Lipoprotein(a) and cardiovascular disease: sifting the evidence to guide future research

Abstract

Lipoprotein(a) (Lp(a)) is a genetically determined causal risk factor for cardiovascular disease including coronary heart disease, peripheral arterial disease, ischaemic stroke, and calcific aortic valve stenosis. Clinical trials of specific and potent Lp(a)-lowering drugs are currently underway. However, in clinical practice, widespread assessment of Lp(a) is still lacking despite several guideline recommendations to measure Lp(a) at least once in a lifetime in all adults to identify those at high or very high risk due to elevated levels. The present review provides an overview of key findings from observational and genetic Lp(a) studies, highlights the main challenges in observational Lp(a) studies, and proposes a minimum set of requirements to enhance the quality and harmonize the collection of Lp(a)-related data. Adherence to the recommendations set forth in the present manuscript is intended to enhance the quality of future observational Lp(a) studies, to better define thresholds for increased risk, and to better inform clinical trial design. The recommendations can also potentially assist in the interpretation and generalization of clinical trial findings, to improve care of patients with elevated Lp(a) and optimize treatment and prevention of cardiovascular disease.

Graphical Abstract

Open in new tabDownload slide

See the editorial comment for this article ‘Integrating lipoprotein(a) into preventive cardiology: probably important to get the measurement right', by S.J. Nicholls, https://doi.org/10.1093/eurjpc/zwae037.

Introduction

Elevated lipoprotein(a) (Lp(a)) is an under-recognized inherited cardiovascular (CV) risk factor that affects about one in five people.^1–3 Over 1.4 billion people globally with elevated Lp(a) are at increased risk of CV disease (CVD),⁴ but most clinicians are unaware of their patients’ Lp(a) levels. Reduction of elevated Lp(a) levels has the potential to reduce the risk of atherosclerotic CVD (ASCVD).^1,2

Lipoprotein(a) research is a rapidly evolving field, and this white paper reviews the key observational and genetic findings on the association of Lp(a) with the risk of disease based on data not collected as part of conventional randomized controlled trials, also commonly known as real-world data.^5,6 The review highlights the main challenges in conducting observational Lp(a) studies and proposes a minimum set of requirements to enhance the quality and harmonize the collection of Lp(a)-related data.

Lipoprotein(a) and cardiovascular risk

Lipoprotein(a) consists of a cholesterol-rich low-density lipoprotein particle containing an additional apolipoprotein, the distinctive and plasminogen-like apolipoprotein(a) (apo(a)), covalently attached to apolipoprotein B.⁷ The principal determinant (70% to ≥90%) of Lp(a) plasma levels is variation in the LPA gene coding for apo(a). The kringle IV type 2 (KIV2) copy number repeat polymorphism determines apo(a) isoform size, which correlates inversely with plasma Lp(a) levels.⁸ Elevated Lp(a) is considered the most common genetically inherited risk factor for CVD.⁹ In contrast to lifestyle risks, this genetic risk is present throughout a person’s lifetime and cannot be sufficiently reduced by lifestyle changes or currently approved therapies.

The effect of Lp(a) on CV risk is independent of conventional CV risk factors, including hypertension, smoking, diabetes, obesity, low-grade inflammation, and high LDL cholesterol (LDL-C).^1,10 Lipoprotein(a) levels can vary by a factor of 1000 between adults.⁷ The concentration distribution of Lp(a) is skewed; two-thirds of the general population have Lp(a) levels that confer low CV risk, 1 in 5 have levels associated with increased risk, and 1 in 20 have levels that confer high or very high risk.⁹ The 2019 consensus statement from HEART UK⁹ categorized the impact of CV risk associated with Lp(a) as minor [32–90 nmol/L (18–40 mg/dL)], moderate [90–200 nmol/L (40–90 mg/dL)], high [200–400 nmol/L (90–180 mg/dL)], and very high [>400 nmol/L (>180 mg/dL)]. The recently updated European Atherosclerosis Society (EAS) consensus paper refines the interpretation of the continuous relationship between Lp(a) concentration and absolute ASCVD risk by estimating how much Lp(a) levels increase individuals’ overall risk of ASCVD.²

Clinical manifestations of elevated Lp(a) include coronary heart disease (CHD), ischaemic stroke, peripheral arterial disease (PAD), and calcific aortic valve stenosis (CAVS).² The proposed mechanisms by which elevated Lp(a) leads to ASCVD and CAVS include proatherogenic, prothrombotic, and proinflammatory effects.^2,7 More specifically, endothelial dysfunction, lipid deposition, and calcification may promote ASCVD and CAVS, partly via oxidized phospholipids on the Lp(a) particle.

Substantial variation in the global prevalence of elevated Lp(a) levels of >50 mg/dL exists (Figure 1).^3,4,11 Individuals with ASCVD are more likely to have elevated Lp(a) than those without ASCVD. A large UK study of >460 000 middle-aged individuals found that 20% of individuals with ASCVD had Lp(a) levels of >150 nmol/L (≍70 mg/dL) vs. 12% of individuals without ASCVD.³ Individuals of White, South Asian, and Black origin in that study had ≥90th percentile Lp(a) thresholds of 168, 140, and 212 nmol/L, with even more pronounced variation in median levels (19, 31, and 75 nmol/L, respectively).^3,11

No specific Lp(a)-lowering pharmacological treatment has yet been approved, but observational studies, including Mendelian randomization studies, suggest that clinical benefit would require a reduction in Lp(a) of 50 mg/dL (105 nmol/L) to 100 mg/dL.^12–14 Extensive clinical research is ongoing into targeted Lp(a) reduction, with active or completed randomized controlled trials of Lp(a) in Phase 2 (NCT02160899, NCT03070782, NCT05537571, and NCT01890967) and two trials in Phase 3 [Lp(a)HORIZON (NCT04023552) and OCEAN(a)¹⁵ (NCT05581303)].

Without specific targeted Lp(a)-lowering treatments, treatment recommendations for individuals with elevated Lp(a) focus on overall CV risk reduction. More intense LDL-C lowering and control of dyslipidaemia and other ASCVD risk factors have been suggested for patients with elevated Lp(a).⁹ Risk associated with elevated Lp(a) is not yet captured by established CV risk assessment models like SCORE, SMART, and the American Heart Association (AHA)/American College of Cardiology (ACC) ASCVD 10-year risk score/calculator.^16–18 However, the updated 2022 EAS Lp(a) consensus paper presents a new model that estimates how much Lp(a) level increases an individual’s overall risk of ASCVD by considering both Lp(a) and baseline absolute global risk of ASCVD.² Including Lp(a) in this model leads to a substantial upward reclassification of total ASCVD risk.

Clinical lipoprotein(a) guidelines

Key Lp(a) guidelines and consensus statements have been published by several national and international medical societies, including the AHA/ACC,¹ American Association of Clinical Endocrinologists/American College Of Endocrinology, European Society of Cardiology (ESC)/EAS, EAS/European Federation of Clinical Chemistry and Laboratory Medicine,¹⁹ National Lipid Association, HEART UK,⁹ Nouvelle Société Francophone d'Athérosclérose, Canadian Cardiovascular Society, Lipid Association of India, Polish Lipid Association, Brazilian Society of Cardiology, and Beijing Heart Society (Figure 2; references included in figure legend).

Figure 2 shows that despite variations amongst guidelines, Lp(a) testing is now recommended by most contemporary guidelines in individuals with a history of premature ASCVD (personal or familial) and in individuals with familial hypercholesterolaemia. Guidelines are gradually transitioning from recommending targeted Lp(a) testing of high-risk individuals to screening of all adults.

Units used to report thresholds for elevated Lp(a) vary, with some guidelines only reporting Lp(a) thresholds in milligrams per decilitre, others only in nanomoles per litre,⁹ but most reporting thresholds in both units.¹⁹ An expert statement from Chinese cardiologists and Canadian guidelines have the lowest Lp(a) threshold expressed in milligrams per decilitre (30 mg/dL), while American guidelines have opted for a higher threshold [50 mg/dL (100–125 nmol/L)]. Despite the impact of ethnicity, Lp(a) < 75 nmol/L is considered normal, while Lp(a) > 125 nmol/L negatively impacts CV risk. Lipoprotein(a) thresholds expressed in nanomoles per litre range from >90 nmol/L in UK guidelines⁹ to ≥125 nmol/L in French guidelines and ≥125 nmol/L (≥50 mg/dL) in ACC/AHA guidelines (Figure 2). For primary prevention patients, European guidelines state that Lp(a) measurement should be considered at least once in each adult’s lifetime. This would identify those with very high inherited Lp(a) levels (>175–180 mg/dL or >378–430 nmol/L) with a lifetime risk of ASCVD equivalent to the risk associated with heterozygous familial hypercholesterolaemia.^2,20 Likewise, the updated 2022 EAS Lp(a) consensus report also clearly recommends measuring Lp(a) at least once in all adults to identify those at high CV risk.²

Observational and genetic evidence landscape in lipoprotein(a) research

Beyond clinical trials: potential insights from observational data

Good quality observational and genetic observational data and evidence may provide alternative and complementary hypothesis-generating insights beyond randomized controlled trials into the effect of Lp(a) on CV risk, disease burden, and healthcare utilization. They would also assist with generalizing information provided by randomized controlled trial findings. Hence, this would help address limitations of randomized controlled trials, such as strict inclusion criteria for patient selection, duration, and operational complexity, to improve healthcare decision-making.^6,21

Observational data based on measurements made in the clinical or community care setting²¹ are particularly important in Lp(a) research for understanding the prevalence and patterns of elevated Lp(a) levels in diverse populations and the associated risk of CVD. Observational population-based data with long-term follow-up may allow us to evaluate the chronic real-life burden of genetically determined, lifelong elevated Lp(a).

Genetic studies, e.g. in the form of Mendelian randomization studies, provide evidence of causality between elevated Lp(a) and disease. Mendelian randomization studies rely on the random allocation of genetic variants at conception to ensure even distributions of known and unknown confounders between comparison groups defined by genotypes.²² Thus, studies demonstrating that LPA genetic variants strongly associated with Lp(a) levels are also associated with CV risk provide evidence of causality. The observed associations are likely free of confounding and cannot result from reverse causality, as disease cannot alter inherited genotypes.²²

By including so-called instrumental variable analyses, Mendelian randomization studies can also provide an estimate of the magnitude of the causal effect of elevated Lp(a) levels. This can guide the design of randomized controlled trials by providing estimates of the magnitude of Lp(a) lowering needed to achieve a meaningful clinical benefit.²³ The highly genetically determined levels of Lp(a) make Lp(a) an ideal candidate for Mendelian randomization studies, and large genetic epidemiologic studies have provided evidence of causality for the association of elevated Lp(a) levels with CHD, CAVS, PAD, heart failure, and CV and all-cause mortality.^1,2

Evolving lipoprotein(a) research over time

Interest in Lp(a) has expanded in recent years, as demonstrated by the increasing volume of Lp(a) research published recently (Figure 3).

Following the discovery of Lp(a) by Kåre Berg in 1963,²⁴ research into Lp(a) was limited until 1987, the year the LPA gene was cloned (Figure 3).²⁵ The finding and the demonstration in 1990 that increased Lp(a) was an independent risk factor for early myocardial infarction (MI) caused interest in Lp(a) to grow.²⁶ However, this increased interest in Lp(a) was quickly dampened by the first prospective studies with negative results, likely due to shortcomings of the assays and the use of long-term frozen samples.²⁷ These early studies negatively impacted research with a subsequent reduction in Lp(a)-focused research and publications, and many clinical laboratories abandoned routine measurement of Lp(a).

The year 2009 was a milestone for Lp(a) research following the publication of the Emerging Risk Factor Collaboration meta-analysis of observational Lp(a) studies and simultaneous publication of two large genetic studies providing evidence of causality.^28,29 First, a classic Mendelian randomization study of >40 000 individuals demonstrated an increased risk of MI in persons with elevated Lp(a) and a corresponding low number of LPA KIV2 repeats.^22,29 Second, a large case-control study including 3100 CHD cases genotyped for 49 000 variants in 2100 candidate genes identified two LPA single-nucleotide polymorphisms (SNPs), associated with elevated Lp(a), as having the strongest association with CHD risk of all SNPs tested.^29,30

In 2013, another key genetic study, using a hypothesis-free genome-wide association approach, identified an LPA SNP associated with elevated Lp(a) to be strongly associated with aortic valve calcification, leading to a higher frequency of CAVS.³⁴ In a subsequent Mendelian randomization study, elevated Lp(a) levels and corresponding LPA risk genotypes were associated with a clear stepwise increase in risk of CAVS in the Danish general population.³¹ Individuals in the top 10% of the concentration distribution had a two-to-three-fold increase in risk of CAVS, comparable with the risk for CHD. Subsequent Mendelian randomization studies conducted in the same population have also demonstrated an increased risk of heart failure, ischaemic stroke, and CV and total mortality in individuals with elevated Lp(a) and corresponding LPA risk genotypes.³¹

In the last 5 years, three further Mendelian randomization studies have investigated the degree of Lp(a) lowering needed to reduce CV risk, with an estimated reduction of 50–100 mg/dL needed to obtain short-term clinical benefit.^12–14 The strong genetic evidence for elevated Lp(a) being a causal risk factor for CVD has prompted randomized controlled trials to investigate novel treatments for elevated Lp(a), like antisense oligonucleotides and small interfering RNA therapeutics specifically targeting hepatic LPA gene expression.^32,33,35

Recent insights from large-scale observational studies

Recent large observational studies of Lp(a) have assessed the prevalence of elevated Lp(a) in patients with established CVD and the magnitude and shape of ASCVD risk associated with Lp(a).³ For example, Lp(a)HERITAGE (NCT03887520), a global multi-centre, cross-sectional epidemiological study, estimated the prevalence of elevated Lp(a) in 49 001 patients with established ASCVD, based on current and historical measurements from local laboratories. The results of the Lp(a)HERITAGE study indicate that Lp(a) is infrequently measured in patients with ASCVD (14%).¹¹ Furthermore, the study showed that in patients with ASCVD, Lp(a) is highest in Black, younger, and female patients. Notably, more than 28% of this global ASCVD population had levels exceeding 50 mg/dL, a common threshold for increased risk.¹¹

Another comprehensive study of 500 000 UK Biobank participants (primary and secondary prevention) followed over 11 years recently investigated the magnitude and shape of ASCVD risk associated with a range of Lp(a) concentrations measured using a Denka Seiken-developed Lp(a) assay with traceability to an internationally recommended calibrator. The study also examined risk variation across racial and clinical subgroups and the clinical importance of a threshold for elevated Lp(a) to guide future development of Lp(a)-reducing therapy.³ The analysis by Patel et al.³ found a linear relationship between Lp(a) and ASCVD, with a hazard ratio of 1.11 (95% confidence interval, 1.10–1.12) per 50 nmol/L increment.

Major challenges in conducting observational studies of lipoprotein(a)

Observational Lp(a) studies can suffer from a variety of methodological weaknesses and biases. Depending on the study type, many of these may be common to observational studies³⁶ or observational studies using Mendelian randomization.³⁷ Specific limitations affecting Lp(a) research will be discussed in this section, such as selection bias, poor definition of study population and outcomes, inconsistency in Lp(a) measurements, variability in applied Lp(a) thresholds, small sample sizes, and limitations in the applied statistical methods.

Selection bias

Currently, Lp(a) measurement is infrequently included as part of routine clinical practice as demonstrated by a recent US study where <1% of a young ASCVD patient population assessed from 2001 to 2016 had been tested for Lp(a) prior to a premature MI, increasing to 21% post-MI.³⁸ If Lp(a) is measured, it is normally measured for a specific reason, e.g. secondary care investigation for hyperlipidaemia and/or CV symptoms. In a German study of 15 381 patients hospitalized with CHD, the frequency of Lp(a) screening was 20% in 5725 patients with premature CHD (mean age, 50.5 ± 7.2 years) compared with 5% in 9656 patients with late-onset CHD (mean age, 69.4 ± 7.4 years).³⁸ Thus, using Lp(a) measurement as an inclusion criterion may introduce bias towards capturing a cohort of individuals with higher CV risk and consequently higher Lp(a) levels.³ Of note, this selection bias likely does not affect Lp(a) studies conducted in a general population setting, where, in contrast, a healthy participant bias may occur.

Patient population/cohort and outcomes definition

An important source of variability among observational studies of Lp(a) is inconsistent definitions of ASCVD in primary vs. secondary prevention cohorts. While in general, ASCVD encompasses MI, ischaemic stroke, transient ischaemic attack, revascularization, PAD, and unstable or stable angina, diverse definitions and use of diagnosis codes in different studies impact the ASCVD patient populations captured. Likewise, definitions of outcomes for major adverse cardiovascular events (MACEs) may differ between studies. For instance, in their studies of Lp(a), Patel et al. assessed recurrent ASCVD, but excluded PAD and death,³ whereas Madsen et al.¹⁴ examined MACE including death. Inadequately defined outcomes in observational Lp(a) studies may limit the interpretation of results, making studies unsuitable for pooled analyses or meta-analyses.

Furthermore, the comparability of cohorts and outcomes between studies is affected by patient population (primary vs. secondary prevention). Outcomes in a primary prevention cohort with no history of ASCVD events may differ from those in a secondary prevention cohort with pre-existing ASCVD. In fact, even within secondary prevention, the different definitions of newly diagnosed ASCVD (incident) vs. established ASCVD cohorts before study inclusion (prevalent) impact comparability. The latter might bring about immortal time bias and survival bias.

A thorough patient history is important for interpreting Lp(a) findings in a clinical context. Median Lp(a) has been reported to be up to almost twice as high in patients with heterozygous familial hypercholesterolaemia than in the general population, and elevated Lp(a) is an independent predictor of CVD risk in patients with familial hypercholesterolaemia.³⁸ Of note, as the Lp(a) cholesterol content confounds the LDL-C measurement in elevated Lp(a) and as elevated Lp(a) confers increased ASCVD risk, some patients are diagnosed with clinical familial hypercholesterolemia (FH) for that reason.³⁹ Furthermore, Lp(a) can be raised in chronic kidney disease and certain inflammatory conditions, like rheumatoid arthritis, psoriasis, and systemic lupus erythematosus.⁴⁰ Patient selection from populations with these co-morbidities may also be a source of selection bias.

Ethnicity

Although Lp(a) levels vary little between men and women⁴¹ and remain relatively stable throughout life,⁴¹ they do differ according to race and ethnicity (Figure 1).⁴ East Asians have been found to have the lowest Lp(a) levels, followed by Hispanics, Whites, South Asians, and Blacks. In the UK Biobank study, Blacks had the highest median Lp(a) levels (75 nmol/L) followed by South Asians (31 nmol/L), Whites (19 nmol/L), and Chinese (16 nmol/L).³ It has been shown that inter-racial variance in LPA gene locus size and SNPs account for ethnic differences in Lp(a). Given these variations, knowing the racial and ethnic composition of mixed study populations is key to appropriately analysing and interpreting results.

Lipoprotein(a) measurement

Methods

Lipoprotein(a) size heterogeneity has made developing accurate Lp(a) immunoassays a challenge.⁴² Lipoprotein(a) size heterogeneity is caused by the variable number of kringle-shaped domains in apo(a), resulting in apo(a) isoforms with molecular masses ranging from ≍300 to ≍800 kDa. Thus, apo(a) contains a single plasminogen-like kringle V domain and 10 types of plasminogen-like KIV domains, where the number of copies of the KIV2 domain varies between 3 and >40 determining isoform size.⁹ Isoform size correlates inversely with hepatic production rates and consequently with Lp(a) plasma concentrations. Polyclonal antibodies typically used in immunoassays are effective at recognizing the repetitive KIV structure of apo(a), but they may bias measurements.⁴³ Levels of large isoforms, with many binding sites generating high signal strength, may be overestimated. Likewise, levels of small isoforms, with fewer binding sites, may be underestimated (Figure 4). Considerable isoform-dependent measurement bias (>100% positive and >50% negative bias) may bias risk associated with high Lp(a) towards underestimation in epidemiologic studies.²

Available immunochemical methods include enzyme-linked immunosorbent assay, immunonephelometry, immunoturbidimetry, and dissociation-enhanced lanthanide fluorescent immunoassay.⁴² A common problem with Lp(a) assays has been the lack of standardization, i.e. traceability to a common calibrator or reference material with an assigned Lp(a) target value, a prerequisite for ensuring accuracy and for comparing measurements from different assays (Figure 5). The lack of Lp(a) assay standardization may explain the highly variable median Lp(a) levels reported by different studies conducted in populations of similar ethnicity.¹² One of the difficulties with immunochemical methods for measuring Lp(a) is assay calibration.

Physicochemical methods, such as mass spectrometry-based approaches, measure Lp(a) without isoform size bias by detecting unique peptide fragments of apo(a). However, these methods are scarce, labour-intensive, and time-consuming compared with immunochemical methods.⁴⁴

Today, commercial assays using robust and precise turbidimetric methods based on the Denka reagent (Denka, Japan) are available from different manufacturers. Denka-based assays are considered the least isoform-sensitive and most reliable of commercially available methods suitable for high-throughput analysis.⁴² Denka-based assays use polyclonal antibodies directed at apo(a) and are thus by definition subject to apo(a) isoform size-dependent bias. However, calibrators with different isoform composition for low and high Lp(a) levels can minimize such potential bias.⁴² Depending on the calibrators supplied, Lp(a) concentrations in these assays may be traceable to the World Health Organization (WHO)/International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) primary reference material SRM-2B (with an assigned target value in nanomoles) used to standardize methods for analysing Lp(a).⁴²

Lipoprotein(a) stability

Cardiovascular risk assessment is complicated by the instability of Lp(a) in stored samples. Lipoprotein(a) degrades in frozen samples over time, even when preserved at very low temperatures (−70°C),⁴⁵ and small Lp(a) isoforms may degrade more quickly than large isoforms. Since individuals with elevated Lp(a) levels generally have small isoforms, degradation during storage is expected to have a correspondingly larger impact on samples from individuals with elevated Lp(a) levels.⁴² Studies measuring Lp(a) retrospectively in frozen samples stored for a long period of time (>3–5 years) might therefore underestimate true Lp(a) levels and CV risk.

Lipoprotein(a) units

Currently, Lp(a) measurements are reported using two different unit systems, mass (mg/dL, mg/L, or g/L), and molar (nmol/L), where the latter is the ideal. The selection of units partly reflects test endorsement by national health authorities (e.g. the US Food and Drug Administration has to date only cleared mg/dL).⁴⁶ However, for clinical purposes, either unit may be used, as both can determine if patients have low, medium, or high Lp(a) levels, with appropriately calibrated assays. Nevertheless, to optimize the scientific reporting of Lp(a), it is important to understand these two different units.

As Lp(a) has no single, defined molecular weight, converting from mass units to molar concentrations, or from molar concentrations to mass units, is generally imprecise and unreliable. Estimated conversion factors vary between assays and thus Lp(a) measurements should always be reported in the same unit as the assay calibrators.⁴² The Canadian Society of Clinical Chemists recommends quantifying apo(a) in molar units (nmol/L) and specifying the Lp(a) assay used in the laboratory report.⁴⁷ Moving worldwide reporting of Lp(a) levels away from mass units towards molar concentrations will eventually improve the clinical interpretation of Lp(a)-mediated risk.⁴⁸

Lipoprotein(a) thresholds

The categorization of Lp(a) levels as ‘elevated’ in international guidelines is complicated by either the absence of cut-offs or the existence of different cut-offs for ‘elevated’ Lp(a).⁹ When present, not only do cut-offs vary in terms of thresholds but also in terms of units measured. This reflects the current lack of Lp(a) assay standardization, which hampers the direct comparison of measurements from different studies using different Lp(a) assays.⁹ As a result, different thresholds are used for ‘elevated’ Lp(a) and for ‘normal’ Lp(a).

Data analysis considerations

Sample size

As with any observational study, appropriate sample sizes are needed to derive meaningful results from statistical analyses. In the case of Lp(a), large sample sizes are required to study associations between Lp(a) concentrations and outcomes. Since Lp(a) is not normally distributed in most ethnicities, almost any sample population of individuals with Lp(a) measurements will be skewed towards lower Lp(a) concentrations. Hence, large sample sizes are not only needed to demonstrate substantial increases in risk at very high Lp(a) levels but also subtle increases in risk at moderately elevated Lp(a) levels.

Investigators should be mindful that large sample sizes are not per se able to overcome all methodological limitations. For instance, if Lp(a) measurements are derived from severely isoform size–biased assays and/or long-term stored samples, confidence intervals will narrow, but risk estimate bias towards the null will likely not be offset. Genetic epidemiological studies with large populations, like the Copenhagen General Population Study and UK Biobank Study, have been at the forefront of addressing the sample size limitations of Lp(a) research.^3,14,34

Statistical analyses

Estimating risk associated with Lp(a) poses challenges due to the highly right-skewed distribution of Lp(a) levels, and analyses that focus on per-unit increase, high vs. low, or quantile comparisons may dilute the increased risk associated with the highest extremes of the Lp(a) distribution. Estimating risk associated with appropriate Lp(a) thresholds (high levels, e.g. top 1%, 5%, or 10%, vs. low levels, e.g. bottom 50%, 33%, or 25%) is therefore crucial for identifying the true risk of extremely high Lp(a) levels.

This skewed distribution also necessitates the use of non-parametric or semi-parametric tests and models, such as the Cox proportional hazards model, which allow hazard ratios of multiple factors to be evaluated simultaneously.³⁴ These models require that all covariates be independent (collinearity between covariates can bias risk estimates) and that their effects are constant over time. As Lp(a) is an independent risk factor, and LPA genotypes are independent of known and unknown potential confounders, hazard ratios that describe the risk of outcome associated with Lp(a) should not be greatly affected when models are adjusted for other risk factors, such as conventional CV risk factors (e.g. lipid levels, hypertension, and smoking). However, when adjusting for total cholesterol, LDL-C, and/or apolipoprotein B, the LDL component of serum Lp(a) levels can introduce collinearity between cholesterol and Lp(a) covariates, and the Lp(a) risk estimates may be underestimated. This potential non-independence between cholesterol and Lp(a) poses a challenge for the correct adjustment of models.

Key recommendations for enhancing observational studies and future research

Key recommendations for future observational studies

Key recommendations on how to address these central challenges in conducting observational studies of Lp(a) are summarized in Table 1. Selection bias in observational Lp(a) studies could be reduced by implementing ESC-recommended once-in-a-lifetime measurement of Lp(a), and the association of Lp(a) with outcomes could be strengthened by accurately defining the cohorts and outcomes examined. Cohort extraction algorithms need to be described in sufficient detail for results to be replicated, the pros and cons of definitions assessed, and definitions compared with those of other studies. Major adverse cardiovascular events should be defined in line with the definitions used in major outcome trials to allow comparison between clinical trials and observational studies. Including large numbers of specific ethnicities in Lp(a) studies or demonstrating the absence of variance between White and specific ethnic groups could mitigate the current underrepresentation of ethnic and racial groups in Lp(a) research.

Lipoprotein(a) measurement poses challenges related to methods, stability, units, and thresholds. Concerns about the isoform sensitivity of commercially available Lp(a) assays should not stop clinicians from measuring Lp(a), as the best quality commercial assays are able to accurately identify individuals with low, medium, or high Lp(a). To minimize assay-related Lp(a) variation, researchers should prioritize assays that are traceable to recommended calibrators and governments should provide incentives for assay manufacturers to move towards this goal. Using fresh samples to measure Lp(a) should be encouraged, as Lp(a) degrades over time when frozen. Otherwise, storage conditions should be validated.

Lipoprotein(a) should preferably be measured in nanomoles per litre using a standardized assay avoiding the use of conversion factors to change molar measurements into mass measurements or vice versa. Until a consensus has been reached on a threshold for elevated Lp(a), assay-specific cut-offs for the 80th percentile of the general population, as described in the 2019 HEART UK consensus statement, could be used.⁹ An alternative approach, detailed in the 2022 EAS consensus statement, is to use thresholds above which (>125 nmol/L or >50 mg/dL) or below which (<75 nmol/L or <30 mg/dL) CV risk is ruled in or out.²

Because there are disproportionately fewer individuals with elevated Lp(a), sample sizes should allow the assessment of risk in individuals with the highest Lp(a) levels and the highest risk of CVD and appropriate statistical methods to analyse non-parametric distributions used (Table 1). Further, assessment of conventional CV risk factors should be performed in study populations, not only to adequately control for known confounders in multivariable-adjusted analyses but also to enable stratified analyses and identification of subgroups with the highest absolute risk of CV events.

Lastly, consideration should be given to the cholesterol content of Lp(a) when adjusting Lp(a) analyses for total cholesterol or LDL-C. Lipoprotein(a) cholesterol content is inevitably included in these measurements, regardless of whether LDL-C is calculated from total cholesterol or measured directly. While estimates of the cholesterol content of Lp(a) vary considerably, a conservative LDL-C correction of 15–30% of total Lp(a) mass seems reasonable for population-based studies.⁴³ Sensitivity analyses where no correction is performed should be included.

Future research: limitations of current research/research gaps

Important gaps exist in the current evidence base for Lp(a). To confirm the causal association between Lp(a) and the risk of incidence and progression of ASCVD and CAVS determined in Mendelian randomization studies, CV outcome trials of new Lp(a)-lowering agents will be needed. These studies could also enhance treatment cost-effectiveness by identifying patients most likely to benefit from treatment. High-quality epidemiological studies are needed in non-White populations to address the lack of ethnicity data, in particular what constitutes elevated Lp(a) across Black, Hispanic, and South Asian populations.² With regard to Lp(a) stability, it is still unclear how quickly different isoforms of Lp(a) degrade over time when frozen, and a properly controlled study to investigate this would be useful. The observed association between low levels of Lp(a) and the risk of diabetes needs to be explored, in particular what drives this (causality, reverse causality, co-morbidities, and/or medication).⁴⁹ Lastly, our incomplete comprehension of the genetic regulation of Lp(a) in diverse ethnicities underlines the need to identify causal variants along with how these variants might affect Lp(a) levels or Lp(a) pathogenicity.²

Conclusion

Our knowledge and understanding of Lp(a) have come a long way since the discovery of Lp(a) by Kåre Berg nearly 60 years ago. The finding that Lp(a) is the single most common genetically inherited risk factor for CVD has made it a key focus of contemporary research interest, a fact reflected in the newest national and international CV guidelines. Epidemiological studies of Lp(a) are particularly susceptible to the negative influences of inadequate sample size, selection bias, and poor assay design. Better awareness of these epidemiological research challenges can help investigators design observational Lp(a) studies that mitigate their effects, leading to better Lp(a) data quality and research in the future. Adherence to the recommendations set forth can enhance the ability of future observational and genetic epidemiologic studies to positively inform clinical trial design, potentially leading to better care of patients with elevated Lp(a) and better overall treatment and prevention of CVD.

Acknowledgements

The authors would like to thank John Plant (Evidera) and Stephen Gilliver (Evidera) for medical writing support, Richard Leason (Evidera) and Hanah Byrne (Novartis) for graphic design support, and Gergana Kucheva (Evidera) for project management support.

Author contributions

P.R.K., R.D.G.N., S.N., A.H., U.L., M.C.-S., C.A., and B.G.N. contributed to the conception of this review. P.R.K., R.D.G.N., and M.C.-S. conducted the literature review and the interpretation of data for the different chapters. P.R.K., R.D.G.N., and M.C.-S. also drafted the manuscript. S.N., U.L., A.H., B.G.N., and C.A. critically revised the manuscript. All gave final approval and agreed to be accountable for all aspects of work ensuring integrity and accuracy.

Funding

This work was supported by Novartis Pharma AG, Basel, Switzerland.

Data availability

No data were generated or analysed for or in support of this paper.

References

Author notes