Polygenic Risk Scores for Coronary Artery Disease Across Ancestries

The Validation Gap

Polygenic risk scores (PRS) identify 8.0% of individuals at ≥3-fold elevated coronary artery disease (CAD) risk—a prevalence 20-fold higher than monogenic familial hypercholesterolemia mutations conferring comparable risk (Khera et al., 2018). However, most PRS are trained predominantly on European-ancestry genomes, producing 25–46% accuracy reductions in African, Hispanic, and South Asian populations due to linkage disequilibrium differences, allele frequency variation, and gene-environment interactions (Patel et al., 2023). The validation gap persists: clinical implementation remains concentrated at academic centers despite randomized trial evidence demonstrating 77% relative increase in statin initiation and meaningful LDL-C reduction (Kullo et al., 2016). This brief synthesizes multi-ancestry validation evidence and addresses equitable implementation pathways.

Evidence for Clinical Validation

Foundational Genome-Wide PRS Demonstrates Clinical-Grade Prediction

The metaGRS developed by Inouye et al. (2018) established clinical CAD risk prediction by integrating 1,745,180 genetic variants in 480,000 UK Biobank participants. External validation demonstrated a C-index of 0.623 (95% CI: 0.615–0.631) for incident CAD—higher than any single conventional risk factor including smoking, diabetes, or family history. The metaGRS stratified individuals into markedly different life-course trajectories: participants in the top quintile exhibited a hazard ratio of 4.17 (95% CI: 3.97–4.38) compared with the bottom quintile. For men in the top 20% with ≥2 conventional risk factors, 10% cumulative CAD risk was reached by age 48, highlighting early-life risk stratification potential.

Simultaneously, Khera et al. (2018) validated a genome-wide PRS incorporating 6,630,150 variants in 409,258 UK Biobank participants, identifying 8.0% of the population at ≥3-fold increased CAD risk. This prevalence was 20-fold higher than the carrier frequency of familial hypercholesterolemia mutations conferring comparable threefold risk. The genome-wide approach achieved an AUC of 0.81 for prevalent CAD, establishing that polygenic architecture—not monogenic mutations—accounts for the majority of inherited CAD risk.

Multi-Ancestry Validation Reveals Substantial Transferability Gaps

The most significant barrier to equitable PRS implementation is marked reduction in predictive accuracy for non-European populations. Patel et al. (2023) developed GPS_Mult by integrating GWAS data from CAD and 10 cardiovascular risk factors across diverse populations, validating performance in the Million Veteran Program (33,096 African, 124,467 European, 16,433 Hispanic ancestry participants) and Genes & Health cohort (16,874 South Asian ancestry). The GPS_Mult score demonstrated:

Despite these improvements, performance gaps persist: African-ancestry participants showed 27% lower effect sizes compared to European ancestry (OR 1.25 vs. 1.72), quantifying the equity challenge. When combined with Pooled Cohort Equations, the C-statistic improved from 0.739 to 0.763 (95% CI: 0.759–0.768), with net reclassification improvement of 7.0% at the clinically relevant 7.5% threshold.

Biological Mechanisms Underlying Ancestry Performance Differences

Four fundamental mechanisms drive transferability gaps across ancestry groups: (1) Linkage Disequilibrium Differences—African populations have shorter LD blocks due to longer evolutionary history, meaning European-selected tag SNPs often fail to capture causal effects; (2) Allele Frequency Variation—protective 9p21 haplotypes responsible for substantial PRS risk stratification are virtually absent in African-ancestry populations; (3) Training Data Bias—approximately 79% of GWAS participants remain European-ancestry, creating systematic underrepresentation; (4) Gene-Environment Interactions—effect sizes may vary based on environmental contexts (diet, stress, healthcare access) that differ systematically between populations.

Randomized Trial Evidence Demonstrates Clinical Utility

The MI-GENES trial (Kullo et al., 2016) provides the strongest randomized evidence for PRS clinical implementation. Among 203 statin-naive participants aged 45–65 with intermediate CHD risk, genetic risk score disclosure significantly increased statin initiation (39% vs. 22%, P<0.01)—a 77% relative increase—and reduced LDL-C at 6 months (96.5±32.7 vs. 105.9±33.3 mg/dL, P=0.04). Participants were 4.88× more likely to visit CHD websites, 2.99× more likely to access risk information via patient portal, and 3.13× more likely to discuss risk with family members. Notably, LDL-C reduction occurred through pharmacotherapy rather than lifestyle modification—dietary fat intake and physical activity did not differ significantly between groups.

Khera et al. (2016) established across 55,685 participants in four prospective cohorts that favorable lifestyle reduced CAD risk by 46% (HR 0.54; 95% CI: 0.47–0.63) even among individuals in the highest genetic risk quintile. In the ARIC cohort, ten-year CAD incidence dropped from 10.7% to 5.1% with favorable lifestyle in high genetic risk participants, establishing that high polygenic risk is modifiable through behavioral intervention.

Validation Implementation Solution

Mathematical Framework for Risk Integration

The combined risk model integrates polygenic and conventional factors:

$$\text{Risk}_{\text{combined}} = \beta_0 + \beta_1(\text{PRS}_{\text{standardized}}) + \beta_2(\text{Age}) + \beta_3(\text{LDL-C}) + \beta_4(\text{SBP}) + \beta_5(\text{Smoking}) + \beta_6(\text{Diabetes})$$

Where PRS is standardized to mean=0, SD=1, enabling interpretation of hazard ratios per standard deviation increase. The net reclassification improvement quantifies clinical benefit:

$$\text{NRI} = P(\text{up}|\text{event}) - P(\text{down}|\text{event}) + P(\text{down}|\text{no event}) - P(\text{up}|\text{no event})$$

For GPS_Mult added to PCE, NRI = 7.0%, meaning 7% net improvement in correctly classifying individuals across the 7.5% ten-year risk threshold used for statin initiation decisions.

Clinical Workflow Integration

CAD polygenic risk scores integrate into cardiovascular prevention workflows via three primary pathways:

Pathway 1: Early Risk Stratification (Ages 30–50)
One-time genome-wide array genotyping (~$50–100) → PRS calculation using validated algorithms (metaGRS, GPS_Mult) → Risk categorization: Low (<20th percentile), Intermediate (20th–80th), High (>80th percentile) → Integration with conventional risk calculators (Pooled Cohort Equations, ASCVD Risk Estimator)

Pathway 2: Shared Decision-Making for Statin Initiation
Present combined PRS + conventional risk estimates → Visual risk communication tools (lifetime risk trajectories by quintile) → Discuss gene-lifestyle interaction evidence → Jointly decide on pharmacotherapy vs. intensive lifestyle vs. surveillance

Pathway 3: EHR Implementation Architecture
Genomic Data Repository stores genotype data in Epic Genomics Module or Cerner Genomics solution → Cloud-Based PRS Calculation submits genotypes to validated calculation engines (Broad Institute, Mass General Brigham Laboratory for Molecular Medicine) → Results Interface displays PRS percentiles and absolute risk estimates → Clinical Decision Support triggers Best Practice Alerts when high-PRS patients are eligible for enhanced prevention

Clinical Workflow Timeline

Current implementation sites include Mass General Brigham Preventive Genomics Clinic (Maamari et al., 2022), Mayo Clinic (MI-GENES program), and Veterans Affairs Healthcare System (GenoVA Study validation infrastructure).

Figure 1: Implementation Workflow

Figure 2: Polygenic Risk Score Performance by Ancestry

Health Equity and Clinical Implementation Impact

PRS provide greatest clinical value for young adults (ages 30–50) before conventional risk factors manifest, individuals with intermediate conventional risk (5–10% ten-year ASCVD risk) where reclassification meaningfully impacts treatment decisions, and patients with strong family history but normal lipid panels. Early implementation data from Mass General Brigham Preventive Genomics Clinic demonstrated 32% of participants had high polygenic scores (top quintile) and 40% of participants without existing CAD had management changes (statin initiation/intensification, coronary imaging), with 80% rating genetic reports very/extremely helpful.

However, algorithmic bias may exacerbate disparities. Current PRS perform 25–46% worse in non-European populations, potentially leading to under-identification of high-risk individuals in minority populations. Mitigation requires mandated ancestry-specific performance reporting, investment in diverse biobank recruitment (All of Us Research Program target: 1 million participants, >50% underrepresented minorities), and prospective monitoring of clinical implementation stratified by race/ethnicity.

The GPS_Mult represents meaningful progress, improving African-ancestry performance by 73% over prior European-only scores. Achieving equity requires 3–5 years for next-generation multi-ancestry PRS to achieve <10% performance gap across major ancestry groups, contingent on sustained research investment in diverse biobanks, multi-ancestry GWAS (target 50% non-European representation), ancestry-adjusted algorithms accounting for LD structure, and environmental factor integration.

Regulatory and Reimbursement Pathway

Polygenic risk assessment currently bills as preventive genetic testing under CPT codes 81401-81408 (molecular pathology procedures). Genome-wide genotyping arrays cost $50–100 per test. Medicare coverage varies by indication; private insurers increasingly cover PRS for high-risk individuals under preventive care benefits. FDA has not required pre-market review for most polygenic score algorithms as they are classified as laboratory-developed tests (LDTs), though regulatory landscape is evolving. Clinical decision support integration typically bundles with evaluation and management (E&M) coding; dedicated PRS interpretation codes remain under development by AMA CPT Editorial Panel.

Implementation Impact and Scalability

Approximately 20.1 million American adults have coronary artery disease (CDC, 2024). Full implementation of PRS enables precision primary prevention, potentially preventing 15,000–30,000 CAD events annually through enhanced early statin therapy in high-risk young adults. Target adoption: 50% of academic cardiology practices within 12 months, scaling to community practices via cloud-based PRS services integrated through Epic App Orchard or Cerner Code. Current evidence gaps include prospective comparative effectiveness trials, cost-effectiveness analyses incorporating diverse populations, and long-term outcomes from deployed systems. Ongoing algorithmic fairness monitoring must ensure equitable benefit across racial and socioeconomic populations.

References

Inouye, M., Abraham, G., Nelson, C. P., Wood, A. M., Sweeting, M. J., Dudbridge, F., ... & Samani, N. J. (2018). Genomic risk prediction of coronary artery disease in 480,000 adults: Implications for primary prevention. Journal of the American College of Cardiology, 72(16), 1883–1893. https://doi.org/10.1016/j.jacc.2018.07.079

Khera, A. V., Chaffin, M., Aragam, K. G., Haas, M. E., Roselli, C., Choi, S. H., ... & Kathiresan, S. (2018). Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nature Genetics, 50(9), 1219–1224. https://doi.org/10.1038/s41588-018-0183-z

Khera, A. V., Emdin, C. A., Drake, I., Natarajan, P., Bick, A. G., Cook, N. R., ... & Kathiresan, S. (2016). Genetic risk, adherence to a healthy lifestyle, and coronary disease. New England Journal of Medicine, 375(24), 2349–2358. https://doi.org/10.1056/NEJMoa1605086

Kullo, I. J., Jouni, H., Austin, E. E., Brown, S. A., Kruisselbrink, T. M., Isseh, I. N., ... & Bailey, K. R. (2016). Incorporating a genetic risk score into coronary heart disease risk estimates: Effect on low-density lipoprotein cholesterol levels (the MI-GENES clinical trial). Circulation, 133(12), 1181–1188. https://doi.org/10.1161/CIRCULATIONAHA.115.020109

Maamari, D. J., Mele, M., Chehade, M., Hennessey, S., Tjong, M., Shah, S. J., ... & Khera, A. V. (2022). Implementation of polygenic risk scores in a preventive genomics program: Clinical and patient-reported outcomes. JACC: Advances, 1(3), 100068. https://doi.org/10.1016/j.jacadv.2022.100068

Patel, A. P., Wang, M., Ruan, Y., Koyama, S., Clarke, S. L., Yang, X., ... & Khera, A. V. (2023). A multi-ancestry polygenic risk score improves risk prediction for coronary artery disease. Nature Medicine, 29(7), 1793–1803. https://doi.org/10.1038/s41591-023-02429-x