7288
1995
5783
4617
1981
1420
1083
3728
1553
3837
13989
5723
No. of
cases
RR (95% CI)
16
19
20
21
42
53
53
60
92
94
114
133
157
168
206
210
218
246
269
272
299
299
383
386
394
407
458
588
843
1842
1.64
1.76
0.81
1.19
1.05
1.27
1.03
1.07
1.05
1.22
1.28
1.13
1.09
1.32
1.05
1.15
1.32
0.89
1.14
1.15
1.61
1.02
1.08
1.12
1.21
1.19
1.01
1.03
1.07
1.14
(0.82,
(1.00,
(0.51,
(0.72,
(0.74,
(0.95,
(0.72,
(0.80,
(0.85,
(0.95,
(1.05,
(0.92,
(0.92,
(1.10,
(0.92,
(0.94,
(1.09,
(0.71,
(1.01,
(1.01,
(1.40,
(0.92,
(0.96,
(1.01,
(1.03,
(1.04,
(0.88,
(0.93,
(1.00,
(1.09,
3.27)
3.13)
1.29)
1.96)
1.50)
1.70)
1.47)
1.43)
1.31)
1.56)
1.56)
1.39)
1.30)
1.59)
1.20)
1.41)
1.61)
1.12)
1.28)
1.31)
1.86)
1.13)
1.22)
1.24)
1.43)
1.37)
1.17)
1.14)
1.15)
1.20)
Overall (random-effects)
1.13 (1.09, 1.18)
Overall (fixed-effect)
1.13 (1.10, 1.15)
.5
1
2
4
RR per 3.5 fold higher usual Lp(a) levels
CI indicates confidence interval. Sizes of data markers are proportional to the inverse of the variance of the risk ratios. The overall adjusted RR in studies with greater
than 500 CHD cases (1.09, 1.03-1.16) was not significantly different from that of studies with less than 500 cases (1.15, 1.09-1.21) (heterogeneity p-value=0.36).
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 2. Mean Lp(a) levels by cohort and (a) assay method principle, or (b) whether the assay
method used was sensitive to apolipoprotein(a) isoform variation.
Other
WHS
NHS
ATTICA
HPFS
FRAMOFF
COPEN
GOH
WHITE2
PROCAM
ULSAM
KIHD
32
ITA / INA
NS
4
8
16
ELISA
FINRISK92
AFTCAPS
ZUTE
GRIPS
PRIME
FLETCHER
ARIC
WOSCOPS
BUPA
QUEBEC
FIA
NHANES3
TARFS
BRUN
CHS1
NPHSII
GOTO33
DUBBO
CHARL
SHS
REYK
BRHS
USPHS
EAS
MRFIT
2
Geometric mean Lp(a), mg/dl (95% CI) - log scale
(A) Assay method
32
sensitive
NS
4
8
16
Not sensitive
COPEN
GOH
NHANES3
WHITE2
AFTCAPS
ATTICA
ZUTE
CHARL
CHS1
FRAMOFF
WOSCOPS
BUPA
QUEBEC
FIA
MRFIT
SHS
DUBBO
GRIPS
WHS
NHS
PRIME
FINRISK92
HPFS
FLETCHER
ARIC
NPHSII
TARFS
GOTO33
KIHD
BRUN
REYK
BRHS
ULSAM
USPHS
EAS
PROCAM
2
Geometric mean Lp(a), mg/dl (95% CI) - log scale
(B) Isoform sensitivity
CI indicates confidence interval; NS indicates not specified; ELISA: Enzyme Linked Immunosorbent Assay; ITA: Immunoturbidimetric Assay;
INA: Immunonephelometric Assay;. Error bars indicate 95% CI of the mean Lp(a) level in each study. Meta-regression showed that there was
no statistically significant difference in mean loge Lp(a) between groups of studies defined by (a) assay method (p=0.10), or (b) isoform
sensitivity (p= 0.98)
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 3. Within-person variability in lipoprotein(a) levels (regression dilution ratios† stratified by study and repeat).
Regression dilution ratio and 95% CI
1
FLETCHER
AFTCAPS
TARFS
COPEN
REYK
ULSAM
TARFS
0.8
PROCAM
0.6
PROCAM
0.4
0.2
No. of repeats = 6597
Mean RDR† = 0.87 (95% CI, 0.81-0.93)
0
0
3
6
9
12
15
18
21
24
Time since baseline (years)
RDR indicates regression dilution ratio; CI confidence interval. † The values provided are estimates of RDR at mean log-Lp(a) concentration,
as the models allowed variation to vary by level. RDRs are adjusted for age and sex. Data shown for repeat measures involving more than
25 individuals. The solid and broken lines indicate the overall RDR and its 95%CI, respectively. Sizes of data markers are proportional to the
inverse of the variance of the study specific RDRs.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 4. Age- and sex- adjusted risk ratios for various vascular and non-vascular endpoints per 3.5 fold
higher usual Lp(a) levels.
Endpoint
No of
studies
No. of
individuals
No. of
Cases
RR (95% CI)
33
121633
9314
1.17 (1.13, 1.22)
Coronary death†
27
107873
2523
1.20 (1.13, 1.27)
Non-fatal MI†
29
113892
6632
1.16 (1.11, 1.21)
Ischaemic stroke
14
78466
1890
1.10 (1.02, 1.19)
Unclassified stroke
12
51305
720
1.03 (0.93, 1.15)
Haemorrhagic stroke
9
62040
305
1.12 (0.97, 1.29)
Non-vascular deaths
29
117237
8094
0.98 (0.94, 1.01)
23
105716
3945
1.00 (0.96, 1.04)
Smoking-related cancer deaths
17
67963
1481
1.00 (0.95, 1.06)
Other cancer deaths
22
105239
2452
1.01 (0.94, 1.08)
24
114220
4101
0.95 (0.89, 1.00)
Non-fatal MI and coronary death
All tumour deaths
Other non-vascular deaths
.8
1
1.2
1.4
1.6
Risk ratio per 3.5 fold higher Lp(a) level
CI indicates confidence interval. Sizes of data markers are proportional to the inverse of the variance of the risk ratios. Risk ratios are adjusted for baseline age and stratified, where
appropriate, by sex and study group. Studies involving fewer than 10 cases of any outcome were excluded from the analysis of that outcome. † These subtotals do not add to the total
number of CHD outcome in the first row because some nested case-control studies did not subdivide outcomes into coronary death or non-fatal MI
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 5. Risk ratios for coronary heart disease by fifths of usual Lp(a) levels, after excluding the first 5 years of follow-up.
b) Further adjustment†
a) Adjustment for age and sex only
Non-fatal MI and coronary death: 5237 Cases
Risk ratio and 95% CI (log scale)
Non-fatal MI and coronary death: 5717 Cases
2
2
1.8
1.8
1.6
1.6
1.4
1.4
1.2
1.2
1
1
.9
.9
3
6
12
24
48
96
Usual Lp(a) (mg/dl)
Geometric mean (log scale)
192
3
6
12
24
48
96
192
Usual Lp(a) (mg/dl)
Geometric mean (log scale)
CI indicates confidence interval. CIs were calculated using floating absolute risk technique. †Further adjustment for systolic blood pressure, smoking
status, history of diabetes, body mass index and total cholesterol. Sizes of data markers are proportional to the inverse of the variance of the risk
ratios.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 6. Risk ratios for coronary heart disease per 3.5 fold higher usual Lp(a) levels, by strata of various study level
characteristics.
Variable/
subgroup
No. of
studies
No. of
individuals
RR (95% CI) Heterogeneity
p-value
No. of
cases
Ethnicity†
White
Black
Other
29
3
2
95753
4546
4207
7540
261
426
1.14 (1.09, 1.19)
1.05 (0.90, 1.23)
1.36 (0.88, 2.09)
0.52
18
10
2
54301
49981
2363
5138
2819
405
1.16 (1.10, 1.23)
1.08 (1.02, 1.15)
1.14 (1.02, 1.28)
0.39
17
13
45650
60995
4631
3731
1.17 (1.10, 1.24)
1.09 (1.04, 1.15)
0.18
11
19
24517
82128
2332
6030
1.17 (1.07, 1.28)
1.11 (1.07, 1.15)
0.26
4
6
16
13409
32128
48530
533
2009
5182
1.18 (1.07, 1.30)
1.07 (1.02, 1.13)
1.14 (1.07, 1.22)
0.48
4
15
8
13409
71359
13486
533
3710
3103
1.18 (1.07, 1.30)
1.16 (1.08, 1.24)
1.09 (1.02, 1.17)
0.50
20
4
4
58455
38284
8776
6483
861
955
1.13 (1.07, 1.20)
1.18 (1.06, 1.32)
1.11 (1.03, 1.19)
0.75
22
6
2
79749
25766
1130
6561
1738
63
1.14 (1.09, 1.20)
1.11 (1.01, 1.22)
1.10 (0.82, 1.46)
0.83
Geographical region
Western Europe
North America
Other
Blood sample
Serum
Plasma
Fasting status
Non-fasted
Fasted
Storage duration†
< 1 wk
1wk-1yr
> 1yr
Storage temperature†
Fresh
-70 or less
-20 to -70
Assay method†
ELISA
ITA/INA
OTHER
Isoform sensitivity‡
insensitive
sensitive
unknown
.75
1
1.5
2.5
RR per 3.5 fold higher Lp(a) levels
CI indicates confidence interval. Sizes of data markers are proportional to the inverse of the variance of the risk ratios. Risk ratios are adjusted for age, and usual levels of systolic
blood pressure, smoking status, history of diabetes, body mass index and total cholesterol, and stratified, where appropriate, by sex and study group. †Although a total of 30 studies
have contributed to the analyses, for different characteristics different number of studies had relevant data; for Storage duration 4 studies, for Storage temperature 3 studies, and for
Assay method 2 studies did not have relevant data. For Ethnicity, the no. of studies add to 34 because 4 studies contributed to 2 categories. ‡Isoform sensitivity refers to whether the
result of an assay is affected by apolipoprotein(a) isoform variation.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 7. Direct comparison of adjusted risk ratios for CHD between Lp(a) and non-HDL-C, for a 1SD higher† baseline or usual levels
RR (95% CI)
Baseline
Lp(a)
1.11 (1.05, 1.17)
Non-HDL-c
1.48 (1.34, 1.64)
Usual
Lp(a)
1.14 (1.09, 1.20)
Non-HDL-c
1.66 (1.40, 1.96)
.75
1
1.5
2.5
RR per 1 SD higher level
CI indicates confidence interval. Sizes of data markers are proportional to the inverse of the variance of the risk ratios.* † RRs presented
are for 1-SD increase in loge Lp(a) or non-HDL-C levels. Analyses were based on data from 26 cohorts involving 97,049 and 5766 cases.
Risk ratios were mutually adjusted for each other, and baseline age, and usual levels of systolic blood pressure, smoking status, history of
diabetes, body mass index and HDL-C
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014
eFigure 8. Association of Lp(a) with fatal vascular and non-vascular outcomes in analyses that did not censor for nonfatal
events† – RRs are per 3.5 fold higher usual Lp(a) levels adjusted for cardiovascular risk factors.
Endpoint
No. of
studies
No. of
individuals
No. of
cases
RR (95% CI)
Coronary death
26
101604
3575
1.15 (1.08, 1.22)
Fatal MI
25
95915
1744
1.20 (1.10, 1.30)
All vascular deaths
28
103452
5553
1.10 (1.06, 1.14)
Non-vascular deaths
26
103170
8390
1.01 (0.99, 1.04)
All tumour deaths
21
92326
3945
1.00 (0.97, 1.04)
Smoking-related cancer deaths
16
63555
1576
0.99 (0.94, 1.05)
Other cancer deaths
20
92209
2357
1.03 (0.97, 1.09)
22
100378
4409
1.02 (0.98, 1.07)
Other non-vascular deaths
.8
1
1.2
1.4
1.6
Risk ratio per 3.5 fold higher Lp(a) level
CI indicates confidence interval. Sizes of data markers are proportional to the inverse of the variance of the risk ratios †Compared to the corresponding main analyses (Figure 2), analyses that
did not censor for non-fatal events involved additional 1917 vascular and 1122 non-vascular fatal outcomes. Risk ratios are adjusted for baseline age, smoking status, systolic blood pressure,
history of diabetes, body mass index and total cholesterol, and stratified, where appropriate, by sex and study group. Studies involving fewer than 10 cases of any outcome were excluded
from the analysis of the outcome.
Downloaded From: http://jama.jamanetwork.com/ on 06/09/2014