Skip to Content
Biotechnology and health

23andMe thinks polygenic risk scores are ready for the masses, but experts aren’t so sure

A new genetic test that estimates your risk for diabetes is probably less useful than standing on a scale.
Sharon McCutcheon | Unspalsh; 23andME


23andMe, the direct-to-consumer DNA testing company in Silicon Valley whose motto is “Everyone has a right to their genetic information,” is rolling out a potentially controversial new type of disease prediction.

Today at the SXSW festival in Austin, and via press release, 23andMe is announcing plans to a commercialize a DNA assessment it says will gauge people’s risk of developing type 2 diabetes from their genes.  

Several million of its customers will begin receiving the health information Monday, according to 23andMe.

The report is based on a polygenic risk score, a genetic calculation that weighs a person’s odds of developing a medical condition. It does so by inspecting DNA information spread across the genome.

In the case of the new diabetes test, 23andMe says its report will evaluate information at 1,244 distinct locations in a person’s genome, each with a small bearing on the overall risk for diabetes.

About 80% of customers will learn that their DNA puts them at average risk, and 20% will learn they have an increased likelihood of getting diabetes. Only those in the high-risk category will be provided exact odds (for instance, a 3-in-5 lifetime chance) of developing the disease.

Experts questioned the introduction of such a report, saying that while polygenic scoring systems look promising, they are not highly accurate and don’t have proven health benefits. “I think it’s a huge experiment,” says Peter Kraft, an epidemiologist at Harvard University’s school of public health. “You’re rolling it out to millions of people, but there is a lot we don’t know.”

An example report, provided by 23andMe to the media last week, is for an imaginary Latino customer named Jamie. Jamie learns his genes predict he’s got a very high chance of getting diabetes. It then encourages him to check out a $19.99-a-month health coaching app, called Lark, sold by a 23andMe partner.

Underlying the report is some razzle-dazzle new science. With the DNA from enough people, it’s now become possible to create statistical models that can predict, from any single individual’s DNA, what traits that person is likely to have, including the chance of getting diabetes or breast cancer, as well as the likelihood of being particularly short or having a higher IQ than average.

More than 216 scientific papers on polygenic risk scores were published last year. The idea also made our 2018 list of 10 Breakthrough Technologies.

To build its diabetes predictor, 23andMe says, it employed its own vast trove of DNA to study more than 70,000 customers who’d told the company they have the condition, as well as a couple of million who said they don’t.

For 23andMe, the promise of the scoring technology is substantial. According to a grant application, 23andMe considers “highly scalable and accurate disease risk estimation” to represent the “next phase” of its research efforts. The company declined to say if it planned to introduce scores for other conditions.

In 2013, the US government forced 23andMe to retire a large slate of health tests, many of which also employed polygenic predictors, including one for type 2 diabetes, because their accuracy was unproved and they might cause people to take unnecessarily medical steps.

Since then, however, the science of prediction has improved and regulations have loosened. According to 23andMe, the current diabetes report needs no regulation at all. That is because it falls into an exemption for low-risk tests and phone apps that offer only “general wellness” suggestions, not real medical advice or diagnoses.

The FDA did not respond to questions about whether polygenic risk scores for common diseases would be exempt from regulation.  

“It’s totally expected they are going this route,” says Cecile Janssens, an epidemiologist at Emory University. “Although it is way too early to share this kind of information, 23andMe thinks as long as they are honest about the shortcomings, who cares.”

Some doctors said genetic prediction has questionable benefit in diabetes, a disease strongly shaped by age, diet, and weight. A bathroom scale, for example, would be about three times as effective in identifying people with diabetes risk as 23andMe’s $199 genetic test. That is because many Americans are overweight. Overall about 9.5% of American adults have diabetes

The genetic predictions are also especially spotty for African-Americans. The company developed its model using DNA from white people of European ancestry, who make up most of its database. The result is that the predictions perform less well for other populations.

“I don’t believe the African-American results,” says James Meigs, an endocrinologist at Massachusetts General Hospital who specializes in diabetes prevention. “These risk scores do not perform in blacks, disease after disease. If you are a black person doing this test, it’s not giving the right answer.”

For that reason, some companies offering polygenic tests have restricted them to people of European ancestry. In a statement, 23andMe says its test is “accurate” for African-American, Latinos, and Asians.

Once they have their scores, customers of 23andMe will be able to dig into the data with an interactive widget, which shows 100 human figures representing people with genes like theirs. More or fewer of these turn dark blue (for diabetes) depending on factors the users can select with a drop-down menu, including age, weight, and frequency of eating fast food.

Selecting for lower risk in real life is not as easy. Meigs gives 23andMe points for educating the public and thinks the approach is well thought out. However, from what he sees treating people with diabetes, “the killer problem is that knowledge does not change human behavior.” It’s just too hard to get off the couch and put down the fries.

He doesn't think that a 23andMe test is going to change that. “It’s a business model,” says Meigs. “They will sell kits, but it can’t be expected to improve public health at all.”

Deep Dive

Biotechnology and health

FDA advisors just said no to the use of MDMA as a therapy

The studies demonstrating MDMA’s efficacy against PTSD left experts with too many questions to greenlight the treatment.

Biotech companies are trying to make milk without cows

The bird flu crisis on dairy farms could boost interest in milk protein manufactured in microorganisms and plants. 

What’s next for MDMA

The FDA is poised to approve the notorious party drug as a therapy. Here’s what it means, and where similar drugs stand in the US. 

Beyond Neuralink: Meet the other companies developing brain-computer interfaces

Companies like Synchron, Paradromics, and Precision Neuroscience are also racing to develop brain implants

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.