Thousands of Variants, One Imperfect Number: The Limits of Polygenic Risk Scores in Clinical Practice

For more than a decade, polygenic risk scores have occupied an almost messianic position in the narrative of precision medicine. The promise was straightforward: gather enough genetic variants associated with a disease, weight them by their individual effect sizes, sum the result, and arrive at a number that tells a patient how likely they are to develop type 2 diabetes, coronary artery disease, breast cancer, or schizophrenia. The logic is mathematically coherent. The clinical reality is considerably more complicated.

What a Polygenic Risk Score Actually Measures

A polygenic risk score (PRS) is not a diagnosis, a prognosis, or a prediction in the strict sense of the word. It is a statistical summary—an aggregate of effect estimates drawn from genome-wide association studies (GWAS), each of which identifies single-nucleotide polymorphisms (SNPs) associated with a given trait in a specific study population. When researchers combine tens of thousands, or even millions, of such variants into a single score, they produce a number that correlates with disease incidence at the population level.

The crucial distinction is between population-level correlation and individual-level prediction. A PRS may explain 10 to 15 percent of the variance in coronary artery disease risk across a large cohort, but that residual variance—the 85 to 90 percent the score does not capture—is substantial. It encompasses rare variants with large effects, gene-environment interactions, epigenetic modifications, lifestyle factors, and the accumulated stochasticity of biological development. A high PRS does not mean a patient will develop disease; a low PRS does not mean they will not.

The Population Diversity Problem

Perhaps the most consequential limitation of current polygenic risk scores is their deeply uneven performance across ancestral populations. The overwhelming majority of GWAS participants have been of European descent. Estimates from the GWAS Catalog suggest that individuals of European ancestry have historically comprised more than 70 percent of study participants, despite representing a minority of the global population.

This imbalance has measurable consequences. SNPs identified as predictive in European cohorts may have different frequencies, different linkage disequilibrium patterns, or different effect sizes in individuals of African, East Asian, South Asian, or Latino ancestry. A PRS trained predominantly on European data can produce systematically miscalibrated scores when applied to patients from other backgrounds—sometimes overestimating risk, sometimes underestimating it. In a clinical context, this is not a theoretical concern. It is a mechanism for amplifying existing health disparities.

Efforts to develop ancestry-diverse training datasets and multi-ancestry PRS models are underway at institutions including the National Human Genome Research Institute and through collaborative initiatives such as the Global Biobank Meta-analysis Initiative. Progress is real but incomplete, and clinicians applying PRS tools today should understand that the evidence base for their use is far stronger in patients of Northern European ancestry than in the broader US population.

The Gap Between Statistical Power and Clinical Utility

A score can be statistically significant and clinically marginal at the same time. This distinction matters enormously when evaluating whether a PRS should influence medical decision-making.

Consider coronary artery disease, one of the conditions for which PRS tools are arguably most developed. A landmark 2018 study published in Nature Genetics demonstrated that individuals in the top percentile of a genome-wide polygenic score had a threefold higher lifetime risk of coronary artery disease compared to the median. That finding generated considerable enthusiasm. What received less attention was that traditional clinical risk calculators—incorporating age, sex, blood pressure, cholesterol, smoking history, and diabetes status—performed comparably or better in most patient populations. Adding the PRS to established risk models improved prediction modestly at best.

This does not make PRS tools useless. For certain conditions with limited environmental risk factors, or for identifying high-risk individuals at younger ages before conventional risk factors have accumulated, polygenic scores may add genuine value. The American College of Medical Genetics and Genomics has noted potential utility in cascade screening and early intervention contexts. But the bar for clinical adoption should be evidence of improved patient outcomes, not merely improved area under a receiver operating characteristic curve.

Interpreting Scores Responsibly

For geneticists, genetic counselors, and primary care physicians navigating this landscape, several principles are worth internalizing.

Context is not optional. A PRS delivered without accompanying information about its training population, its predictive performance metrics, and the absolute risk it implies is an incomplete clinical tool. Patients who receive a "high risk" designation without understanding that the score places them in, say, the 85th percentile for a condition that affects 5 percent of the population may draw conclusions far more alarming than the data warrant.

Ancestry matters, and patients deserve to know. Clinicians should be transparent with patients about whether their ancestral background is well-represented in the training data for the score being applied. This is not a minor caveat; it is central to the score's validity for that individual.

Risk is not fate. This point cannot be overstated. Even individuals with the highest polygenic risk scores for most common diseases will not develop those diseases with certainty. Behavioral and environmental factors remain powerful modifiers of genetic predisposition, and communicating this clearly is an ethical imperative.

Regulatory and professional guidance is evolving. The US Food and Drug Administration has begun examining direct-to-consumer PRS products, and professional societies are developing updated frameworks for clinical use. Staying current with these guidelines is essential for responsible practice.

The Road Ahead

Polygenic risk scores represent a genuine scientific achievement. The ability to distill complex genetic architecture into a quantitative metric—and to demonstrate that this metric correlates with disease across large populations—reflects decades of rigorous work in statistical genetics and genomic epidemiology. The challenge now is translating that achievement into tools that are equitable, interpretable, and genuinely useful at the bedside.

That translation requires continued investment in diverse biobanks, more sophisticated models that integrate environmental and epigenetic data alongside germline variants, and an honest reckoning with what these scores currently cannot do. The patients who will benefit most from precision medicine are precisely those who have been most underrepresented in the research generating it. Closing that gap is both a scientific priority and a moral one.