Genetic Prediction of Circulating Lipoprotein(a) Levels in Diverse Populations

BackgroundCirculating lipoprotein(a) [Lp(a)] levels are highly heritable and linked to atherosclerotic cardiovascular disease, yet clinical measurement rates remain low (<1%) in the United States. The high heritability of Lp(a) across populations makes genetic prediction an attractive approach for closing this testing gap, but existing polygenic scores transfer poorly across populations. Haplotype-based prediction models, which use standard genome-wide genotype data to capture common-, rare-, and structural-variation at the LPA locus, could bridge this gap, enabling opportunistic identification of individuals with elevated Lp(a) levels across diverse populations within existing large, genotyped cohorts. ObjectivesThis study sought to develop and validate a haplotype-based prediction model using genome-wide genotype data to identify individuals with elevated Lp(a) levels across diverse populations. MethodsWe developed an LPA-haplotype model using data from the All of Us Research Program and validated it in the Penn Medicine BioBank (PMBB), Mass General Brigham Biobank (MGBB), and Mount Sinai BioMe cohorts. Primary outcomes included model performance for predicting continuous Lp(a) concentrations (r{superscript 2}) and identifying elevated Lp(a) levels (>125 nmol/L) through positive predictive value (PPV) and number needed to test (NNT). ResultsAmong PMBB (n = 1856), MGBB (n = 1401), and BioMe (n = 1686) participants with available genotype and Lp(a) measurements, average age was 60 years, and 51% were female. Overall r{superscript 2} of the haplotype model was 0.46 (95% Credible Interval [CrI] 0.32 to 0.6), with similar performance across genetically inferred ancestries and cohorts. For identifying elevated Lp(a) levels >125 nmol/L the overall PPV was 0.81 (95% CrI 0.6 to 0.89), corresponding to a NNT of 1.2 (95% CrI 1.1 to 1.7) individuals predicted to have elevated levels needing to undergo clinical testing to identify one true elevation. In the full PMBB cohort (n = 49310), the haplotype model identified elevated Lp(a) at a rate of 128 per 1000 (95% CrI 125 to 130), corresponding to an estimated 14.4-fold improvement (95% CrI 13.1 to 15.9; P(improvement) = 1) in identification rate compared with the existing rate of clinical assessment. ConclusionsA haplotype-based genetic model effectively identified individuals with elevated Lp(a) levels across diverse populations, with potential utility for opportunistic screening among cohorts where genotype data is available, but Lp(a) testing rates are low.