A study by Misra et al. describes the development and validation of a clinically implementable polygenic risk score (PRS) panel for eight cardiovascular conditions, addressing a key gap between genomic research and routine clinical use.
Cardiovascular diseases remain the leading cause of mortality globally and have a strong heritable component. While PRS can quantify inherited risk by aggregating the effects of numerous genetic variants, their clinical adoption has been limited by lack of standardisation, variability across scores, and insufficient frameworks for reporting and validation.
The primary objective of this study was to create integrated, consensus PRS for eight cardiovascular traits—coronary artery disease (CAD), atrial fibrillation (AF), type 2 diabetes mellitus (T2DM), venous thromboembolism (VTE), thoracic aortic aneurysm (TAA), extreme hypertension, severe hypercholesterolaemia, and elevated lipoprotein(a) [Lp(a)]—and to establish a scalable framework for their clinical reporting. The authors also aimed to evaluate predictive performance across large, diverse cohorts and assess potential clinical utility.
The study utilised genomic and electronic health record data from three major biobanks: the All of Us Research Program (AOU; n=245,394), the Mass General Brigham Biobank (MGBB; n=53,306), and the UK Biobank (for Lp(a) training). Phenotypes were defined using structured clinical data, including diagnostic codes and laboratory thresholds. For example, severe hypercholesterolaemia was defined as LDL cholesterol ≥190 mg/dL, and extreme hypertension as blood pressure ≥180/120 mm Hg.
PRS were constructed by integrating multiple publicly available trait-specific scores sourced from the PGS Catalog. The authors used an elastic-net approach (PRSmix) to combine these into a single optimised PRS per condition. Training was performed in 80% of the AOU cohort (or UK Biobank for Lp(a)), with the remaining 20% reserved for internal validation. External validation was conducted in the MGBB cohort. Models were adjusted for age, sex, and genetic ancestry using principal components derived from a reference population.
Results demonstrated that integrated PRS consistently performed as well as or better than individual component scores across all traits. In the MGBB validation cohort, individuals in the highest risk categories showed substantially increased odds of disease compared with those at average risk. For example, the top decile of PRS was associated with a 3.7-fold increased risk of CAD, 3.1-fold for T2DM, and 3.0-fold for AF. Hypercholesterolaemia showed a 4.1-fold increase, while Lp(a) exhibited a particularly strong association, reflecting its high heritability. Lower but still significant associations were observed for VTE and TAA.
The integrated PRS also demonstrated good discrimination and calibration across cohorts, although some differences were observed. Risk tended to be overestimated in younger individuals and underrepresented in certain subgroups, particularly those of non-European ancestry, reflecting known limitations in genomic datasets. External validation in the MGBB showed acceptable calibration overall, though some underprediction occurred in the highest risk groups. Recalibration improved performance for clinical deployment.
The study showed that incorporating PRS into existing clinical risk models improved risk classification. For CAD, adding PRS to the Pooled Cohort Equations or PREVENT model resulted in net reclassification improvements of 0.17 and 0.18, respectively. For AF, adding PRS to the CHARGE-AF model yielded an improvement of 0.14. These findings suggest that PRS can refine risk estimates, particularly among individuals at borderline or intermediate clinical risk.
Prospective analyses further demonstrated that high PRS was associated with increased incidence of cardiovascular events over time. Hazard ratios for high versus average PRS were approximately 2.08 for CAD, 2.32 for AF, 2.10 for T2DM, 1.60 for VTE, and 1.49 for TAA. These associations were consistent even among individuals younger than 50 years, highlighting the potential for early risk identification before traditional risk factors manifest.
The proportion of variance explained by PRS varied by trait. PRS accounted for modest heritability in most conditions (e.g. ~11% for CAD, ~8% for AF), but a substantially larger proportion for Lp(a) (~43%). The study also found that a large proportion of individuals had elevated genetic risk for at least one condition, suggesting broad population-level relevance.
To support clinical implementation, the authors developed a structured PRS report that presents individual risk as percentiles and relative risk categories, alongside explanatory information. The test is now available as a clinically orderable assay within a certified molecular diagnostics laboratory. The authors emphasise transparency, reproducibility, and periodic updates as essential components of responsible clinical deployment.
Source: JACC
Image Credit: iStock
