Polygenic risk scores rank people by inherited risk, not by fate. A myth-versus-fact guide to what the evidence supports, with ancestry gaps and the catalog of 6,991 published scores.
Polygenic risk scores add up the small effects of many DNA variants, often thousands, into a single number that ranks a person’s inherited tendency toward a condition such as coronary artery disease. Such a score is good at sorting large groups, so a few percent of people land in a clearly higher-risk tail. It is much weaker at telling any one person what will happen, and it works noticeably less well for people whose ancestry is poorly represented in the underlying studies.
That is the short version. The rest of this article uses a myth-versus-fact format to separate what the evidence supports from what marketing and headlines sometimes imply. It is general information about the science, not medical advice.
How a score is built
Genome-wide association studies compare the DNA of many people with and without a condition and estimate, for each of millions of common variants, a tiny shift in risk. A polygenic score multiplies each variant a person carries by its estimated effect and totals the results. The National Human Genome Research Institute’s glossary entry describes it as an estimate of the chance of having or developing a condition from genomic information alone, and stresses that it does not account for environmental or other factors.
The output is relative. Your score is placed against a reference group, and you are described as, say, in the top 5 percent. Whether that matters depends on how large the effect is, how common the condition is, and what could be done differently as a result.
Myth versus fact
| Common belief | What the evidence supports |
|---|---|
| A high score means you will get the disease. | It means higher-than-average risk. Many high-scoring people never develop the condition, and many people with low scores do. |
| A low score means you are safe. | It lowers estimated genetic risk only. Diet, smoking, blood pressure, other genes and plain chance remain. |
| Scores work the same for everyone. | They are several times more accurate in people of European ancestry than in other groups, according to a 2019 analysis. |
| Polygenic scores are as informative as a single-gene test like BRCA1. | For a tail of the population, the risk can be comparable. For most people, the effect is far smaller and the result is less actionable. |
| The science is settled. | The field is active, scores are being revised constantly, and clinical use is still being worked out. |
What the strongest evidence shows
The paper most often cited for clinical promise came from Amit Khera, Sekar Kathiresan and colleagues, published in Nature Genetics in 2018. Using 288,978 participants of the UK Biobank as a test set, they found that a genome-wide score for coronary artery disease put about 8.0 percent of people at more than three times the average risk. They described that share as roughly 20 times the carrier frequency of rare single-gene mutations that confer a similar risk.
They reported the same threefold-risk tail for atrial fibrillation (6.1 percent), type 2 diabetes (3.5 percent), inflammatory bowel disease (3.2 percent) and breast cancer (1.5 percent). The point was not that polygenic scores are more powerful than rare mutations. It was that far more people carry the polygenic kind of high risk, so a population strategy could in principle reach more of them.
Two cautions go with that headline. First, “more than threefold” is relative to the average, and the starting level matters a great deal. Second, the UK Biobank is a research volunteer cohort that skews healthier and more European than the general population, so a score validated there may not transfer unchanged.
Relative risk versus absolute risk
Here is an illustration with invented numbers, to show the logic. Suppose a condition affects 2 out of 100 people over a lifetime. A person with three times the average risk would have about 6 out of 100, which means 94 out of 100 people in that high-risk group would not get the condition, a point that is easy to lose in a headline about tripled risk. Now suppose the condition affects 20 out of 100 people. The same threefold multiplier would imply 60 out of 100, a far larger absolute change. One relative number, two very different real-world meanings, which is why a percentile or a ratio alone should never be read as a probability.
This is why researchers pair a score with age, sex, family history, blood pressure, cholesterol and other known risk factors, and ask whether the score improves prediction beyond them. An addition that reclassifies a small number of people is not the same as a test that transforms care.
The ancestry problem
Most large genetic studies have enrolled people of European descent. Because a polygenic score is trained on those effect sizes and on the linkage patterns between variants that are specific to a population, it tends to lose accuracy elsewhere. A 2019 Nature Genetics analysis by Alicia Martin and colleagues concluded that current scores are several times more accurate in European-ancestry people than in other groups, and warned that clinical use could widen health disparities.
Practical consequences follow. A person of African, Asian, Latino or mixed ancestry may receive a score that is less precise, or placed on a distribution built from a different population. Researchers are developing methods that adjust for ancestry, and the PGS Catalog’s 2024 flagship publication, per its About page, covers tools for calculation and ancestry normalization. That is progress, but it does not remove the need for diverse training data.
What a score cannot see
Several kinds of risk sit outside a polygenic score by design. Rare variants with large effects, such as the single-gene changes behind inherited cancer syndromes, are handled by separate tests. Environmental exposure, medications, other diseases and plain luck shape outcomes too, and the NHGRI notes that a score ignores non-genomic influences. Age and sex typically need to be modeled alongside it.
There is also a ceiling. A trait’s heritability, the share of its variation tied to genetic differences, limits how well any DNA-based score can predict it, however large the study behind it. For conditions with modest heritability, even a perfectly built score would leave most of the picture blank.
How to read a percentile
A report might say you are at the 92nd percentile for a condition. That means your score is higher than 92 percent of the reference group, not that your chance of illness is 92 percent. The reference group matters, and so does the question of what the top tail is being compared against. Ask for the absolute lifetime or ten-year risk, and ask how it compares with the risk estimated from your age, sex, family history and measurements. If the new number would not change what you or your doctor do, its practical value is small.
A crowded and uneven landscape
Scores are plentiful. The open PGS Catalog, whose latest release is dated September 17, 2026, lists 6,991 polygenic scores for 813 traits drawn from 836 publications. Many traits have multiple competing scores built with different methods and samples. Two scores for the same disease can rank the same person differently, which is another reason an individual result should be read cautiously.
The catalog requires publications to describe the samples used to develop and test each score, so a reader can check who was studied. That transparency is useful, but a published score is not an endorsed clinical test.
Where scores are being studied for care
Several research programs are evaluating how to return polygenic results to patients and how doing so affects behavior and care. The eMERGE Network, which links DNA biorepositories to electronic medical records, lists among its recent publications work on genetic risk disclosure and behavior change (Naderian and Kullo, February 2025) and on factors that affect score performance across diverse groups (Hui and colleagues, January 2025), per its site. I did not review those papers, so their findings are not summarized here.
The NHGRI glossary is candid that the science is still developing and the scores are not in wide use. It notes that people in the top 1 percent or higher for risk may benefit from more frequent screening or earlier lifestyle changes, a statement of possibility rather than a guideline.
Questions to ask if you are offered a score
- Which condition does it address, and which population was it built and tested in?
- Is my result expressed as relative risk, absolute risk, or just a percentile?
- Does it change what is recommended for me beyond what family history and standard tests already show?
- Who will see the result and how is the data stored?
A genetic counselor or physician can help answer them for a specific situation. The honest summary today is narrow: polygenic scores are a useful research and population tool, a promising but unproven clinical one, and an unreliable crystal ball for any single person.
Sources
