Skip to Main Content

When 23andMe received government permission in 2017 to sell health-related genetic testing, it came with several conditions, including that the company tell customers that its brand of testing can miss disease-causing variants. Now, in a study presented on Thursday at the annual meeting of the American Society of Human Genetics, a competitor has analyzed the likelihood of such “false negatives,” concluding that in some ethnic groups the chance that 23andMe’s tests will miss a cancer-causing DNA variant is 100%.

23andMe pushed back hard on that conclusion from DNA testing company Invitae. “It is patently wrong to state that 23andMe delivers ‘false negatives’ for variants that it does not test nor claim to test for,” said a company spokesman. “This is a false and misleading characterization of 23andMe’s test.”

That the kind of testing 23andMe does has the potential to miss a disease-causing variant is not in dispute. The company does not sequence DNA. Instead, it uses DNA chips, or microarrays, which are packed with millions of microscopic beads holding bits of DNA. The DNA “matches one of the genetic variants that we test,” the company explains on its website. If the company does not test for a variant, needless to say, its microarrays won’t detect it in a customer’s DNA from a spit sample.


It chooses which variants to test for based on the rigor of the science linking them to a disease, their disease-causing potential (some variants raise the risk of disease only a little, others a lot), and other factors.  For its “BRCA1 and BRCA2 (Selected Variants) report,” for instance, it tests for “some of the most well-studied variants, associated with extremely high risk,” said the spokesman.

Because a gene can have many variants — an A where there should be a T, a GA where there should be a TC, and myriad other misspellings, insertions, deletions, and other biological typos — the limitations of microarrays are well known and “obviously a concern,” said Erica Ramos, a personalized medicine expert at the National Society of Genetic Counselors. “Whether consumers understand that has been on the radar of genetic counselors for a long time.”


She did not comment directly on 23andMe’s tests, but noted that the fewer variants of a gene that a company tests for, the more people with other disease-causing variants it can miss. 23andMe tests for three variants of the BRCA1/2 genes, which can cause breast or ovarian cancer; MyHeritage tests for 13.

Microarray tests “have always been known to have this problem,” said Dr. Robert Nussbaum, chief medical officer of Invitae, who co-led the study presented at ASHG. “If you have a different variant, they don’t catch it.”

For the Invitae study, he and his colleagues studied 270,806 patients who, because of a family history of cancer, had been referred to Invitae by physicians for testing of the MUTYH gene, which has variants linked to colorectal cancer, and 119,328 who had been referred for testing of BRCA1/2. 23andMe tests for two MUTYH variants that are common in people who trace their ancestry to northern Europe, and for three BRCA1/2 variants that are common in Ashkenazi Jews. Genetic research has studied white Europeans more than other ethnic groups, which is why the best-known disease-causing variants tend to be found in those populations (and are called “founder variants”).

None of the nearly 400,000 patients were actually 23andMe customers. But if they had been, Nussbaum and his colleagues found, 40% of those with a disease-causing mutation in both of their MUTYH genes would have been told none was found. Having two such copies makes colorectal cancer a near certainty, Nussbaum said. Another 22% of those with a single MUTYH mutation, which doubles the lifetime risk of colorectal cancer, would also have been missed.

Although the BRCA1/2 variants that 23andMe tests for are most common in Ashkenazi Jews, even among them 19% of their BRCA1/2 mutations would be missed. Among non-Ashkenazis, 94% of the BRCA1/2 mutations would be missed.

The likelihood of a pathogenic variant being missed by 23andMe’s tests varied by ethnic background, according to the study. For MUTYH, 100% of the variants in Asians, 75% in African-Americans, 46% in Hispanics, and 33% in Caucasians wouldn’t be picked up.

For BRCA1/2, 98% of the variants in Asians, 99% of those in African Americans, 94% of those in Hispanics, and 94% of those in Caucasians would slip through the microarray cracks. What 23andMe calls a “0 variant” — nothing disease-causing found — “might give people a false sense of security,” Nussbaum said.

Since the study was presented at a scientific meeting and not published in a journal, it has not been peer reviewed. It is an expansion of research the authors presented earlier this year at a meeting of the American College of Medical Genetics.

Ramos agreed that ethnic minorities “are at particular risk” of having rare, disease-causing DNA variants that commercial gene chip tests don’t even look for. “One might argue that such [gene chip] tests do more harm than good,” said Dennis Grishin, the chief scientific officer of DNA testing company Nebula Genomics.

“Having an understanding of what you were tested for, and what was not tested for, can be challenging,” Ramos said. “But there are cases of people saying, oh, I don’t have [a cancer-causing] BRCA, when the test couldn’t tell them that.”

Before 23andMe received Food and Drug Administration approval to sell its health-related tests, it required the company to test whether customers would understand how it reported results. For one thing, “We are very clear with customers that we test only for certain genetic variants,” said the spokesman. “We state this clearly in our pre-purchase information” that customers must read before accessing their results  “and within the report names themselves,” such as “BRCA1/BRCA2 (Selected Variants).”

As for whether customers understand “that we do not test for all variants associated with a condition,” he said, 23andMe’s tests of customer comprehension show that 100% understood that being told “0 variants” were found does not mean that cancer-causing variants that are not part of 23andMe’s testing might not be lurking in the customer’s genome.

Seeing an opening, newer DNA testing companies have been either offering tests for many more variants than 23andMe (as in MyHeritage’s 13 BRCA variants vs. 23andMe’s three) or moving away from gene chips. Nebula, for instance, sequences customers’ genomes. Sequencing is like reading a book from cover to cover; gene chips are like hitting “control F” and looking only for pre-specified DNA sequences, which risks missing the vast majority of the text.