Polygenic Risk Scores Explained: What They Measure, What They Miss, and Whether to Care

Genetic medicine has long promised personalised risk estimates, yet for most of medical history only a small category of patients received them: those who happened to carry rare, high-penetrance mutations in genes like BRCA1, BRCA2, or those responsible for familial hypercholesterolaemia. For the vast majority of people whose risk of heart disease, diabetes, or cancer is shaped not by one decisive mutation but by thousands of tiny genetic nudges spread across the entire genome, personalised risk quantification remained out of reach.

That changed with the maturation of genome-wide association studies and the development of polygenic risk scores (PRS). A PRS aggregates the small, incremental effects of hundreds of thousands of common variants into a single number that positions you within a population distribution of inherited susceptibility. The methodology is now sophisticated enough that the best coronary artery disease PRS algorithms rival established clinical risk scores in predictive power, and screening programs in several countries are beginning to use breast cancer PRS to personalise mammography scheduling. But the technology also carries limitations that are frequently understated in consumer marketing, particularly around ancestry bias and the gap between population-level statistics and individual predictions. This article explains what polygenic risk scores actually measure, how they are built, which diseases have the strongest validated scores, and what to do with your number if you have one.

What Is a Polygenic Risk Score?

A polygenic risk score is a numerical summary of your inherited genetic predisposition to a specific disease. Unlike a diagnostic test, which reports whether you have or do not have a condition, a PRS is probabilistic: it tells you how your genetic makeup compares to a reference population in terms of inherited disease risk. The word "polygenic" is important because it distinguishes these scores from single-gene tests. Most common diseases, including coronary artery disease, type 2 diabetes, and the majority of cancers, are not caused by one defective gene. They arise from the cumulative action of hundreds or thousands of genetic variants, each conferring a fractionally higher or lower risk.

The individual variants in a PRS are called single-nucleotide polymorphisms, or SNPs. A SNP is a position in the genome where people commonly differ by one DNA letter, for example an A at a position where others have a G. Any given SNP typically shifts disease risk by less than one percent. The power of a PRS comes from aggregating these tiny signals across the entire genome, where their combined influence can be substantial. Someone at the 95th percentile of a well-validated CAD PRS carries roughly twice the average lifetime risk of coronary events. Someone at the 99th percentile may carry three to four times the average risk, a magnitude that is clinically actionable.

The output scale of a PRS depends on the algorithm and how it is reported. Some systems express the raw weighted sum; others convert it to a percentile within a reference population; others express it as odds ratios or absolute lifetime risk estimates. Percentile framing is common in consumer products because it is intuitive, but absolute risk framing is more useful for clinical decision-making. A score at the 90th percentile sounds alarming, but if the baseline lifetime risk of the condition in your demographic is 5 percent, the 90th-percentile genetic risk might translate to roughly 10 to 12 percent absolute lifetime risk, still a minority probability.

How Polygenic Risk Scores Are Calculated

From GWAS to a Weighted Sum

Every PRS begins with a genome-wide association study. Researchers genotype hundreds of thousands to millions of participants, compare those who develop a given disease with those who do not, and identify SNPs that appear significantly more or less often in the affected group. Each SNP receives a beta coefficient (in linear models) or a log odds ratio (in logistic models) reflecting the strength of its association with the outcome. These weights are derived from the GWAS summary statistics.

To calculate your individual PRS, an algorithm multiplies your genotype at each included SNP (coded 0, 1, or 2 copies of the risk allele) by the corresponding weight, then sums across all variants. This is the weighted allele sum that defines your score. Early PRS models used only the variants that crossed a stringent genome-wide significance threshold (p < 5 x 10-8), typically a few dozen to a few hundred SNPs. Modern approaches use far more variants, including those with weaker statistical evidence, because sub-threshold variants still carry real signal when their effects are aggregated across millions of base pairs.

Clumping, Thresholding, and Bayesian Refinement

A key methodological challenge is linkage disequilibrium (LD): nearby SNPs on the same chromosome are correlated with each other, so including all of them would effectively double-count the same signal. The classic solution is clumping and thresholding (C+T): retain only the most significant SNP in each LD block, then include all retained SNPs above a p-value threshold. Tools like PRSice-2 automate this process and were instrumental in early clinical validation studies.

More recent Bayesian approaches, especially LDpred and its successor LDpred2, explicitly model the LD structure of the genome and use a prior distribution on SNP effect sizes to shrink noisy estimates toward zero while preserving true signals. These methods consistently outperform C+T approaches, particularly when the underlying GWAS is large. The practical implication for patients is that a PRS generated with LDpred2 from a GWAS of one million participants will generally be more predictive than one built with C+T from a GWAS of fifty thousand participants, even if both claim to test the same condition.

What Diseases Have Validated PRS Tests?

Not every condition has a clinically useful PRS. Utility requires that the condition has substantial heritability captured by common variants, that a large and diverse enough GWAS exists to produce reliable weights, that the disease is prevalent enough for stratification to matter, and that there are actionable interventions a high-risk person can take. The conditions with the strongest current evidence include the following.

Coronary artery disease (CAD) has the best-validated PRS in medicine. Khera et al. (Nature Genetics, 2018) developed a genome-wide PRS from over six million SNPs and showed it identified 8 percent of the population as carrying more than three-fold elevated CAD risk. Independently, Inouye et al. (JACC, 2018) validated a genome-wide CAD PRS in the UK Biobank and EPIC-Norfolk cohorts, reporting an AUC of 0.81 and demonstrating that individuals in the top 20 percent of PRS had event rates approaching those seen in familial hypercholesterolaemia carriers. These findings have driven cardiology societies to consider PRS integration into risk frameworks alongside LDL cholesterol, blood pressure, age, and smoking status, following the updated PREVENT framework from the American Heart Association.

Breast cancer PRS algorithms now combine thousands of common SNPs with rare pathogenic variants and achieve AUC values of 0.63 to 0.68, sufficient to stratify women meaningfully into screening risk tiers. The BOADICEA model, used in UK clinical practice, integrates both common variant PRS and rare variant status (including BRCA gene testing results) to produce a lifetime risk estimate that guides screening intensity decisions.

Type 2 diabetes PRS achieves AUC values of approximately 0.64 to 0.72 in European-ancestry cohorts. The score identifies people who will develop diabetes years before clinical glucose abnormalities appear, creating an opportunity for lifestyle intervention during the window when prevention is most effective. Atrial fibrillation and prostate cancer also have well-validated scores, with atrial fibrillation PRS showing AUC values approaching 0.77 in the largest validation studies and already influencing anticoagulation decisions in some cardiology practices.

What Is a Polygenic Risk Score and Should I Care?

This is the practical question that matters most for individuals. The answer depends on where your score falls and what disease it concerns. Here is a framework for thinking about it clearly.

High PRS: A Signal to Act, Not a Verdict

If your PRS for coronary artery disease or breast cancer places you in the top 10 to 20 percent, that is clinically meaningful information. For CAD, it suggests your physician should apply stricter targets for modifiable risk factors and consider whether preventive medications like statins are warranted at a younger age than population guidelines specify. For breast cancer, it may justify starting mammography at 40 or 45 rather than 50. These are conversations to have with your clinician, not unilateral decisions to make based on a consumer report.

The distinction between relative risk and absolute risk is critical here. A PRS at the 90th percentile may be described as conferring "twice the average risk." If the baseline 10-year risk for your age and sex is 5 percent, twice the average is 10 percent: elevated, worth addressing, but still a 90 percent chance of not having an event in the next decade. Framing the absolute number alongside the relative elevation prevents both complacency (for those at average risk who see a high percentile) and unnecessary panic.

Low PRS: Not a Licence to Ignore Risk Factors

A low PRS does not confer protection against lifestyle-driven risk. Most people who develop common diseases carry average or below-average polygenic risk, simply because the middle of the distribution contains far more people than the tails. Smoking, physical inactivity, poor dietary patterns, hypertension, and obesity elevate risk substantially regardless of what your genome says. A low PRS is reassuring but does not replace healthy habits or routine clinical screening appropriate for your age.

The Major Limitations You Must Understand

The Ancestry Problem

The single most important limitation of current polygenic risk scores is their reduced accuracy in people of non-European ancestry. Martin et al. (Nature Genetics, 2019) quantified this disparity systematically: PRS for a range of conditions performed substantially worse in African, East Asian, South Asian, and admixed populations compared with European-ancestry individuals. The root cause is that GWAS datasets have been overwhelmingly composed of participants of European descent. The effect-size weights derived from these studies reflect European-specific allele frequencies and linkage disequilibrium patterns. When applied to people with different population structure, the weights are miscalibrated, producing scores that are less predictive and potentially misleading.

This is not a minor caveat. It represents a fundamental equity problem in precision medicine. Efforts to address it include the H3Africa consortium, the All of Us Research Program, and the Global Biobank Meta-analysis Initiative, all of which are generating the large, diverse GWAS datasets needed to build ancestry-appropriate PRS. Multi-ancestry PRS methods that jointly model data from multiple populations are also under active development. Until these efforts mature, any PRS result for a person of African, Latino, or Asian ancestry should be interpreted with explicit acknowledgment that the score's predictive accuracy has not been validated to the same degree as for people of European descent.

What PRS Does Not Capture

Common variants identified in GWAS explain only a portion of the genetic architecture of most diseases, a phenomenon called the "missing heritability" problem. Rare variants with larger individual effects, structural variants, somatic mutations acquired during life, and epigenetic modifications are not captured in standard PRS. Gene-environment interactions, where the effect of a genetic variant depends on diet, exercise, or environmental exposures, are also not modelled. Disease heterogeneity is another limitation: "coronary artery disease" and "type 2 diabetes" are not monolithic entities but collections of overlapping mechanisms, and a single PRS cannot distinguish between subtypes that may have different genetic drivers and different optimal treatments.

PRS vs Direct-to-Consumer Tests Like 23andMe

Consumer genetic services like 23andMe genotype a panel of roughly 600,000 to 700,000 SNPs chosen primarily for ancestry inference and a limited set of trait associations. Their health reports use a subset of this panel, typically a few hundred to a few thousand variants, to estimate disease risk. This is categorically different from a clinical-grade PRS built from millions of variants weighted by the latest large-scale GWAS using Bayesian LD-aware algorithms.

The practical consequence is that consumer DNA health reports can tell you something meaningful about your genetic background but should not be used to make clinical decisions in the way that a validated clinical PRS can. A consumer service might flag moderate elevation in type 2 diabetes risk using 100 SNPs. A clinical PRS for the same condition incorporates 200 times as many variants, uses more accurate effect-size estimates, and has been validated against clinical outcomes in independent cohorts. For more detail on what consumer tests can and cannot tell you, see our comparison of 23andMe vs clinical genomics. For context on the sequencing methods that generate the raw data for these analyses, see our guide to whole genome vs exome sequencing.

If you already have raw genotyping data from a consumer service, third-party platforms can calculate condition-specific clinical-grade PRS from your existing file, often with greater methodological transparency. The key questions to ask of any PRS service: What variants are included? What reference population was used to derive the weights? What method was used (C+T vs LDpred2)? Has the algorithm been validated in a population similar to mine? What absolute lifetime risk does my percentile translate to?

How Clinicians Are Using Polygenic Risk Scores Now

Cardiology: Risk Stratification and Statin Decisions

In cardiology, PRS is beginning to supplement traditional risk calculators. The American Heart Association's updated PREVENT framework (2023-2024) incorporates novel risk-enhancing factors, and PRS is increasingly viewed as one such factor. Patients with borderline or intermediate 10-year CAD risk by PREVENT score who also carry a high CAD PRS are frequently reclassified to high risk, justifying earlier initiation of statin therapy or aspirin use. The coronary artery calcium (CAC) score from CT imaging is another risk-enhancer in this framework, and some cardiologists now order both CAC and PRS together to get orthogonal lines of evidence: CAC reflects disease that has already accumulated, while PRS reflects inherited predisposition that existed from birth. Together they give a more complete picture of both current arterial burden and lifetime susceptibility. This approach exemplifies the broader promise of precision medicine for cardiovascular disease.

Breast Cancer Screening Age Decisions

In oncology, breast cancer PRS is being used to personalise the age at which women are invited for mammographic screening. The WISDOM trial in the United States is comparing annual fixed-schedule mammography with a PRS-informed adaptive schedule in a randomised design, with preliminary results suggesting that risk-stratified scheduling is non-inferior for cancer detection while reducing unnecessary imaging in lower-risk women. The UK PROCAS and BOADICEA models integrate common variant PRS with family history, lifestyle factors, and rare variant status to produce an integrated lifetime risk estimate that determines whether a woman is offered standard, enhanced, or high-risk surveillance pathways.

The Role of Genetic Counselling

A recurring theme across all clinical PRS applications is the necessity of genetic counselling for high-risk results. A PRS is one input among many. Family history often adds predictive information beyond what the PRS captures, particularly for conditions where rare high-penetrance variants make a meaningful contribution. The BRCA gene testing framework illustrates this: a high breast cancer PRS without a BRCA mutation and a low breast cancer PRS with a BRCA mutation are very different clinical situations that require different management. A genetic counsellor is equipped to synthesise both rare and common variant information into a coherent personalised risk assessment. The intersection of AI and genomics is also accelerating this synthesis, with machine learning models beginning to integrate PRS, rare variant calls, biomarker data, and imaging into unified risk models that no single data type could power alone.

How to Interpret Polygenic Risk Scores: A Step-by-Step Clinical Guide

Interpreting a polygenic risk score requires moving through several layers of context before drawing any clinical conclusion. Step one: identify the reference population. A PRS in the 90th percentile means you scored higher than 90% of people in the population against which the score was calibrated — if that population was predominantly European-ancestry, the percentile is only meaningful for people of similar ancestry. Ask your provider or the testing service what reference population was used. Step two: convert the percentile to absolute risk. A percentile is relative; you need the baseline absolute risk for your age, sex, and demographic to calculate what the score means in practice. A CAD PRS in the 85th percentile for a 40-year-old woman with no other risk factors may translate to a 10-year absolute risk of 4–6%, compared to a 2–3% baseline — elevated but not alarming. For a 55-year-old man with hypertension, the same PRS percentile translates to a much higher absolute risk. Step three: layer the PRS over traditional risk factors. The PRS captures genetic predisposition; traditional risk factors (LDL, blood pressure, smoking, BMI) capture lifestyle-determined risk. A high PRS combined with modifiable risk factors is the most actionable scenario — it gives you both the motivation and the specific targets to address. Step four: determine the management implication. Does this result change what screening you should have, at what age, how frequently? Would it alter a decision about preventive medication? If the honest answer to both questions is “no,” the clinical utility of the PRS in your specific situation may be limited. Step five: discuss with a qualified clinician or genetic counsellor. A PRS report without clinical interpretation context is data without decision support, and a number without context is more likely to generate anxiety than actionable insight.