Polygenic Risk Scores for Coronary Artery Disease: How Much Do They Actually Add?

A polygenic risk score for coronary artery disease is a single number that ranks your inherited risk by summing the tiny effects of thousands of common DNA variants. The published evidence is genuinely two-sided: these scores do sort people by who later develops heart disease, and they can refine prediction at the edges, but the versions that dominate the literature were trained almost entirely in people of European genetic ancestry and lose much of their accuracy in everyone else. That is a large part of why no major cardiology guideline currently recommends them for routine risk stratification.

Key points#

• A coronary polygenic risk score is a relative ranking against a reference population, not a diagnosis or a fixed probability.
• On top of age, blood pressure, lipids, smoking, and diabetes, it adds incremental accuracy, most useful in younger adults whose standard risk factors have not yet declared themselves.
• Because the source data are mostly European-ancestry, the scores predict worst in the populations least represented in research.
• Multi-ancestry scores narrow that gap but have not closed it.
• No trial has yet shown that acting on a score improves hard outcomes better than acting on conventional risk assessment.

From a million small nudges to one number#

Coronary artery disease is highly polygenic. Rather than one broken gene, inherited risk is scattered across hundreds of thousands of common variants, each shifting risk up or down by a barely perceptible amount. A genome-wide association study (GWAS) estimates how strongly each variant tracks disease, and a polygenic risk score multiplies each variant you carry by that weight and adds everything together.

A helpful way to picture the output is a percentile on a curve. The score places a person somewhere along a distribution relative to the group it was trained on. It says "your inherited loading is higher or lower than most people here." It does not say the disease will or will not happen, and it does not replace a blood-pressure cuff or a lipid panel.

The prediction gain is real, but incremental#

The measured, honest version of the evidence is that a polygenic score adds information on top of the risk factors clinicians already use, and that the added information is modest rather than transformative. In a middle-aged population, age, blood pressure, lipids, smoking, and diabetes already capture most of the risk that can be predicted at all. A genetic score earns its keep mainly by reclassifying a subset of people and by pulling out the extremes of the distribution.

The 2023 Nature Medicine study by Aragam and colleagues shows what a carefully built score can manage. Their model, GPSMult, drew on GWAS data across five ancestries, with more than 269,000 coronary cases and over 1.1 million controls, plus ten coronary risk factors. In UK Biobank participants of European ancestry, the score carried an odds ratio near 2.1 per standard deviation and flagged roughly 20 percent of people as having about threefold higher risk, alongside roughly 14 percent with about threefold lower risk. That is a meaningful spread, but it remains a population-level statistic: a high stratum lifts the odds without making disease certain, and a low score is not permission to ignore tobacco or hypertension.

Where a score might change a decision#

The case for optimism, laid out by Klarin and Natarajan in their 2021 Nature Reviews Cardiology review, rests on a few concrete uses. Across cohorts, coronary polygenic scores associate consistently with incident disease. Reanalyses of statin and PCSK9-inhibitor trials suggest that people in the top genetic-risk strata may see a larger absolute benefit from preventive therapy, which is exactly where treatment decisions turn. And there is a research use that is easy to miss: enrolling high-genetic-risk participants can shrink the sample size a prevention trial needs to detect an effect, by roughly threefold in some estimates. Those are real signals that the biology is valid, even before the clinical case is settled.

Why the score travels poorly across ancestries#

Here the evidence turns cautionary, and the limitation is structural rather than cosmetic. Because the underlying GWAS data have been overwhelmingly European, the resulting scores predict best in European-ancestry groups and noticeably worse in African, East Asian, South Asian, and Hispanic or Latino populations. Klarin and Natarajan state it directly: predictions are poor in people of non-European ancestry, with the potential to create racial disparities. Writing in Nature Genetics in 2019, Martin and colleagues quantified the problem and warned that rolling out current scores could widen health inequities rather than narrow them, since the people helped least are those research has already underserved.

The mechanism is worth understanding, because it explains why more cautious clinical use cannot patch it. Variant frequencies differ across ancestral populations, the statistical correlation between neighboring markers differs, and the true effect sizes differ. A score calibrated in one group is therefore systematically miscalibrated in another. The problem lives in the training data, not in the bedside application.

Multi-ancestry scores are the field's response, and they do help. In the GPSMult work, the authors traced their roughly 38 percent overall improvement over an earlier score to three sources: about 26 percent from a larger coronary GWAS, about 9 percent from adding multi-ancestry summary statistics, and about 3 percent from borrowing signal across correlated risk factors. The multi-ancestry contribution is genuine but partial, and prediction still tended to be strongest in the European-ancestry groups that supply most of the data. Progress here is measured by how much the gap narrows, not by whether it disappears.

So why is it not routine yet#

Set side by side, the pieces support a measured conclusion. Coronary polygenic scores are biologically valid and steadily improving, but they add only modest accuracy over established risk factors, they perform unevenly across ancestral groups, and no trial has yet shown that acting on a score changes hard outcomes better than acting on a good conventional risk assessment. Open questions about calibration, standardization across the many competing scores, and equitable performance are the reason guideline bodies have stopped short of recommending these scores for general screening. The technology is promising and moving quickly; it is not yet a routine test.