Why Today's Genetic Risk Scores Miss the Mark for Many

Imagine a doctor telling you, based on a quick cheek swab, that you have a high chance of developing heart disease or breast cancer. That’s the appeal of polygenic risk scores—those handy calculators that bundle thousands of tiny DNA differences into a single risk number. In theory, they could make prevention as personal as a prescription.

In practice, however, the picture is messier. Most of the data that train these scores come from people whose ancestry is European, and that skew has a ripple effect. When the same algorithms are turned on DNA from African, Latino, or Asian backgrounds, the predictions can wobble, sometimes barely better than a guess.

Take the UK Biobank, for example. Launched in the early 2000s, it now holds genetic and health information on half a million volunteers—yet about 94 % of them identify as White. The U.S. Department of Veterans Affairs runs the Million Veteran Program, another massive repository, and roughly three‑quarters of its participants are also of European descent. Those numbers sound impressive until you remember that the United States and the world are far more diverse.

“We keep seeing the same pattern,” says Dr. Eimear Kenny, who leads the Institute for Genomic Health at Mount Sinai in New York and co‑chairs a national consortium testing these scores. “The tools work really well for the groups they were built on, but the performance drops off dramatically for others.”

That drop‑off isn’t just a statistical footnote. It could translate into missed early‑intervention opportunities for millions of people who already face health inequities. If a risk score underestimates a Black patient’s chance of developing hypertension, that individual might not get the lifestyle counseling or medication that could prevent a future heart attack.

Recognizing the problem, the National Institutes of Health recently announced the expansion of its All of Us program—a massive effort to gather genomic and health data from a truly representative slice of the U.S. population. By pulling in DNA from people of all backgrounds, researchers hope to rebuild the training sets from the ground up, making the scores more universally reliable.

But rebuilding isn’t instant. It will take years of data collection, new statistical tricks, and, frankly, a willingness to accept a little messiness in the models. In the meantime, clinicians are urged to treat polygenic risk scores as one piece of the puzzle—not the whole picture—especially when working with patients from under‑represented groups.

So the promise of personalized genetics is still there, just not yet evenly distributed. The next wave of research must focus on inclusion, or risk widening the very health gaps these tools were meant to close.