Clinical Utility of Lipoprotein(a) and LPA Genetic Risk Score in Risk Prediction of Incident Atherosclerotic Cardiovascular Disease

Key Points Question Does measurement of lipoprotein(a) and/or LPA genetic risk score (GRS) have clinical utility in risk prediction of incident atherosclerotic cardiovascular disease (ASCVD)? Findings In this cohort of 283 540 adults recruited by the UK Biobank, both measured lipoprotein(a) and LPA GRS were associated with comparable risk of incident ASCVD events. The LPA GRS did not yield additional prognostic information beyond measured lipoprotein(a), and both measured lipoprotein(a) and LPA GRS yielded modest improvements in the discrimination of ASCVD risk relative to the Pooled Cohort Equations or QRISK3. Meaning Cardiovascular risk assessment with lipoprotein(a) may be achieved with either direct measurement or an LPA GRS.

L ipoprotein(a) is a plasma lipoprotein composed of a lowdensity lipoprotein (LDL) particle that is covalently linked to apolipoprotein(a) by a disulfide bond. On an equimolar basis, lipoprotein(a) is more atherogenic than LDL because the additional apolipoprotein(a) component may exacerbate atherothrombosis by promoting vascular inflammation and its potential antifibrinolytic activity is associated with inhibition of plasminogen. [1][2][3] Lipoprotein(a) levels are 75% to 95% heritable and predominately determined by singlenucleotide variants at the LPA gene and copy number variants specifically in the kringle IV type 2 domain. 4-6 Genetic association studies have identified genetic variants explaining approximately 60% of the variability in lipoprotein(a) levels in European populations. 7,8 Elevated lipoprotein(a), defined as lipoprotein(a) level of 120 nmol/L or greater or approximately 50 mg/dL, is common and varies in prevalence by ancestry (affects 1 in 5 European individuals). [9][10][11][12] Mendelian randomization studies 7,13 suggest that lipoprotein(a) is a causal contributor to ASCVD, 7,13 and ongoing randomized clinical trials are testing this hypothesis. [14][15][16] Both the diagnostic yield and clinical value of genetic testing of LPA are not well understood. Specifically, unlike LDL cholesterol, it remains unclear whether genetic factors related to lipoprotein(a) may provide information regarding lifetime exposure relevant to ASCVD risk prediction and selection of patients for primary prevention with statin therapy. 4, [17][18][19] Here, we report our investigation of the prognostic utility of testing for the genetic determinants of lipoprotein(a) relative to direct measurement of lipoprotein(a) levels in a large cohort of middle-to late-age adults not using lipid-lowering medication and without baseline ASCVD.

UK Biobank Cohort
The UK Biobank is a prospective observational study of approximately 500 000 volunteer adults aged 40 to 69 years recruited from 22 sites across the United Kingdom between 2006 and 2010, with follow-up ongoing. This study included individuals of third-degree relatedness or less (Figure 1). 20 Biochemical measurements, physical examination measurements, and medical histories were assessed at the time of study enrollment. Self-reported ethnicities were categorized as mixed, African/Black, European/White, East Asian, South Asian, and unknown.

Key Points
Question Does measurement of lipoprotein(a) and/or LPA genetic risk score (GRS) have clinical utility in risk prediction of incident atherosclerotic cardiovascular disease (ASCVD)?
Findings In this cohort of 283 540 adults recruited by the UK Biobank, both measured lipoprotein(a) and LPA GRS were associated with comparable risk of incident ASCVD events. The LPA GRS did not yield additional prognostic information beyond measured lipoprotein(a), and both measured lipoprotein(a) and LPA GRS yielded modest improvements in the discrimination of ASCVD risk relative to the Pooled Cohort Equations or QRISK3.
Meaning Cardiovascular risk assessment with lipoprotein(a) may be achieved with either direct measurement or an LPA GRS.

Lipoprotein(a) Measurement
Lipoprotein(a) was measured in nanomoles per liter at study enrollment using an immunoturbidimetric method on the Beckman Coulter AU5800 platform (Randox Bioscience, UK), which is isoform insensitive. 10 To convert lipoprotein(a) values to milligrams per deciliter, divide by 2.15.

LPA Genetic Risk Score
We used genotyping array data from UK Biobank participants to calculate a previously described weighted LPA genetic risk score (GRS), composed of 43 single-nucleotide variants that were conditionally and significantly associated with lipoprotein(a) levels in data sets external to the UK Biobank (eTable 1 and eMethods in the Supplement). 7 The units of the LPA GRS were converted from milligrams per deciliter to nanomoles per liter by multiplying by 2.15 to harmonize units with the measured lipoprotein(a). 21 An LPA GRS of at least 120 was considered elevated. 7,8,11,12,17,22 Atherosclerotic Cardiovascular Disease Outcomes The outcomes of myocardial infarction, coronary artery disease, ischemic stroke, peripheral arterial disease, cardiovascular mortality, and a composite of the aforementioned outcomes were based on International Statistical Classification of Diseases and Related Health Problems and Office of Population Censuses and Surveys Classification of Interventions and Procedures codes (eTable 2 in the Supplement). 23 Incident events were defined as the first event occurring between the date of enrollment and the end of follow-up of March 31, 2020. Individuals were censored at death or loss to follow-up. Individuals reporting use of cholesterol-lowering medication or prevalent angina, coronary artery disease, history of myocardial infarction, history of stroke, or peripheral arterial disease/ revascularization at study enrollment were excluded from analyses of incident events. A sensitivity analysis was performed with individuals without prevalent ASCVD but using cholesterol-lowering medication.

Statistical Analyses
Analyses were performed using R, version 4.0.0 (R Core Team [2020]). The association between LPA GRS and measured lipoprotein(a) levels were assessed by Spearman correlation and adjusted linear regression models (covariates of age, sex, assessment center, genotyping batch, and the first 5 principal components of ancestry).
We assessed the risk of incident ASCVD events Cox proportional hazards models with the covariates of sex, age, assessment center, genotyping batch, and the first 5 principal components of ancestry using the survival package (version 3.1-12). 24 When explicitly stated, Cox proportional hazards models included the additional covariate of continuous, measured lipoprotein(a) or LPA GRS.
To compare whether an elevated LPA GRS was associated with greater risk of ASCVD for a given lipoprotein(a) level, individuals with and without an elevated LPA GRS were matched at a 1:1 ratio based on measured lipoprotein(a) levels using the nearest neighbor algorithm from the MatchIt, version 3.0.2 (R Core Team [2020]), package with a caliper set to 0.02. 25 Measured lipoprotein(a) levels were compared between matched individuals with and without an elevated LPA GRS using a Mann-Whitney U test. The risks of incident ASCVD events were compared between matched individuals with and without an elevated LPA GRS using a log-rank test.
We calculated the incidence rates for composite ASCVD events for each percentile of measured lipoprotein(a) and LPA GRS using the Poisson rate exact method from the epitools package, version 0.5-10.1. 26 Additionally, for individuals with and without an elevated LPA GRS (≥120 nmol/L), we calculated the incidence rates for composite ASCVD events for percentiles of measured lipoprotein(a) and LPA GRS or for 50 nmol/L bins of measured lipoprotein(a) levels with the maximum lipoprotein (a) bin was set to lipoprotein(a) levels 250 nmol/L or greater.
Lastly, we assessed whether lipoprotein(a) levels and/or LPA GRS could refine the ASCVD risk discrimination for individuals without diabetes or severe hypercholesterolemia (LDL cholesterol level ≥190 mg/dL; to convert to millimoles per liter, multiply by 0.0259) classified as borderline-intermediate risk by the Pooled Cohort Equations 27,28 and QRISK3, 29 respectively (predicted 10-year ASCVD risk: 5%-20%). This borderline-intermediate risk group consists of individuals in whom statin therapy could be warranted if risk enhancers are present for a given patient (ie, elevated lipoprotein[a]). 27, 28 We used multiple imputation by chained equations to impute missing values needed to calculate ASCVD risk according to the Pooled Cohort Equations and QRISK3 with the mice package, version 3.11.0 (The R Foundation). 30 The associations between continuous, measured lipoprotein(a) levels, LPA GRS, and Pooled Cohort Equations-derived or QRISK3-derived clinical risk scores 27 with incident myocardial infarction, ischemic stroke, or cardiovascular mortality were assessed by the area under the curve from receiver operating characteristic models 31 and the Harrell C statistic 32 from Cox proportional hazards models. We assessed whether Cox proportional hazards models and correlated receiver operating characteristic curves were statistically different using an analysis of variance test and Delong test, respectively. Statistical significance was claimed when 2-sided P values were less than .05.

Cohort Characteristics
A flowchart of the subgroups used in this study is shown in Figure 1. For the overall study population of 374 099 individuals, the mean (SD) age at study enrollment was 57.6 (8.0) years, and 204 355 individuals were female (54.6%; Table 1 For 300 839 individuals without prevalent ASCVD and not using cholesterol-lowering medication, the mean (SD) age at study enrollment was 56.6 (8.0) years, and 174 555 individuals were female (58.0%). The median level of lipoprotein(a) was 24.1 nmol/L (interquartile range, 73.6 nmol/L) and the median length of follow-up for composite ASCVD events was 11.1 years (interquartile range, 1.4 years; eTable 4 in the Supplement). A 120-nmol/L increase in measured lipoprotein(a) was significantly associated with increased risk of incident, composite ASCVD events for White/European individuals (hazard ratio

Association of Measured Lipoprotein(a) and LPA GRS With Atherosclerotic Cardiovascular Diseases
For individuals of White/European ethnicity without prevalent ASCVD and not using cholesterol-lowering medication (n = 283 540), a 120-nmol/L increase in either measured lipoprotein (a) or LPA GRS was associated with increased risk of incident peripheral arterial disease (HR, 1.25; 95% CI, 1.19-1. 31 Figure 2). In a sensitivity analysis of individuals of White/European ethnicity without prevalent ASCVD but using cholesterol-lowering medication (n = 43 829), the effect estimates for the association between lipoprotein(a) and risk of incident ASCVD were modestly attenuated (eFigure 4 in the Supplement). Notably, the risk of incident, composite ASCVD events and probability of ASCVD by age 75 years was more pronounced at the extremes of measured lipoprotein(a) than LPA GRS (eFigure 5 in the Supplement).

Using LPA GRS for Cardiovascular Disease Risk Prediction Compared With Measured Lipoprotein(a) Levels
The associations between LPA GRS and risk of incident peripheral arterial disease, coronary artery disease, myocardial infarction, ischemic stroke, cardiovascular mortality, and composite ASCVD were substantially attenuated when models were additionally adjusted for measured lipoprotein(a) levels ( Figure 2). Additionally, to assess the association of an elevated LPA GRS with ASCVD risk at comparable levels of measured lipoprotein(a), we matched individuals of White/ European ethnicity with and without an elevated LPA GRS (≥120 nmol/L) based on measured lipoprotein(a). Measured lipoprotein(a) levels were comparable between matched individuals with and without elevated LPA GRS (median,  There was no significant difference in the risk of an incident, composite ASCVD events between individuals with and without an elevated LPA GRS matched for lipoprotein(a) levels (log-rank P = .60; Figure 3A).

Research Original Investigation
To further explore the influence of an elevated LPA GRS on risk of ASCVD at comparable levels of lipoprotein(a), we assessed the incidence rate of composite ASCVD events across 50-nmol/L bins of measured lipoprotein(a) for individuals of White/European ethnicity with and without an elevated LPA GRS (≥120 nmol/L). In general, there was a trend toward increased incidence rate of ASCVD with increasing bins of lipoprotein(a) levels for both individuals with and without an elevated LPA GRS. However, for each elevated LPA GRS threshold, an elevated lipoprotein(a) GRS did not significantly associate with a greater incidence rate of ASCVD compared with a nonelevated LPA GRS within a given bin of measured lipoprotein(a) (Figure 3B and C).

Lipoprotein(a) and Improvement in Risk Discrimination Among Borderline-Intermediate Risk Patients
The distribution of QRISK3 and Pooled Cohort Equations 10year ASCVD risk scores for the 300 839 individuals of European and non-European ethnicity not using cholesterollowering medication and without prevalent ASCVD at enrollment are displayed in eFigure 6 in the Supplement. Prediction with QRISK3 had significantly greater ASCVD risk discrimination compared with the Pooled Cohort Equations for these individuals (eTable 5 in the Supplement). We used a subgroup of 113 703 and 144 350 individuals of White/European ethnicity defined as borderline to intermediate ASCVD risk by the Pooled Cohort Equation and QRISK3 to assess the role of measured lipoprotein(a) and LPA GRS as a risk enchancer. 12,27 Adding continuous, measured lipoprotein(a) levels to QRISK3 provided modest improvements to the predicted risk of incident ASCVD events as assessed by AUROC and Harrell C statistic ( Table 2). The AUROC improved from 0.640 (95% CI, 0.633-0.647) to 0.642 (95% CI, 0.634-0.649) and 0.642 (95% CI, 0.634-0.649) when measured, continuous lipoprotein(a) levels and LPA GRS were added to QRISK3 (P = .005 and P = .01, respectively). Similar results were observed when continuous, measured lipoprotein(a) was added to the Pooled Cohort Equations and when analyses were restricted to only individuals with complete clinical data (no imputation and exclusion of individuals with missing data; eTables 6-8 in the Supplement).

Discussion
Here we used the UK Biobank cohort to demonstrate that measured lipoprotein(a) and an LPA GRS are associated with  increased risk of incident ASCVD in a primary prevention setting. The major advance from this study is the demonstration that an LPA GRS offered comparable ASCVD risk prediction to directly measured lipoprotein(a). In a large observational cohort, we observed that increased lipoprotein(a) is independently associated with future risk for ASCVD but with modest discrimination in addition to clinical risk scores at the level of population screening for primary prevention.
In contrast to LDL cholesterol, this work suggests that profiling the genetic determinants of plasma lipoprotein(a) does not provide additional value to measured levels of lipoprotein(a) in terms of ASCVD risk prediction. These data suggest that the LPA GRS is likely not a better marker for exposure to lipoprotein(a) than an isolated lipoprotein(a) measurement. We suspect that this is explained by 2 key factors: (1) lipoprotein(a) displays higher heritability than LDL cholesterol lev-  Our findings are in accordance with other studies demonstrating that adding continuous lipoprotein(a) levels to clinical risk scores leads to modest improvements in risk discrimination for ASCVD. 22,41,42 While some guidelines support broad population-based screening with lipoprotein(a), 43 the small improvement in the C statistics relative to clinical risk scores (Δ C statistic of approximately 0.002) suggests that measured lipoprotein(a) or LPA GRS may be an inefficient approach for refinement of ASCVD risk among asymptomatic middle-aged adults broadly. 44 However, because lipoprotein(a) levels display an extremely right-skewed distribution in the general population (potentially varying more than 1000-fold between individuals, approximately 0.2 to ≥200 mg/dL), individuals with extreme lipoprotein(a) levels greater than 200 mg/dL could have a 3-to 4-fold increased lifetime risk of ASCVD. 7,13,17,45 In such cases, the modest improvement in ASCVD risk discrimination that was observed in this study when continuous levels of measured lipoprotein(a) or LPA GRS were added to clinical risk scores may underestimate cardiovascular risk. Inclusion of the entire range of lipoprotein(a) levels in ASCVD risk discrimination and reclassification models often fails to accurately quantify the cardiovascular risk associated with the extremes of elevated lipoprotein(a) (ie, because extremely elevated lipoprotein[a] is rare there is limited predictive power). 17,41 Efficient strategies to identify individuals with extremely elevated lipoprotein(a) levels require further research.
Previous studies suggest that small apolipoprotein(a) isoforms, which are also associated with higher lipoprotein(a) levels, may increase an individual's risk of ASCVD more than expected by changes in lipoprotein(a) concentration alone. 46 However, in this study we did not find that the LPA GRS, which contains several genetic variants associated with small apolipoprotein(a) size, 6 explained additional ASCVD risk beyond measured levels of lipoprotein(a). Unlike isoformindependent lipoprotein(a) molar concentration, the associations of apolipoprotein(a), and by extension isoform-dependent measurements, with ASCVD may be confounded by pleiotropic effects on high-density lipoprotein cholesterol and triglycerides. 19 In some cases, the genetic determinants of elevated lipoprotein(a) may help discern the familial risk of lipoprotein(a)-associated ASCVD that is not always conclusive from lipoprotein(a) measurement alone. For instance, a single copy of the rs10455872-G allele, which is common in European populations, is known to associate with extremely elevated lipoprotein(a) that can present phenotypically similar to familial hypercholesteremia and thus may be amenable to cascade screening. 47,48 Limitations First, a major limitation of this study is that it focused on individuals of White/European ethnicity, and the generalizability of these findings, in particular the LPA GRS, to other ethnic groups requires further research. Notably, we show that the association of measured lipoprotein(a) with incident ASCVD risk may vary between ethnic groups and that, similar to other GRSs, the LPA GRS performed suboptimally in non-White/ European ethnicities. An LPA GRS needs to be developed and tested in non-White/European cohorts to determine whether these findings extend to individuals of other ethnic groups. Second, most individuals enrolled in this study were recruited at middle age (mean [SD], 56.8 [8.0] years), with median follow-up of 11.1 years, and thus the influence of measured lipoprotein(a) and LPA GRS on incident premature and lifelong ASCVD risk remains to be determined in prospective primary prevention cohorts with younger age of enrollment and longer follow-up. Third, additional limitations include the ascertainment bias associated with volunteer recruitment, use of hospitalization and operation codes to define ASCVD events, and that the LPA GRS was constructed with winsorization (lipoprotein[a] levels greater than 130 mg/dL were set to 130 mg/dL), which may fail to explain some of the genetic variation associated with extremely elevated levels of measured lipoprotein(a). 7 Despite this, the LPA GRS is still a reasonable genetic predictor that explains greater than 60% of the variation in lipoprotein(a) levels, an association that is considerably greater than contemporary GRSs for other lipid traits such as LDL cholesterol. 49

Conclusions
In conclusion, the results from this study suggest that an LPA GRS provides comparable risk prediction for incident ASCVD compared with measured lipoprotein(a). Measured lipoprotein(a) and the LPA GRS both provided modest improvement in risk discrimination beyond guideline-supported risk scores, which supports the role of lipoprotein(a) as a risk-enhancing factor.