Correlation between Serum and Plasma Lipoprotein(a) Measurements: ARIC Study

Abstract

Lipoprotein(a) [Lp(a)] is recommended for atherosclerotic cardiovascular disease (ASCVD) risk assessment and is targeted by new drugs in development. However, the association of Lp(a) with ASCVD was identified from epidemiologic studies that typically used plasma samples. To determine the agreement between serum- and plasma-based Lp(a) measurements, we measured Lp(a) in both EDTA plasma and serum using the same validated, isoform-insensitive assay in 100 ARIC participants equally distributed across age, sex, race, and Lp(a) levels (0 to >200 nmol/L). We assessed correlation (Spearman), mean differences, and bias (Deming regression). The correlation between serum and plasma was excellent (Spearman’s r=0.99). Deming regression yielded a slope of 1.05 (95% CI 1.03, 1.08) and intercept of 0.26 nmol/L (95% CI −1.97, +2.41). Bland–Altman analysis identified no statistically significant mean difference (serum minus plasma: +7.7 nmol/L [95% CI −16.7, +32.0]). We demonstrated excellent agreement between serum and plasma Lp(a), supporting the use of either specimen for Lp(a) measurement and pooling data from both sources.

Introduction

Lipoprotein(a) [Lp(a)], an apolipoprotein B–containing lipoprotein whose level is primarily determined by genetics, is a causal and independent risk factor for atherosclerotic cardiovascular disease (ASCVD).^1,2 Globally, nearly one in five individuals is estimated to have elevated Lp(a) levels,³ with the highest prevalence among people of African and South Asian ancestry.⁴ The development of potent Lp(a)-lowering therapies capable of reducing levels by more than 80%^5,6 has spurred renewed clinical, regulatory, and public health interest in accurately estimating Lp(a) burden, defining novel Lp(a) phenotypes, and determining appropriate treatment indications. However, this resurgence has exposed long-standing measurement challenges, including measurement units, assay calibration, and sample types (plasma versus serum).⁷ In this study, we focus on sample type.

Large epidemiologic studies that demonstrate associations between Lp(a) and CVD have typically used plasma^8,9 with far fewer studies using serum¹⁰ samples. Regardless of sample type, these studies consistently show strong associations between Lp(a) and adverse ASCVD outcomes.^8,10 Yet whether measurements of Lp(a) in serum and plasma (using the same assay) are directly comparable remains unknown. As meta-analyses and pooled cohorts have become commonplace, the challenge of harmonizing biomarker data from these different sources is growing in importance.¹¹ This issue is especially relevant for Lp(a), as combining data from the many available longitudinal cohorts and biobanks would result in a richly phenotyped, ancestrally diverse combined cohort that would uniquely position researchers to clarify Lp(a)’s role in ASCVD in diverse populations. We, therefore, sought to evaluate the degree of agreement between serum and plasma Lp(a) measurements using an apolipoprotein(a) [apo(a)] isoform–insensitive assay.

Methods

Study Population

The Atherosclerosis Risk in Communities (ARIC) Study is a prospective cohort study that enrolled 15,792 participants aged 45–64 years from four U.S. communities during 1987–1989, as described previously.¹² For the present analysis, blood samples were collected in gel separator tubes containing clot activator for preparation of serum samples and spray-coated K₂EDTA-containing tubes for preparation of plasma samples at visit 7 (2018–2019); samples were processed and stored at −80°C, and never thawed prior to analysis. We previously demonstrated that decades of storage under these conditions preserve the reliability of Lp(a) measurements.⁸ Protocols for the ARIC study were approved by the institutional review board at each participating site, and all participants provided written informed consent.

Sample Selection

We selected a sample of 50 Black and 50 White participants. We chose this sample size because power calculations (using the blandPower package in R) indicated about 88% power to detect a mean difference of 7.5 nmol/L (assuming the true mean [SD] difference between serum and plasma Lp(a) is 5 [1] nmol/L) at α = 0.05. Because participants had Lp(a) measured in their plasma samples at visit 5 and because Lp(a) is relatively stable in older adults, we could select individuals uniformly across the plasma Lp(a) distribution (from 0 to >200 nmol/L) within each racial subgroup for profiling at visit 7.

Assays

In 2025, stored visit 7 blood samples from 100 participants were thawed, and Lp(a) concentrations were measured contemporaneously in singleton in both serum and EDTA plasma subfractions in pairwise fashion for each participant. Lp(a) levels were quantified in nmol/L using an automated immunoturbidimetric assay from Denka Seiken (Tokyo, Japan) on a Beckman Coulter AU480 clinical chemistry analyzer according to manufacturer protocol.¹³ This assay is largely insensitive to variations in kringle IV type 2 repeat number and, consequently, to apo(a) isoform heterogeneity.¹⁴ Interassay coefficients of variation were estimated as approximately 2.0% for participants of European ancestry and 2.5% for participants of African ancestry, based on data from ARIC visit 4, demonstrating excellent analytical performance.

Statistical Analysis

For both serum and plasma measurements, we report the median and 25th and 75th percentiles of Lp(a) measurements. We assessed correlation between serum and plasma Lp(a) measurements via the Spearman correlation coefficient. We then fitted Deming regression models to evaluate potential constant bias (intercept) and proportional bias (slope). To further examine agreement and bias, we constructed a Bland–Altman plot, which displays, for each participant, the difference (serum minus plasma) against the average of the two measurements.

In a supplementary analysis, we conducted cross-tabulation analysis at clinically relevant thresholds (75 nmol/L, 125 nmol/L, and 200 nmol/L) to evaluate for concordance between plasma and serum measurements and to quantify the proportion of participants reclassified across these thresholds.

Results

Both serum and plasma Lp(a) measurements exhibited a right skew. Serum Lp(a) values ranged from 4.8 to 519.0 nmol/L, and plasma values ranged from 5.0 to 494.0 nmol/L. The median (25th, 75th percentile) Lp(a) was 121.6 (42.5, 216.0) nmol/L for serum samples and 114.7 (40.6, 195.8) nmol/L for plasma samples. We observed a strong positive correlation between serum and plasma measurements (Spearman's r = +0.99). The Deming regression analysis revealed minimal proportional bias (slope = 1.05, 95% confidence interval [CI] 1.03, 1.08) and no statistically significant constant bias (intercept = 0.26 nmol/L, 95% CI −1.97, +2.41 nmol/L) (Figure 1). In the Bland–Altman analysis, the mean paired difference (serum minus plasma) was +7.7 (95% CI −16.7, +32.0) nmol/L, which was not statistically significant. At lower mean Lp(a) levels, differences clustered tightly around zero; at higher Lp(a) levels, variability increased and differences were spread farther apart (Figure 2). In supplementary analyses (not shown), 1%, 1%, and 2% of participants were reclassified at the 75-nmol/L, 125-nmol/L, and 200-nmol/L thresholds, respectively. Overall, 96% of participants remained in the same risk category when comparing serum with plasma measurements.

Figure 1. Deming regression plot.

The Deming regression analysis revealed minimal proportional bias (slope = 1.05, 95% confidence interval [CI] 1.03, 1.08) and no statistically significant constant bias (intercept = 0.26 nmol/L, 95% CI −1.97, +2.41 nmol/L). Lp(a): lipoprotein(a).

Figure 2. Bland-Altman plot.

In the Bland–Altman analysis, the mean paired difference (serum minus plasma) was +7.7 (95% confidence interval −16.7, +32.0) nmol/L, which was not statistically significant. At lower mean lipoprotein(a) [Lp(a)] levels, differences clustered tightly around zero; at higher Lp(a) levels, variability increased and differences were spread farther apart.

Discussion

Previous studies comparing serum and plasma measurements of lipoproteins and other biomarkers have generally demonstrated good concordance.^15-17 In this study, we extend that work to Lp(a): serum Lp(a) measured by the Denka Seiken immunoturbidimetric assay correlated very closely with plasma measurements, and we observed no statistically significant difference in mean levels. Agreement was excellent at lower concentrations; variability increased at higher Lp(a) levels, with a Deming regression slope of 1.05 implying that serum Lp(a) values tend to be ~0.05 nmol/L higher for each 1 nmol/L of plasma Lp(a). Higher concentrations in serum compared with plasma have been reported for several biomarkers. This has been attributed in part to volume displacement during clot formation, which reduces serum sample volume and increases measured concentrations, whereas plasma levels may be slightly lower because of dilution from anticoagulants in collection tubes.¹⁶ It is important to note that in our study, we used spray-coated K₂EDTA blood collection tubes to obtain the plasma subfraction, and therefore dilution effects from anticoagulant in the collection tube are expected to be minimal. However, the dilution effect from anticoagulants is expected to be greater in blood collection tubes containing liquid solutions of anticoagulants (e.g., 3.2% sodium citrate). The increased scatter we observed at high concentrations is consistent with prior literature showing higher measurement variability at elevated Lp(a) values, even when using the same specimen type.¹⁸ Despite this, overall, reclassification rates were low across clinically relevant thresholds (1–2%).

Although no commercial assay is completely free from apo(a) isoform sensitivity,¹⁴ the Denka Seiken method is among the least sensitive to isoform heterogeneity because it uses multiple calibrators with varied apo(a) isoform distributions, each traceable to the WHO/IFCC reference material.¹⁹ Indeed, Lp(a) measurements using this assay show good agreement with other commercially available assays, suggesting our findings can likely be replicated using other assays.²⁰ Both the Roche assay, which has been approved by the US Food and Drug Administration for measurement of Lp(a) in nmol/L, and the Randox assay, which was used in UK Biobank, use reagents supplied by Denka Seiken, further suggesting high likelihood of replication.

Because serum and plasma Lp(a) values are remarkably similar over much of the concentration range, our findings support interchangeability of sample source in the clinical setting and for pooling data in epidemiologic analyses and meta-analyses, thereby facilitating cross-cohort harmonization. Indeed, there have already been efforts to pool Lp(a) measurements from both sample types,²¹ and this becomes particularly relevant in the era of emerging potent Lp(a)-lowering therapies, as researchers seek to define Lp(a) phenotypes beyond traditional cardiovascular endpoints.

Our study has several strengths: it was adequately powered to detect clinically meaningful differences; we measured Lp(a) using the same assay, which was apo(a) isoform insensitive; we sampled across the full Lp(a) distribution; and we specifically included a substantial proportion of Black participants, an important consideration given racial disparities in Lp(a) levels and related disease burden. Our post hoc power analyses revealed that despite the observed variability (SD of paired differences of 12.4 nmol/L), our study had >99% power to detect the observed mean difference.

We also acknowledge limitations. Our results may not generalize to children, younger adults, or other racial/ethnic groups. Concordance was less tight at very high Lp(a) levels, which is a known limitation of Lp(a) measurement, not necessarily specific to sample type. It is a limitation of the study that the measurements were carried out as single measurements rather than duplicate or triplicate measurements of each sample. Multiple measurements of each sample and sample pair would have minimized the technical inaccuracies and would have allowed to make a more accurate identification of differences between plasma and serum samples. Finally, although the Denka Seiken assay is relatively robust with respect to isoform sensitivity, residual bias from extreme apo(a) isoform variation cannot be fully excluded,¹⁴ and our inability to validate these findings with other assays may limit the replicability of our findings across different Lp(a) assays.

Highlights.

• We found strong correlations between serum and plasma measurements of lipoprotein(a).

• These findings support the clinical use of either sample type for lipoprotein(a) measurement.

• Our results also support pooling of serum- and plasma-derived lipoprotein(a) measurements in epidemiological studies.