Genetic prediction models promise to transform medicine by identifying disease risk before symptoms appear. But researchers have discovered a troubling limitation: these tools work far less accurately for people of non-European ancestry.
The problem stems from how these models are built. Scientists train genetic risk algorithms primarily on DNA from European populations, which creates blind spots for other groups. A person of African, Asian, or Latino descent may receive misleading risk assessments because the model simply lacks data about how genetic variants behave in their population.
This disparity has real consequences. Someone might learn their genetic risk for heart disease or diabetes is low when their actual risk is moderate or high, based solely on ancestry. Conversely, they might receive unnecessary interventions based on inflated risk scores. Either way, precision medicine becomes imprecise.
Researchers including those at institutions like the National Institutes of Health and major academic medical centers have flagged this issue as urgent. The genetic sequencing field has grown exponentially, yet roughly 77 percent of genetic study participants remain of European descent. This creates a cascade effect: models trained on limited diversity produce worse predictions across diverse populations, potentially widening existing health disparities rather than closing them.
The solution requires deliberate action. Scientists need to actively recruit participants from underrepresented populations into genetic studies. Funding agencies must prioritize diversity in research design. Tech companies and healthcare systems should audit their existing algorithms and clearly communicate their limitations to clinicians and patients.
Some institutions have started. The All of Us Research Program aims to enroll one million Americans from diverse backgrounds to build more inclusive genetic databases. Other initiatives focus on specific conditions affecting underrepresented groups.
Until the field builds more balanced training data, genetic risk tools should carry clear disclaimers about their population limitations. Clinicians ordering these tests need to understand they cannot rely solely on algorithmic predictions for patients whose ancestry differs from the model's training population. The promise